## Lexical and Computational Semantics and Semantic Evaluation (formerly Workshop on Sense Evaluation) (2022)

up

#### pdf (full)bib (full)Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)

Word embeddings have advanced the state of the art in NLP across numerous tasks. Understanding the contents of dense neural representations is of utmost interest to the computational semantics community. We propose to focus on relating these opaque word vectors with human-readable definitions, as found in dictionaries This problem naturally divides into two subtasks: converting definitions into embeddings, and converting embeddings into definitions. This task was conducted in a multilingual setting, using comparable sets of embeddings trained homogeneously.
This paper describes our system for the Se- mEval2022 task of matching dictionary glosses to word embeddings. We focus on the Reverse Dictionary Track of the competition, which maps multilingual glosses to reconstructed vector representations. More specifically, models convert the input of sentences to three types of embeddings: SGNS, Char, and Electra. We pro- pose several experiments for applying neural network cells, general multilingual and multi-task structures, and language-agnostic tricks to the task. We also provide comparisons over different types of word embeddings and ablation studies to suggest helpful strategies. Our initial transformer-based model achieves relatively low performance. However, trials on different retokenization methodologies indicate improved performance. Our proposed Elmo- based monolingual model achieves the highest outcome, and its multitask, and multilingual varieties show competitive results as well.
This paper describes the BLCU-ICALL system used in the SemEval-2022 Task 1 Comparing Dictionaries and Word Embeddings, the Definition Modeling subtrack, achieving 1st on Italian, 2nd on Spanish and Russian, and 3rd on English and French. We propose a transformer-based multitasking framework to explore the task. The framework integrates multiple embedding architectures through the cross-attention mechanism, and captures the structure of glosses through a masking language model objective. Additionally, we also investigate a simple but effective model ensembling strategy to further improve the robustness. The evaluation results show the effectiveness of our solution. We release our code at: https://github.com/blcuicall/SemEval2022-Task1-DM.
This paper introduces the approach of Team LingJing’s experiments on SemEval-2022 Task 1 Comparing Dictionaries and Word Embeddings (CODWOE). This task aims at comparing two types of semantic descriptions and including two sub-tasks: the definition modeling and reverse dictionary track. Our team focuses on the reverse dictionary track and adopts the multi-task self-supervised pre-training for multilingual reverse dictionaries. Specifically, the randomly initialized mDeBERTa-base model is used to perform multi-task pre-training on the multilingual training datasets. The pre-training step is divided into two stages, namely the MLM pre-training stage and the contrastive pre-training stage. The experimental results show that the proposed method has achieved good performance in the reverse dictionary track, where we rank the 1-st in the Sgns targets of the EN and RU languages. All the experimental codes are open-sourced at https://github.com/WENGSYX/Semeval.
What is the relation between a word and its description, or a word and its embedding? Both descriptions and embeddings are semantic representations of words. But, what information from the original word remains in these representations? Or more importantly, which information about a word do these two representations share? Definition Modeling and Reverse Dictionary are two opposite learning tasks that address these questions. The goal of the Definition Modeling task is to investigate the power of information laying inside a word embedding to express the meaning of the word in a humanly understandable way – as a dictionary definition. Conversely, the Reverse Dictionary task explores the ability to predict word embeddings directly from its definition. In this paper, by tackling these two tasks, we are exploring the relationship between words and their semantic representations. We present our findings based on the descriptive, exploratory, and predictive data analysis conducted on the CODWOE dataset. We give a detailed overview of the systems that we designed for Definition Modeling and Reverse Dictionary tasks, and that achieved top scores on SemEval-2022 CODWOE challenge in several subtasks. We hope that our experimental results concerning the predictive models and the data analyses we provide will prove useful in future explorations of word representations and their relationships.
We propose a pair of deep learning models, which employ unsupervised pretraining, attention mechanisms and contrastive learning for representation learning from dictionary definitions, and definition modeling from such representations. Our systems, the Transformers for Learning Dictionaries and Representations (TLDR), were submitted to the SemEval 2022 Task 1: Comparing Dictionaries and Word Embeddings (CODWOE), where they officially ranked first on the definition modeling subtask, and achieved competitive performance on the reverse dictionary subtask. In this paper we describe our methodology and analyse our system design hypotheses.
This paper presents a novel and linguistic-driven system for the Spanish Reverse Dictionary task of SemEval-2022 Task 1. The aim of this task is the automatic generation of a word using its gloss. The conclusion is that this task results could improve if the quality of the dataset did as well by incorporating high-quality lexicographic data. Therefore, in this paper we analyze the main gaps in the proposed dataset and describe how these limitations could be tackled.
This paper presents a winning submission to the SemEval 2022 Task 1 on two sub-tasks: reverse dictionary and definition modelling. We leverage a recently proposed unified model with multi-task training. It utilizes data symmetrically and learns to tackle both tracks concurrently. Analysis shows that our system performs consistently on diverse languages, and works the best with sgns embeddings. Yet, char and electra carry intriguing properties. The two tracks’ best results are always in differing subsets grouped by linguistic annotations. In this task, the quality of definition generation lags behind, and BLEU scores might be misleading.
Described are our two entries “emukans” and “guntis” for the definition modeling track of CODWOE SemEval-2022 Task 1. Our approach is based on careful scaling of a GRU recurrent neural network, which exhibits double descent of errors, corresponding to significant improvements also per human judgement. Our results are in the middle of the ranking table per official automatic metrics.
We present the Uppsala University system for SemEval-2022 Task 1: Comparing Dictionaries and Word Embeddings (CODWOE). We explore the performance of multilingual reverse dictionaries as well as the possibility of utilizing annotated data in other languages to improve the quality of a reverse dictionary in the target language. We mainly focus on character-based embeddings.In our main experiment, we train multilingual models by combining the training data from multiple languages. In an additional experiment, using resources beyond the shared task, we use the training data in Russian and French to improve the English reverse dictionary using unsupervised embeddings alignment and machine translation. The results show that multilingual models occasionally but not consistently can outperform the monolingual baselines. In addition, we demonstrate an improvement of an English reverse dictionary using translated entries from the Russian training data set.
This paper describes our two deep learning systems that competed at SemEval-2022 Task 1 “CODWOE: Comparing Dictionaries and WOrd Embeddings”. We participated in the subtask for the reverse dictionary which consists in generating vectors from glosses. We use sequential models that integrate several neural networks, starting from Embeddings networks until the use of Dense networks, Bidirectional Long Short-Term Memory (BiLSTM) networks and LSTM networks. All glosses have been preprocessed in order to consider the best representation form of the meanings for all words that appears. We achieved very competitive results in reverse dictionary with a second position in English and French languages when using contextualized embeddings, and the same position for English, French and Spanish languages when using char embeddings.
The reverse dictionary task is a sequence-to-vector task in which a gloss is provided as input, and the output must be a semantically matching word vector. The reverse dictionary is useful in practical applications such as solving the tip-of-the-tongue problem, helping new language learners, etc. In this paper, we evaluate the effect of a Transformer-based model with cross-lingual zero-shot learning to improve the reverse dictionary performance. Our experiments are conducted in five languages in the CODWOE dataset, including English, French, Italian, Spanish, and Russian. Even if we did not achieve a good ranking in the CODWOE competition, we show that our work partially improves the current baseline from the organizers with a hypothesis on the impact of LSTM in monolingual, multilingual, and zero-shot learning. All the codes are available at https://github.com/honghanhh/codwoe2021.
This paper presents the shared task on Multilingual Idiomaticity Detection and Sentence Embedding, which consists of two subtasks: (a) a binary classification task aimed at identifying whether a sentence contains an idiomatic expression, and (b) a task based on semantic text similarity which requires the model to adequately represent potentially idiomatic expressions in context. Each subtask includes different settings regarding the amount of training data. Besides the task description, this paper introduces the datasets in English, Portuguese, and Galician and their annotation procedure, the evaluation metrics, and a summary of the participant systems and their results. The task had close to 100 registered participants organised into twenty five teams making over 650 and 150 submissions in the practice and evaluation phases respectively.
This paper describes the University of Helsinki submission to the SemEval 2022 task on multilingual idiomaticity detection. Our system utilizes several models made available by HuggingFace, along with the baseline BERT model for the task. We focus on feature engineering based on properties that typically characterize idiomatic expressions. The additional features lead to improvements over the baseline and the final submission achieves 15th place out of 20 submissions. The paper provides error analysis of our model including visualisations of the contributions of individual features.
In this paper, we describe our system for SemEval-2022 Task 2: Multilingual Idiomaticity Detection and Sentence Embedding. The task aims at detecting idiomaticity in an input sequence (Subtask A) and modeling representation of sentences that contain potential idiomatic multiword expressions (MWEs) (Subtask B) in three languages. We focus on the zero-shot setting of Subtask A and propose two span-based idiomaticity classification methods: MWE span-based classification and idiomatic MWE span prediction-based classification. We use several cross-lingual pre-trained language models (InfoXLM, XLM-R, and others) as our backbone network. Our best-performing system, fine-tuned with the span-based idiomaticity classification, ranked fifth in the zero-shot setting of Subtask A and exhibited a macro F1 score of 0.7466.
We describe the University of Alberta systems for the SemEval-2022 Task 2 on multilingual idiomaticity detection. Working under the assumption that idiomatic expressions are noncompositional, our first method integrates information on the meanings of the individual words of an expression into a binary classifier. Further hypothesizing that literal and idiomatic expressions translate differently, our second method translates an expression in context, and uses a lexical knowledge base to determine if the translation is literal. Our approaches are grounded in linguistic phenomena, and leverage existing sources of lexical knowledge. Our results offer support for both approaches, particularly the former.
We propose a unified framework that enables us to consider various aspects of contextualization at different levels to better identify the idiomaticity of multi-word expressions. Through extensive experiments, we demonstrate that our approach based on the inter- and inner-sentence context of a target MWE is effective in improving the performance of related models. We also share our experience in detail on the task of SemEval-2022 Tasks 2 such that future work on the same task can be benefited from this.
This paper describes our system for SemEval-2022 Task 2 Multilingual Idiomaticity Detection and Sentence Embedding sub-task B. We modify a standard BERT sentence transformer by adding embeddings for each idiom, which are created using BERTRAM and a small number of contexts. We show that this technique increases the quality of idiom representations and leads to better performance on the task. We also perform analysis on our final results and show that the quality of the produced idiom embeddings is highly sensitive to the quality of the input contexts.
Large Language Models have been successful in a wide variety of Natural Language Processing tasks by capturing the compositionality of the text representations. In spite of their great success, these vector representations fail to capture meaning of idiomatic multi-word expressions (MWEs). In this paper, we focus on the detection of idiomatic expressions by using binary classification. We use a dataset consisting of the literal and idiomatic usage of MWEs in English and Portuguese. Thereafter, we perform the classification in two different settings: zero shot and one shot, to determine if a given sentence contains an idiom or not. N shot classification for this task is defined by N number of common idioms between the training and testing sets. In this paper, we train multiple Large Language Models in both the settings and achieve an F1 score (macro) of 0.73 for the zero shot setting and an F1 score (macro) of 0.85 for the one shot setting. An implementation of our work can be found at https://github.com/ashwinpathak20/Idiomaticity_Detection_Using_Few_Shot_Learning.
This paper describes the experiments ran for SemEval-2022 Task 2, subtask A, zero-shot and one-shot settings for idiomaticity detection. Our main approach is based on fine-tuning transformer-based language models as a baseline to perform binary classification. Our system, CardiffNLP-Metaphor, ranked 8th and 7th (respectively on zero- and one-shot settings on this task. Our main contribution lies in the extensive evaluation of transformer-based language models and various configurations, showing, among others, the potential of large multilingual models over base monolingual models. Moreover, we analyse the impact of various input parameters, which offer interesting insights on how language models work in practice.
We present NEAMER - Named Entity Augmented Multi-word Expression Recognizer. This system is inspired by non-compositionality characteristics shared between Named Entity and Idiomatic Expressions. We utilize transfer learning and locality features to enhance idiom classification task. This system is our submission for SemEval Task 2: Multilingual Idiomaticity Detection and Sentence Embedding Subtask A OneShot shared task. We achieve SOTA with F1 0.9395 during post-evaluation phase. We also observe improvement in training stability. Lastly, we experiment with non-compositionality knowledge transfer, cross-lingual fine-tuning and locality features, which we also introduce in this paper.
Multiword expressions (MWEs) or idiomaticity are common phenomenon in natural languages. Current pre-trained language models cannot effectively capture the meaning of these MWEs. The reason is that two normal words, after combining together, could have an abruptly different meaning than the compositionality of the meanings of each word, whereas pre-trained language models reply on words compositionality. We proposed an improved method of adding an LSTM layer to the BERT model in order to get better results on a text classification task (Subtask A). Our result is slightly better than the baseline. We also tried adding TextCNN to BERT and adding both LSTM and TextCNN to BERT. We find that adding only LSTM gives the best performance.
This paper describes an approach to detect idiomaticity only from the contextualized representation of a MWE over multilingual pretrained language models.Our experiments find that larger models are usually more effective in idiomaticity detection. However, using a higher layer of the model may not guarantee a better performance.In multilingual scenarios, the convergence of different languages are not consistent and rich-resource languages have big advantages over other languages.
This paper presents our contribution to the SemEval-2022 Task 2: Multilingual Idiomaticity Detection and Sentence Embedding.We explore the impact of three different pre-trained multilingual language models in the SubTaskA.By enhancing the model generalization and robustness, we use the exponential moving average (EMA) method and the adversarial attack strategy.In SubTaskB, we add an effective cross-attention module for modeling the relationships of two sentences.We jointly train the model with a contrastive learning objective and employ a momentum contrast to enlarge the number of negative pairs.Additionally, we use the alignment and uniformity properties to measure the quality of sentence embeddings.Our approach obtained competitive results in both subtasks.
Idioms are lexically-complex phrases whose meaning cannot be derived by compositionally interpreting their components. Although the automatic identification and understanding of idioms is essential for a wide range of Natural Language Understanding tasks, they are still largely under-investigated.This motivated the organization of the SemEval-2022 Task 2, which is divided into two multilingual subtasks: one about idiomaticity detection, and the other about sentence embeddings. In this work, we focus on the first subtask and propose a Transformer-based dual-encoder architecture to compute the semantic similarity between a potentially-idiomatic expression and its context and, based on this, predict idiomaticity. Then, we show how and to what extent Named Entity Recognition can be exploited to reduce the degree of confusion of idiom identification systems and, therefore, improve performance.Our model achieves 92.1 F1 in the one-shot setting and shows strong robustness towards unseen idioms achieving 77.4 F1 in the zero-shot setting. We release our code at https://github.com/Babelscape/ner4id.
This paper will present the methods we use as the YNU-HPCC team in the SemEval-2022 Task 2, Multilingual Idiomaticity Detection and Sentence Embedding. We are involved in two subtasks, including four settings. In subtask B of sentence representation, we used novel approaches with ideas of contrastive learning to optimize model, where method of CoSENT was used in the pre-train setting, and triplet loss and multiple negatives ranking loss functions in fine-tune setting. We had achieved very competitive results on the final released test datasets. However, for subtask A of idiomaticity detection, we simply did a few explorations and experiments based on the xlm-RoBERTa model. Sentence concatenated with additional MWE as inputs did well in a one-shot setting. Sentences containing context had a poor performance on final released test data in zero-shot setting even if we attempted to extract effective information from CLS tokens of hidden layers.
We propose a multilingual adversarial training model for determining whether a sentence contains an idiomatic expression. Given that a key challenge with this task is the limited size of annotated data, our model relies on pre-trained contextual representations from different multi-lingual state-of-the-art transformer-based language models (i.e., multilingual BERT and XLM-RoBERTa), and on adversarial training, a training method for further enhancing model generalization and robustness. Without relying on any human-crafted features, knowledgebase, or additional datasets other than the target datasets, our model achieved competitive results and ranked 6thplace in SubTask A (zero-shot) setting and 15thplace in SubTask A (one-shot) setting
The same multi-word expressions may have different meanings in different sentences. They can be mainly divided into two categories, which are literal meaning and idiomatic meaning. Non-contextual-based methods perform poorly on this problem, and we need contextual embedding to understand the idiomatic meaning of multi-word expressions correctly. We use a pre-trained language model, which can provide a context-aware sentence embedding, to detect whether multi-word expression in the sentence is idiomatic usage.
We report the results of the SemEval 2022 Task 3, PreTENS, on evaluation the acceptability of simple sentences containing constructions whose two arguments are presupposed to be or not to be in an ordered taxonomic relation. The task featured two sub-tasks articulated as: (i) binary prediction task and (ii) regression task, predicting the acceptability in a continuous scale. The sentences were artificially generated in three languages (English, Italian and French). 21 systems, with 8 system papers were submitted for the task, all based on various types of fine-tuned transformer systems, often with ensemble methods and various data augmentation techniques. The best systems reached an F1-macro score of 94.49 (sub-task1) and a Spearman correlation coefficient of 0.80 (sub-task2), with interesting variations in specific constructions and/or languages.
This paper presents the results and main findings of our system on SemEval-2022 Task 3 Presupposed Taxonomies: Evaluating Neural Network Semantics (PreTENS). This task aims at semantic competence with specific attention on the evaluation of language models, which is a task with respect to the recognition of appropriate taxonomic relations between two nominal arguments. Two sub-tasks including binary classification and regression are designed for the evaluation. For the classification sub-task, we adopt the DeBERTa-v3 pre-trained model for fine-tuning datasets of different languages. Due to the small size of the training datasets of the regression sub-task, we transfer the knowledge of classification model (i.e., model parameters) to the regression task. The experimental results show that the proposed method achieves the best results on both sub-tasks. Meanwhile, we also report negative results of multiple training strategies for further discussion. All the experimental codes are open-sourced at https://github.com/WENGSYX/Semeval.
This paper describes our system created for the SemEval 2022 Task 3: Presupposed Taxonomies - Evaluating Neural-network Semantics. This task is focused on correctly recognizing taxonomic word relations in English, French and Italian. We developed various datageneration techniques that expand the originally provided train set and show that all methods increase the performance of modelstrained on these expanded datasets. Our final system outperformed the baseline system from the task organizers by achieving an average macro F1 score of 79.6 on all languages, compared to the baseline’s 67.4.
Recognizing lexical relationships between words is one of the formidable tasks in computational linguistics. It plays a vital role in the improvement of various NLP tasks. However, the diversity of word semantics, sentence structure as well as word order information make it challenging to distill the relationship effectively. To address these challenges, SemEval-2022 Task 3 introduced a shared task PreTENS focusing on semantic competence to determine the taxonomic relations between two nominal arguments. This paper presents our participation in this task where we proposed an approach through exploiting an ensemble of multilingual transformer methods. We employed two fine-tuned multilingual transformer models including XLM-RoBERTa and mBERT to train our model. To enhance the performance of individual models, we fuse the predicted probability score of these two models using weighted arithmetic mean to generate a unified probability score. The experimental results showed that our proposed method achieved competitive performance among the participants’ methods.
In human languages, there are many presuppositional constructions that impose a constrain on the taxonomic relations between two nouns depending on their order. These constructions create a challenge in validating taxonomic relations in real-world contexts. In SemEval2022-Task3 Presupposed Taxonomies: Evaluating Neural Network Semantics (PreTENS), the organizers introduced a task regarding validating the taxonomic relations within a variety of presuppositional constructions. This task is divided into two subtasks: classification and regression. Each subtask contains three datasets in multiple languages, i.e., English, Italian and French. To tackle this task, this work proposes to fine-tune different BERT-based models pre-trained on different languages. According to the experimental results, the fine-tuned BERT-based models are effective compared to the baselines in classification. For regression, the fine-tuned models show promising performance with the possibility of improvement.
Synonym and antonym practice are the most common practices in our early childhood. It correlated our known words to a better place deep in our intuition. At the beginning of life for a machine, we would like to treat the machine as a baby and built a similar training for it as well to present a qualified performance. In this paper, we present an ensemble model for sentence logistics classification, which outperforms the state-of-art methods. Our approach essentially builds on two models including ERNIE-M and DeBERTaV3. With cross validation and random seeds tuning, we select the top performance models for the last soft ensemble and make them vote for the final answer, achieving the top 6 performance.
This paper presents our strategy to address the SemEval-2022 Task 3 PreTENS: Presupposed Taxonomies Evaluating Neural Network Semantics. The goal of the task is to identify if a sentence is deemed acceptable or not, depending on the taxonomic relationship that holds between a noun pair contained in the sentence. For sub-task 1—binary classification—we propose an effective way to enhance the robustness and the generalizability of language models for better classification on this downstream task. We design a two-stage fine-tuning procedure on the ELECTRA language model using data augmentation techniques. Rigorous experiments are carried out using multi-task learning and data-enriched fine-tuning. Experimental results demonstrate that our proposed model, UU-Tax, is indeed able to generalize well for our downstream task. For sub-task 2 —regression—we propose a simple classifier that trains on features obtained from Universal Sentence Encoder (USE). In addition to describing the submitted systems, we discuss other experiments that employ pre-trained language models and data augmentation techniques. For both sub-tasks, we perform error analysis to further understand the behaviour of the proposed models. We achieved a global F1$Binary$ score of 91.25% in sub-task 1 and a rho score of 0.221 in sub-task 2.
This paper describes our system submitted for SemEval Task 3: Presupposed Taxonomies: Evaluating Neural Network Semantics (Zamparelli et al., 2022). We participated in both the binary classification and the regression subtask. Target sentences are classified according to their taxonomical relation in subtask 1 and according to their acceptability judgment in subtask 2. Our approach in both subtasks is based on a neural network BERT model. We used separate models for the three languages covered by the task, English, French, and Italian. For the second subtask, we used median averaging to construct an ensemble model. We ranked 15th out of 21 groups for subtask 1 (F1-score: 77.38%) and 11th out of 17 groups for subtask 2 (RHO: 0.078).
In the paper, we describe a unified system for task 3 of SemEval-2022. The task aims to recognize the semantic structures of sentences by providing two nominal arguments and to evaluate the degree of taxonomic relations. We utilise the strategy that adding language prefix tag in the training set, which is effective for the model. We split the training set to avoid the translation information to be learnt by the model. For the task, we propose a unified model fine-tuned on the multilingual pretrained model, XLM-RoBERTa. The model performs well in subtask 1 (the binary classification subtask). In order to verify whether our model could also perform better in subtask 2 (the regression subtask), the ranking score is transformed into classification labels by an up-sampling strategy. With the ensemble strategy, the performance of our model can be also improved. As a result, the model obtained the second place for subtask 1 and subtask 2 in the competition evaluation.
This paper presents an overview of Task 4 at SemEval-2022, which was focused on detecting Patronizing and Condescending Language (PCL) towards vulnerable communities. Two sub-tasks were considered: a binary classification task, where participants needed to classify a given paragraph as containing PCL or not, and a multi-label classification task, where participants needed to identify which types of PCL are present (if any). The task attracted more than 300 participants, 77 teams and 229 valid submissions. We provide an overview of how the task was organized, discuss the techniques that were employed by the different participants, and summarize the main resulting insights about PCL detection and categorization.
Classification of language that favors or condones vulnerable communities (e.g., refugees, homeless, widows) has been considered a challenging task and a critical step in NLP applications. Moreover, the spread of this language among people and on social media harms society and harms the people concerned. Therefore, the classification of this language is considered a significant challenge for researchers in the world. In this paper, we propose JUST-DEEP architecture to classify a text and determine if it contains any form of patronizing and condescending language (Task 4- Subtask 1). The architecture uses state-of-art pre-trained models and empowers ensembling techniques that outperform the baseline (RoBERTa) in the SemEval-2022 task4 with a 0.502 F1 score.
This paper describes the second-placed system for subtask 2 and the ninth-placed system for subtask 1 in SemEval 2022 Task 4: Patronizing and Condescending Language Detection. We propose an ensemble of prompt training and label attention mechanism for multi-label classification tasks. Transfer learning is introduced to transfer the knowledge from binary classification to multi-label classification. The experimental results proved the effectiveness of our proposed method. The ablation study is also conducted to show the validity of each technique.
The subtle and typically unconscious use of patronizing and condescending language (PCL) in large-audience media outlets undesirably feeds stereotypes and strengthens power-knowledge relationships, perpetuating discrimination towards vulnerable communities. Due to its subjective and subtle nature, PCL detection is an open and challenging problem, both for computational methods and human annotators. In this paper we describe the systems submitted by the DH-FBK team to SemEval-2022 Task 4, aiming at detecting PCL towards vulnerable communities in English media texts. Motivated by the subjectivity of human interpretation, we propose to leverage annotators’ uncertainty and disagreement to better capture the shades of PCL in a multi-task, multi-view learning framework. Our approach achieves competitive results, largely outperforming baselines and ranking on the top-left side of the leaderboard on both PCL identification and classification. Noticeably, our approach does not rely on any external data or model ensemble, making it a viable and attractive solution for real-world use.
Patronizing and condescending language (PCL) has a large harmful impact and is difficult to detect, both for human judges and existing NLP systems. At SemEval-2022 Task 4, we propose a novel Transformer-based model and its ensembles to accurately understand such language context for PCL detection. To facilitate comprehension of the subtle and subjective nature of PCL, two fine-tuning strategies are applied to capture discriminative features from diverse linguistic behaviour and categorical distribution. The system achieves remarkable results on the official ranking, including 1st in Subtask 1 and 5th in Subtask 2. Extensive experiments on the task demonstrate the effectiveness of our system and its strategies.
Patronizing behavior is a subtle form of bullying and when directed towards vulnerable communities, it can arise inequalities. This paper describes our system for Task 4 of SemEval-2022: Patronizing and Condescending Language Detection (PCL). We participated in both the sub-tasks and conducted extensive experiments to analyze the effects of data augmentation and loss functions used, to tackle the problem of class imbalance. We explore whether large transformer-based models can capture the intricacies associated with PCL detection. Our solution consists of an ensemble of the RoBERTa model which is further trained on external data and other language models such as XLNeT, Ernie-2.0, and BERT. We also present the results of several problem transformation techniques such as Classifier Chains, Label Powerset, and Binary relevance for multi-label classification.
This paper presents our solutions systems for Task4 at SemEval2022: Patronizing and Condescending Language Detection. This shared task contains two sub-tasks. The first sub-task is a binary classification task whose goal is to predict whether a given paragraph contains any form of patronising or condescending language(PCL). For the second sub-task, given a paragraph, we have to find which PCL categories express the condescension. Here we have a total of 7 overlapping sub-categories for PCL. Our proposed solution uses BERT based ensembled models with hard voting and techniques applied to take care of class imbalances. Our paper describes the system architecture of the submitted solution and other experiments that we conducted.
This paper describes the authors’ submission to the SemEval-2022 task 4: Patronizing and Condescending Language (PCL) Detection. The aim of the task is the detection and classification of PCL in an annotated dataset. Subtask 1 includes a binary classification task (PCL or not PCL). Subtask 2 is a multi label classification task where the system identifies different categories of PCL. The authors of this paper submitted two different models: one RoBERTa model and one DistilBERT model. Both systems performed better than the random and RoBERTA baseline given by the task organizers. The RoBERTA model finetuned by the authors performed better in both subtasks than the DistilBERT model.
In this description paper we outline the system architecture submitted to Task 4, Subtask 1 at SemEval-2022. We leverage the generative power of state of the art generative pretrained transformer models to increase training set size and remedy class imbalance issues. Our best submitted system is trained on a synthetically enhanced dataset with 10.3 times as many positive samples as the original dataset and reaches an F1 score of 50.62%, which is 10 percentage points higher than our initial system trained on an undersampled version of the original dataset. We explore possible reasons for the comparably low score in the overall task ranking and report on experiments conducted during the post-evaluation phase.
In this paper, we present our submission to the SemEval 2022 - Task 4 on Patronizing and Condescending Language (PCL) detection. Weapproach this problem as a traditional text classification problem with machine learning (ML)methods. We experiment and investigate theuse of various ML algorithms for detecting PCL in news articles. Our best methodology achieves an F1- Score of 0.39 for subtask1 witha rank of 63 out of 80, and F1-score of 0.082for subtask2 with a rank of 41 out of 48 on the blind dataset provided in the shared task.
An understanding of patronizing and condescending language detection is an important part of identifying and addressing discrimination and prejudice in various forms of communication. In this paper, we investigate several methods for detecting patronizing and condescending language in short statements as part of SemEval-2022 Task 4. For Task 1a, we investigate applying both lightweight (tree-based and linear) machine learning classification models and fine-tuned pre-trained large language models. Our final system achieves an F1-score of 0.4321, recall-score of 0.5016, and a precision-score of 0.3795 (ranked 53 / 78) on Task 1a.
The act of appearing kind or helpful via the use of but having a feeling of superiority condescending and patronizing language can have have serious mental health implications to those that experience it. Thus, detecting this condescending and patronizing language online can be useful for online moderation systems. Thus, in this manuscript, we describe the system developed by Team UTSA SemEval-2022 Task 4, Detecting Patronizing and Condescending Language. Our approach explores the use of several deep learning architectures including RoBERTa, convolutions neural networks, and Bidirectional Long Short-Term Memory Networks. Furthermore, we explore simple and effective methods to create ensembles of neural network models. Overall, we experimented with several ensemble models and found that the a simple combination of five RoBERTa models achieved an F-score of .6441 on the development dataset and .5745 on the final test dataset. Finally, we also performed a comprehensive error analysis to better understand the limitations of the model and provide ideas for further research.
This paper presents the AliEdalat team’s methodology and results in SemEval-2022 Task 4: Patronizing and Condescending Language (PCL) Detection. This task aims to detect the presence of PCL and PCL categories in text in order to prevent further discrimination against vulnerable communities. We use an ensemble of three basic models to detect the presence of PCL: fine-tuned bigbird, fine-tuned mpnet, and BERT+BiGRU. The ensemble model performs worse than the baseline due to overfitting and achieves an F1-score of 0.3031. We offer another solution to resolve the submitted model’s problem. We consider the different categories of PCL separately. To detect each category of PCL, we act like a PCL detector. Instead of BERT+BiGRU, we use fine-tuned roberta in the models. In PCL category detection, our model outperforms the baseline model and achieves an F1-score of 0.2531. We also present new models for detecting two categories of PCL that outperform the submitted models.
This paper describes our system for Task 4 of SemEval 2022: Patronizing and Condescending Language (PCL) Detection. For sub-task 1, where the objective is to classify a text as PCL or non-PCL, we use a T5 Model fine-tuned on the dataset. For sub-task 2, which is a multi-label classification problem, we use a RoBERTa model fine-tuned on the dataset. Given that the key challenge in this task is classification on an imbalanced dataset, our models rely on an augmented dataset that we generate using paraphrasing. We found that these two models yield the best results out of all the other approaches we tried.
In this paper, we describe our efforts at SemEval 2022 Shared Task 4 on Patronizing and Condescending Language (PCL) Detection. This is the first shared task to detect PCL which is to identify and categorize PCL language towards vulnerable communities. The shared task consists of two subtasks: Patronizing and Condescending language detection (Subtask A) which is the binary task classification and identifying the PCL categories that express the condescension (Subtask B) which is the multi-label text classification. For PCL language detection, We proposed the ensemble strategies of a system combination of BERT, Roberta, Distilbert, Roberta large, Albert achieved the official results for Subtask A with a macro f1 score of 0.5172 on the test set which is improved by baseline score. For PCL Category identification, We proposed a multi-label classification model to ensemble the various Bert-based models and the official results for Subtask B with a macro f1 score of 0.2117 on the test set which is improved by baseline score.
This paper introduces the related work and the results of Team Sapphire’s system for SemEval-2022 Task 4: Patronizing and Condescending Language Detection. We only participated in subtask 1. The task goal is to judge whether a news text contains PCL. This task can be considered as a task of binary classification of news texts. In this binary classification task, the BERT-base model is adopted as the pre-trained model used to represent textual information in vector form and encode it. Capsule networks is adopted to extract features from the encoded vectors. The official evaluation metric for subtask 1 is the F1 score over the positive class. Finally, our system’s submitted prediction results on test set achieved the score of 0.5187.
In this paper we propose four deep learning models for the task of detecting and classifying Patronizing and Condescending Language (PCL) using a corpus of over 13,000 annotated paragraphs in English. The task, hosted at SemEval-2022, consists of two different subtasks. The Subtask 1 is a binary classification problem. Namely, given a paragraph, a system must predict whether or not it contains any form of PCL. The Subtask 2 is a multi-label classification task. Given a paragraph, a system must identify which PCL categories express the condescension. A paragraph might contain one or more categories of PCL. To face with the first subtask we propose a multi-channel Convolutional Neural Network (CNN) and an Hybrid LSTM. Using the multi-channel CNN we explore the impact of parallel word emebeddings and convolutional layers involving different kernel sizes. With Hybrid LSTM we focus on extracting features in advance, thanks to a convolutional layer followed by two bidirectional LSTM layers. For the second subtask a Transformer BERT-based model (i.e. DistilBERT) and an XLNet-based model are proposed. The multi-channel CNN model is able to reach an F1 score of 0.2928, the Hybrid LSTM modelis able to reach an F1 score of 0.2815, the DistilBERT-based one an average F1 of 0.2165 and the XLNet an average F1 of 0.2296. In this paper, in addition to system descriptions, we also provide further analysis of the results, highlighting strengths and limitations.We make all the code publicly available and reusable on GitHub.
We propose the use of a contextual embedding based-neural model on strictly textual inputs to detect the presence of patronizing or condescending language (PCL). We finetuned a pre-trained BERT model to detect whether or not a paragraph contained PCL (Subtask 1), and furthermore finetuned another pre-trained BERT model to identify the linguistic techniques used to convey the PCL (Subtask 2). Results show that this approach is viable for binary classification of PCL, but breaks when attempting to identify the PCL techniques. Our system placed 32/79 for subtask 1, and 40/49 for subtask 2.
Patronizing and condescending language (PCL) can find its way into many mediums of public discourse. Presence of PCL in text can produce negative effects in the society. The challenge presented by the task emerges from the subtleties of PCL and various data dependent constraints. Hence, developing techniques to detect PCL in text, before it is propagated is vital. The aim of this paper is twofold, a) to present systems that can be used to classify a text as containing PCL or not, and b) to present systems that assign the different categories of PCL present in text. The proposed systems are primarily rooted in transformer-based pre-trained language models. Among the models submitted for Subtask 1, the best F1-Score of 0.5436 was achieved by a deep learning based ensemble model. This system secured the rank 29 in the official task ranking. For Subtask 2, the best macro-average F1-Score of 0.339 was achieved by an ensemble model combining transformer-based neural architecture with gradient boosting label-balanced classifiers. This system secured the rank 21 in the official task ranking. Among subsequently carried out experiments a variation in architecture of a system for Subtask 2 achieved a macro-average F1-Score of 0.3527.
Patronizing and Condescending Language (PCL) towards vulnerable communities in general media has been shown to have potentially harmful effects. Due to its subtlety and the good intentions behind its use, the audience is not aware of the language’s toxicity. In this paper, we present our method for the SemEval-2022 Task4 titled “Patronizing and Condescending Language Detection”. In Subtask A, a binary classification task, we introduce adversarial training based on Fast Gradient Method (FGM) and employ pre-trained model in a unified architecture. For Subtask B, framed as a multi-label classification problem, we utilize various improved multi-label cross-entropy loss functions and analyze the performance of our method. In the final evaluation, our system achieved official rankings of 17/79 and 16/49 on Subtask A and Subtask B, respectively. In addition, we explore the relationship between PCL and emotional polarity and intensity it contains.
This paper describes the system for the Semeval-2022 Task4 ”Patronizing and Condescending Language Detection”.An entity engages in Patronizing and Condescending Language(PCL) when its language use shows a superior attitude towards others or depicts them in a compassionate way. The task contains two parts. The first one is to identify whether the sentence is PCL, and the second one is to categorize PCL. Through experimental verification, the Roberta-based model will be used in our system. Respectively, for subtask 1, that is, to judge whether a sentence is PCL, the method of retraining the model with specific task data is adopted, and the method of splicing [CLS] and the keyword representation of the last three layers as the representation of the sentence; for subtask 2, that is, to judge the PCL type of the sentence, in addition to using the same method as task1, the method of selecting a special loss for Multi-label text classification is applied. We give a clear ablation experiment and give the effect of each method on the final result. Our project ranked 11th out of 79 teams participating in subtask 1 and 6th out of 49 teams participating in subtask 2.
Patronizing and condescending language (PCL) is everywhere, but rarely is the focus on its use by media towards vulnerable communities. Accurately detecting PCL of this form is a difficult task due to limited labeled data and how subtle it can be. In this paper, we describe our system for detecting such language which was submitted to SemEval 2022 Task 4: Patronizing and Condescending Language Detection. Our approach uses an ensemble of pre-trained language models, data augmentation, and optimizing the threshold for detection. Experimental results on the evaluation dataset released by the competition hosts show that our work is reliably able to detect PCL, achieving an F1 score of 55.47% on the binary classification task and a macro F1 score of 36.25% on the fine-grained, multi-label detection task.
This paper describes a system built in the SemEval-2022 competition. As participants in Task 4: Patronizing and Condescending Language Detection, we implemented the text sentiment classification system for two subtasks in English. Both subtasks involve determining emotions; subtask 1 requires us to determine whether the text belongs to the PCL category (single-label classification), and subtask 2 requires us to determine to which PCL category the text belongs (multi-label classification). Our system is based on the bidirectional encoder representations from transformers (BERT) model. For the single-label classification, our system applies a BertForSequenceClassification model to classify the input text. For the multi-label classification, we use the fine-tuned BERT model to extract the sentiment score of the text and a fully connected layer to classify the text into the PCL categories. Our system achieved relatively good results on the competition’s official leaderboard.
Patronizing and Condescending Language is an ever-present problem in our day-to-day lives. There has been a rise in patronizing language on social media platforms manifesting itself in various forms. This paper presents two performing deep learning algorithms and results for the “Task 4: Patronizing and Condescending Language Detection.” of SemEval 2022. The task incorporates an English dataset containing sentences from social media from around the world. The paper focuses on data augmentation to boost results on various deep learning methods as BERT and LSTM Neural Network.
This paper describes our system for Task 4 of SemEval 2022: Patronizing and Condescending Language Detection. Patronizing and Condescending Language (PCL) refers to language used with respect to vulnerable communities that portrays them in a pitiful way and is reflective of a sense of superiority. Task 4 involved binary classification (Subtask 1) and multi-label classification (Subtask 2) of Patronizing and Condescending Language (PCL). For our system, we experimented with fine-tuning different transformer-based pre-trained models including BERT, DistilBERT, RoBERTa and ALBERT. Further, we have used token separated metadata in order to improve our model by helping it contextualize different communities with respect to PCL. We faced the challenge of class imbalance, which we solved by experimenting with different class weighting schemes. Our models were effective in both subtasks, with the best performance coming out of models with Effective Number of Samples (ENS) class weighting and token separated metadata in both subtasks. For subtask 1 and subtask 2, our best models were finetuned BERT and RoBERTa models respectively.
This paper describes the system used by the Machine Learning Group of LTU in subtask 1 of the SemEval-2022 Task 4: Patronizing and Condescending Language (PCL) Detection. Our system consists of finetuning a pretrained text-to-text transfer transformer (T5) and innovatively reducing its out-of-class predictions. The main contributions of this paper are 1) the description of the implementation details of the T5 model we used, 2) analysis of the successes & struggles of the model in this task, and 3) ablation studies beyond the official submission to ascertain the relative importance of data split. Our model achieves an F1 score of 0.5452 on the official test set.
This paper describes my participation in the SemEval-2022 Task 4: Patronizing and Condescending Language Detection. I participate in both subtasks: Patronizing and Condescending Language (PCL) Identification and Patronizing and Condescending Language Categorization, with the main focus put on subtask 1. The experiments compare pre-BERT neural network (NN) based systems against post-BERT pretrained language model RoBERTa. This research finds NN-based systems in the experiments perform worse on the task compared to the pretrained language models. The top-performing RoBERTa system is ranked 26 out of 78 teams (F1-score: 54.64) in subtask 1, and 23 out of 49 teams (F1-score: 30.03) in subtask 2.
This paper describes the use of AutoNLP techniques applied to the detection of patronizing and condescending language (PCL) in a binary classification scenario. The proposed approach combines meta-learning, in order to identify the best performing combination of deep learning architectures, with the synthesis of adversarial training examples; thus boosting robustness and model generalization. A submission from this system was evaluated as part of the first sub-task of SemEval 2022 - Task 4 and achieved an F1 score of 0.57%, which is 16 percentage points higher than the RoBERTa baseline provided by the organizers.
A logistic regression model only fed with character and word n-grams is proposed for the SemEval-2022 Task 4 on Patronizing and Condescending Language Detection (PCL). It obtained an average level of performance, well above the performance of a system that tries to guess without using any knowledge about the task, but much lower than the best teams. To facilitate the interpretation of the performance scores, the F1 measure, the best level of performance of a system that tries to guess without using any knowledge is calculated and used to correct the F1 scores in the manner of a Kappa. As the proposed model is very similar to the one that performed well on a task requiring to automatically identify hate speech and offensive content, this paper confirms the difficulty of PCL detection.
This work describes the development of different models to detect patronising and condescending language within extracts of news articles as part of the SemEval 2022 competition (Task-4). This work explores different models based on the pre-trained RoBERTa language model coupled with LSTM and CNN layers. The best models achieved 15th rank with an F1-score of 0.5924 for subtask-A and 12th in subtask-B with a macro-F1 score of 0.3763.
This paper presents a combination of data augmentation methods to boost the performance of state-of-the-art transformer-based language models for Patronizing and Condescending Language (PCL) detection and multi-label PCL classification tasks. These tasks are inherently different from sentiment analysis because positive/negative hidden attitudes in the context will not necessarily be considered positive/negative for PCL tasks. The oblation study observes that the imbalance degree of PCL dataset is in the extreme range. This paper presents a modified version of the sentence paraphrasing deep learning model (PEGASUS) to tackle the limitation of maximum sequence length. The proposed algorithm has no specific maximum input length to paraphrase sequences. Our augmented underrepresented class of annotated data achieved competitive results among top-16 SemEval-2022 participants. This paper’s approaches rely on fine-tuning pretrained RoBERTa and GPT3 models such as Davinci and Curie engines with an extra-enriched PCL dataset. Furthermore, we discuss Few-Shot learning technique to overcome the limitation of low-resource NLP problems.
This paper details our implementations for finding Patronizing and Condescending Language in texts, as part of the SemEval Workshop Task 4. We have used a variety of methods from simple machine learning algorithms applied on bag of words, all the way to BERT models, in order to solve the binary classification and the multi-label multi-class classification.
This paper narrates the work of the team Amrita_CEN for the shared task on Patronizing and Condescending Language Detection at SemEval 2022. We implemented machine learning algorithms such as Support Vector Machine (SVV), Logistic regression, Naive Bayes, XG Boost and Random Forest for modelling the tasks. At the same time, we also applied a feature engineering method to solve the class imbalance problem with respect to training data. Among all the models, the logistic regression model outperformed all other models and we have submitted results based upon the same.
In this paper, we describe our submissions to SemEval-2022 subtask 4-A - “Patronizing and Condescending Language Detection: Binary Classification”. We developed different models for this subtask. We applied 11 supervised machine learning methods and 9 preprocessing methods. Our best submission was a model we built with BertForSequenceClassification. Our experiments indicate that pre-processing stage is a must for a successful model. The dataset for Subtask 1 is highly imbalanced dataset. The f1-scores on the oversampled imbalanced training dataset were higher the results on the original training dataset.
We describe the ULFRI system used in the Subtask 1 of SemEval-2022 Task 4 Patronizing and condescending language detection. Our models are based on the RoBERTa model, modified in two ways: (1) by injecting additional knowledge (coreferences, named entities, dependency relations, and sentiment) and (2) by leveraging the task uncertainty by using soft labels, Monte Carlo dropout, and threshold optimization.We find that the injection of additional knowledge is not helpful but the uncertainty management mechanisms lead to small but consistent improvements. Our final system based on these findings achieves F1 = 0.575 in the online evaluation, ranking 19th out of 78 systems.
Social media is an idea created to make theworld smaller and more connected. Recently,it has become a hub of fake news and sexistmemes that target women. Social Media shouldensure proper women’s safety and equality. Filteringsuch information from social media is ofparamount importance to achieving this goal.In this paper, we describe the system developedby our team for SemEval-2022 Task 5: MultimediaAutomatic Misogyny Identification. Wepropose a multimodal training methodologythat achieves good performance on both thesubtasks, ranking 4th for Subtask A (0.718macro F1-score) and 9th for Subtask B (0.695macro F1-score) while exceeding the baselineresults by good margins.
This paper describes our submission for task 5 Multimedia Automatic Misogyny Identification (MAMI) at SemEval-2022. The task is designed to detect and classify misogynous memes. To utilize both textual and visual information presented in a meme, we investigate several of the most recent visual language transformer-based multimodal models and choose ERNIE-ViL-Large as our base model. For subtask A, with observations of models’ overfitting on unimodal patterns, strategies are proposed to mitigate problems of biased words and template memes. For subtask B, we transform this multi-label problem into a multi-class one and experiment with oversampling and complementary techniques. Our approach places 2nd for subtask A and 5th for subtask B in this competition.
Research is progressing in a fast manner in the field of offensive, hate speech, abusive and sarcastic data. Tackling hate speech against women is urgent and really needed to give respect to the lady of our life. This paper describes the system used for identifying misogynous content using images and text. The system developed by the team TECHSSN uses transformer models to detect the misogynous content from text and Convolutional Neural Network model for image data. Various models like BERT, ALBERT, XLNET and CNN are explored and the combination of ALBERT and CNN as an ensemble model provides better results than the rest. This system was developed for the task 5 of the competition, SemEval 2022.
In current times, memes have become one of the most popular mediums to share jokes and information with the masses over the internet. Memes can also be used as tools to spread hatred and target women through degrading content disguised as humour. The task, Multimedia Automatic Misogyny Identification (MAMI), is to detect misogyny in these memes. This task is further divided into two sub-tasks: (A) Misogynous meme identification, where a meme should be categorized either as misogynous or not misogynous and (B) Categorizing these misogynous memes into potential overlapping subcategories. In this paper, we propose models leveraging task-specific pretraining with transfer learning on Visual Linguistic models. Our best performing models scored 0.686 and 0.691 on sub-tasks A and B respectively.
Hate speech expressions in social media are not limited to textual messages; they can appear in videos, images, or multimodal formats like memes. Existing work towards detecting such expressions has been conducted almost exclusively over textual content, and the analysis of pictures and videos has been very scarce. This paper describes our team proposal in the Multimedia Automatic Misogyny Identification (MAMI) task at SemEval 2022. The challenge consisted of identifying misogynous memes from a dataset where images and text transcriptions were provided. We reported a 71% of F-score using a multimodal system based on the CLIP model.
Online misogyny meme detection is an image/text multimodal classification task, the complicated relation of image and text challenges the intelligent system’s modality fusion learning capability. In this paper, we investigate the single-stream UNITER and dual-stream CLIP multimodal pretrained models on their capability to handle strong and weakly correlated image/text pairs. The XGBoost classifier with image features extracted by the CLIP model has the highest performance and being robust on domain shift. Based on this, we propose the PBR system, an ensemble system of Pretraining models, Boosting method and Rule-based adjustment, text information is fused into the system using our late sequential fusion scheme. Our system ranks 1st place on both sub-task A and sub-task B of the SemEval-2022 Task 5 Multimedia Automatic Misogyny Identification, with 0.834/0.731 macro F1 scores for sub-task A/B correspondingly.
Women are frequently targeted online with hate speech and misogyny using tweets, memes, and other forms of communication. This paper describes our system for Task 5 of SemEval-2022: Multimedia Automatic Misogyny Identification (MAMI). We participated in both the sub-tasks, where we used transformer-based architecture to combine features of images and text. We explore models with multi-modal pre-training (VisualBERT) and text-based pre-training (MMBT) while drawing comparative results. We also show how additional training with task-related external data can improve the model performance. We achieved sizable improvements over baseline models and the official evaluation ranked our system 3rd out of 83 teams on the binary classification task (Sub-task A) with an F1 score of 0.761, and 7th out of 48 teams on the multi-label classification task (Sub-task B) with an F1 score of 0.705.
In the context of the Multimedia Automatic Misogyny Identification (MAMI) competition 2022, we developed a framework for extracting lexical-semantic features from text and combine them with semantic descriptions of images, together with image content representation. We enriched the text modality description by incorporating word representations for each object present within the images. Images and text are then described at two levels of detail, globally and locally, using standard dimensionality reduction techniques for images in order to obtain 4 embeddings for each meme. These embeddings are finally concatenated and passed to a classifier. Our results overcome the baseline by 4%, falling behind the best performance by 12% for Sub-task B.
Gender discrimination is a serious and widespread problem on social media and online in general. Besides offensive messages, memes are one of the main means of dissemination for such content. With these premises, the MAMI task was proposed at the SemEval-2022, which consists of identifying memes with misogynous characteristics. In this work, we propose a solution to this problem based on Mask R-CNN and VisualBERT that leverages the multimodal nature of the task. Our study focuses on observing how the two sources of data in memes (text and image) and their possible combinations impact performances. Our best result slightly exceeds the higher baseline, but the experiments allowed us to draw important considerations regarding the importance of correctly exploiting the visual information and the relevance of the elements present in the memes images.
In recent times, the detection of hate-speech, offensive, or abusive language in online media has become an important topic in NLP research due to the exponential growth of social media and the propagation of such messages, as well as their impact. Misogyny detection, even though it plays an important part in hate-speech detection, has not received the same attention. In this paper, we describe our classification systems submitted to the SemEval-2022 Task 5: MAMI - Multimedia Automatic Misogyny Identification. The shared task aimed to identify misogynous content in a multi-modal setting by analysing meme images together with their textual captions. To this end, we propose two models based on the pre-trained UNITER model, one enhanced with an image sentiment classifier, whereas the second leverages a Vocabulary Graph Convolutional Network (VGCN). Additionally, we explore an ensemble using the aforementioned models. Our best model reaches an F1-score of 71.4% in Sub-task A and 67.3% for Sub-task B positioning our team in the upper third of the leaderboard. We release the code and experiments for our models on GitHub.
This work presents an ensemble system based on various uni-modal and bi-modal model architectures developed for the SemEval 2022 Task 5: MAMI-Multimedia Automatic Misogyny Identification. The challenge organizers provide an English meme dataset to develop and train systems for identifying and classifying misogynous memes. More precisely, the competition is separated into two sub-tasks: sub-task A asks for a binary decision as to whether a meme expresses misogyny, while sub-task B is to classify misogynous memes into the potentially overlapping sub-categories of stereotype, shaming, objectification, and violence. For our submission, we implement a new model fusion network and employ an ensemble learning approach for better performance. With this structure, we achieve a 0.755 macro-average F1-score (11th) in sub-task A and a 0.709 weighted-average F1-score (10th) in sub-task B.
Detecting MEME images to be misogynous or not is an application useful on curbing online hateful information against women. In the SemEval-2022 Multimedia Automatic Misogyny Identification (MAMI) challenge, we designed a system using two simple but effective principles. First, we leverage on recently emerging Transformer models pre-trained (mostly in a self-supervised learning way) on massive data sets to obtain very effective visual (V) and language (L) features. In particular, we used the CLIP model provided by OpenAI to obtain coherent V and L features and then simply used a logistic regression model to make binary predictions. Second, we emphasized more on data rather than tweaking models by following the data-centric AI principle. These principles were proven to be useful and our final macro-F1 is 0.778 for the MAMI task A and ranked the third place among participant teams.
We describe our system for the SemEval 2022 task on detecting misogynous content in memes. This is a pressing problem and we explore various methods ranging from traditional machine learning to deep learning models such as multimodal transformers. We propose a multimodal BERT architecture that uses information from both image and text. We further incorporate common world knowledge from pretrained CLIP and Urban dictionary. We also provide qualitative analysis to support out model. Our best performing model achieves an F1 score of 0.679 on Task A (Rank 5) and 0.680 on Task B (Rank 13) of the hidden test set. Our code is available at https://github.com/paridhimaheshwari2708/MAMI.
We present a multi-modal deep learning system for the Multimedia Automatic Misogyny Identification (MAMI) challenge, a SemEval task of identifying and classifying misogynistic messages in online memes. We adapt multi-task learning for the multimodal subtasks of the MAMI challenge to transfer knowledge among the correlated subtasks. We also leverage on ensemble learning for synergistic integration of models individually trained for the subtasks. We finally discuss about errors of the system to provide useful insights for future work.
In this paper, we describe the system proposed by the MilaNLP team for the Multimedia Automatic Misogyny Identification (MAMI) challenge. We use Perceiver IO as a multimodal late fusion over unimodal streams to address both sub-tasks A and B. We build unimodal embeddings using Vision Transformer (image) and RoBERTa (text transcript). We enrich the input representation using face and demographic recognition, image captioning, and detection of adult content and web entities. To the best of our knowledge, this work is the first to use Perceiver IO combining text and image modalities. The proposed approach outperforms unimodal and multimodal baselines.
We present our submission to SemEval 2022 Task 5 on Multimedia Automatic Misogyny Identification. We address the two tasks: Task A consists of identifying whether a meme is misogynous. If so, Task B attempts to identify its kind among shaming, stereotyping, objectification, and violence. Our approach combines a BERT Transformer with CLIP for the textual and visual representations. Both textual and visual encoders are fused in an early-fusion fashion through a Multimodal Bidirectional Transformer with unimodally pretrained components. Our official submissions obtain macro-averaged F1=0.727 in Task A (4th position out of 69 participants)and weighted F1=0.710 in Task B (4th position out of 42 participants).
This paper provides a comparison of different deep learning methods for identifying misogynous memes for SemEval-2022 Task 5: Multimedia Automatic Misogyny Identification. In this task, we experiment with architectures in the identification of misogynous content in memes by making use of text and image-based information. The different deep learning methods compared in this paper are: (i) unimodal image or text models (ii) fusion of unimodal models (iii) multimodal transformers models and (iv) transformers further pretrained on a multimodal task. From our experiments, we found pretrained multimodal transformer architectures to strongly outperform the models involving the fusion of representation from both the modalities.
In this paper we describe our work towards building a generic framework for both multi-modal embedding and multi-label binary classification tasks, while participating in task 5 (Multimedia Automatic Misogyny Identification) of SemEval 2022 competition.Since pretraining deep models from scratch is a resource and data hungry task, our approach is based on three main strategies. We combine different state-of-the-art architectures to capture a wide spectrum of semantic signals from the multi-modal input. We employ a multi-task learning scheme to be able to use multiple datasets from the same knowledge domain to help increase the model’s performance. We also use multiple objectives to regularize and fine tune different system components.
In this paper we present our approach and system description on Task 5 A in MAMI: Multimedia Automatic Misogyny Identification. In our experiments we compared several architectures based on deep learning algorithms with various other approaches to binary classification using Transformers, combined with a nudity image detection algorithm to provide better results. With this approach, we achieved an F1-score of 0.665 in the evaluation process
This paper describes INF-UFRGS submission for SemEval-2022 Task 5 Multimodal Automatic Misogyny Identification (MAMI). Unprecedented levels of harassment came with the ever-growing internet usage as a mean of worldwide communication. The goal of the task is to improve the quality of existing methods for misogyny identification, many of which require dedicated personnel, hence the need for automation. We experimented with five existing models, including ViLBERT and Visual BERT - both uni and multimodally pretrained - and MMBT. The datasets consist of memes with captions in English. The results show that all models achieved Macro-F1 scores above 0.64. ViLBERT was the best performer with a score of 0.698.
Nowadays, memes have become quite common in day-to-day communications on social media platforms. They appear to be amusing, evoking and attractive to audiences. However, some memes containing malicious contents can be harmful to the targeted group and arouse public anger in the long run. In this paper, we study misogynous meme detection, a shared task in SemEval 2022 - Multimedia Automatic Misogyny Identification (MAMI). The challenge of misogynous meme detection is to co-represent multi-modal features. To tackle with this challenge, we propose a Multi-modal Multi-task Variational AutoEncoder (MMVAE) to learn an effective co-representation of visual and textual features in the latent space, and determine if the meme contains misogynous information and identify its fine-grained categories. Our model achieves 0.723 on sub-task A and 0.634 on sub-task B in terms of F1 scores. We carry out comprehensive experiments on our model’s architecture and show that our approach significantly outperforms several strong uni-modal and multi-modal approaches. Our code is released on github.
Women are influential online, especially in image-based social media such as Twitter and Instagram. However, many in the network environment contain gender discrimination and aggressive information, which magnify gender stereotypes and gender inequality. Therefore, the filtering of illegal content such as gender discrimination is essential to maintain a healthy social network environment.In this paper, we describe the system developed by our team for SemEval-2022Task 5: Multimedia Automatic Misogyny Identification. More specifically, we introduce two novel system to analyze these posts: a multimodal multi-task learning architecture that combines Bertweet for text encoding with ResNet-18 for image representation, and a single-flow transformer structure which combines text embeddings from BERT-Embeddings and image embeddings from several different modules such as EfficientNet and ResNet. In this manner, we show that the information behind them can be properly revealed. Our approach achieves good performance on each of the two subtasks of the current competition, ranking 15th for Subtask A (0.746 macro F1-score), 11th for Subtask B (0.706 macro F1-score) while exceeding the official baseline results by high margins.
This paper describes the participation of the University of Hildesheim at the SemEval task 5. The task deals with Multimedia Automatic Misogyny Identification (MAMI). Hateful memes need to be detected within a data collection. For this task, we implemented six models for text and image analysis and tested the effectiveness of their combinations. A fusion system implements a multi-modal transformer to integrate the embeddings of these models. The best performing models included BERT for the text of the meme, manually derived associations for words in the memes and a Faster R-CNN network for the image. We evaluated the performance of our approach also with the data of the Facebook Hateful Memes challenge in order to analyze the generalisation capabilities of the approach.
Everyday more users are using memes on social media platforms to convey a message with text and image combined. Although there are many fun and harmless memes being created and posted, there are also ones that are hateful and offensive to particular groups of people. In this article present a novel approach based on the CLIP network to detect misogynous memes and find out the types of misogyny in that meme. We participated in Task A and Task B of the Multimedia Automatic Misogyny Identification (MaMi) challenge and our best scores are 0.694 and 0.681 respectively.
This paper presents our submission to task 5 ( Multimedia Automatic Misogyny Identification) of the SemEval 2022 competition. The purpose of the task is to identify given memes as misogynistic or not and further label the type of misogyny involved. In this paper, we present our approach based on language processing tools. We embed meme texts using GloVe embedding and classify misogyny using BERT model. Our model obtains an F1-score of 66.24% and 63.5% in misogyny classification and misogyny labels, respectively.
With the growth of the internet, the use of social media based on images has drastically increased like Twitter, Instagram, etc. In these social media, women have a very high contribution as of 75% women use social media multiple times compared to men which is only 65% of men uses social media multiple times a day. However, with this much contribution, it also increases systematic inequality and discrimination offline is replicated in online spaces in the form of MEMEs. A meme is essentially an image characterized by pictorial content with an overlaying text a posteriori introduced by humans, with the main goal of being funny and/or ironic. Although most of them are created with the intent of making funny jokes, in a short time people started to use them as a form of hate and prejudice against women, landing to sexist and aggressive messages in online environments that subsequently amplify the sexual stereotyping and gender inequality of the offline world. This leads to the need for automatic detection of Misogyny MEMEs. Specifically, I described the model submitted for the shared task on Multimedia Automatic Misogyny Identification (MAMI) and my team name is IIT DHANBAD CODECHAMPS.
In this paper, we describe our submission to the misogyny classification challenge at SemEval-2022. We propose two models for the two subtasks of the challenge: The first uses joint image and text classification to classify memes as either misogynistic or not. This model uses a majority voting ensemble structure built on traditional classifiers and additional image information such as age, gender and nudity estimations. The second model uses a RoBERTa classifier on the text transcriptions to additionally identify the type of problematic ideas the memes perpetuate. Our submissions perform above all organizer submitted baselines. For binary misogyny classification, our system achieved the fifth place on the leaderboard, with a macro F1-score of 0.665. For multi-label classification identifying the type of misogyny, our model achieved place 19 on the leaderboard, with a weighted F1-score of 0.637.
In this manuscript we describe the participation of the UMUTeam on the MAMI shared task proposed at SemEval 2022. This task is concerning the identification of misogynous content from a multi-modal perspective. Our participation is grounded on the combination of different feature sets within the same neural network. Specifically, we combine linguistic features with contextual transformers based on text (BERT) and images (BEiT). Besides, we also evaluate other ensemble learning strategies and the usage of non-contextual pretrained embeddings. Although our results are limited, we outperform all the baselines proposed, achieving position 36 in the binary classification task with a macro F1-score of 0.687, and position 28 in the multi-label task of misogynous categorisation, with an macro F1-score of 0.663.
This paper describes our system used in the SemEval-2022 Task5 Multimedia Automatic Misogyny Identification (MAMI). This task is to use the provided text-image pairs to classify emotions. In this paper, We propose a multi-label emotion classification model based on pre-trained LXMERT. We use Faster-RCNN to extract visual representation and utilize LXMERT’s cross-attention for multi-modal alignment. Then we use the Bilinear-interaction layer to fuse these features. Our experimental results surpass the F1 score of baseline. For Sub-task A, our F1 score is 0.662 and Sub-task B’s F1 score is 0.633. The code of this study is available on GitHub.
The detection of offensive, hateful content on social media is a challenging problem that affects many online users on a daily basis. Hateful content is often used to target a group of people based on ethnicity, gender, religion and other factors. The hate or contempt toward women has been increasing on social platforms. Misogynous content detection is especially challenging when textual and visual modalities are combined to form a single context, e.g., an overlay text embedded on top of an image, also known as meme. In this paper, we present a multimodal architecture that combines textual and visual features to detect misogynous memes. The proposed architecture is evaluated in the SemEval-2022 Task 5: MAMI - Multimedia Automatic Misogyny Identification challenge under the team name TIB-VA. We obtained the best result in the Task-B where the challenge is to classify whether a given document is misogynous and further identify the following sub-classes: shaming, stereotype, objectification, and violence.
This paper describes the multimodal deep learning system proposed for SemEval 2022 Task 5: MAMI - Multimedia Automatic Misogyny Identification. We participated in both Subtasks, i.e. Subtask A: Misogynous meme identification, and Subtask B: Identifying type of misogyny among potential overlapping categories (stereotype, shaming, objectification, violence). The proposed architecture uses pre-trained models as feature extractors for text and images. We use these features to learn multimodal representation using methods like concatenation and scaled dot product attention. Classification layers are used on fused features as per the subtask definition. We also performed experiments using unimodal models for setting up comparative baselines. Our best performing system achieved an F1 score of 0.757 and was ranked 3rd in Subtask A. On Subtask B, our system performed well with an F1 score of 0.690 and was ranked 10th on the leaderboard. We further show extensive experiments using combinations of different pre-trained models which will be helpful as baselines for future work.
This paper describes the multimodal late fusion model proposed in the SemEval-2022 Multimedia Automatic Misogyny Identification (MAMI) task. The main contribution of this paper is the exploration of different late fusion methods to boost the performance of the combination based on the Transformer-based model and Convolutional Neural Networks (CNN) for text and image, respectively. Additionally, our findings contribute to a better understanding of the effects of different image preprocessing methods for meme classification. We achieve 0.636 F1-macro average score for the binary subtask A, and 0.632 F1-macro average score for the multi-label subtask B. The present findings might help solve the inequality and discrimination women suffer on social media platforms.
This paper presents a deep learning system that contends at SemEval-2022 Task 5. The goal is to detect the existence of misogynous memes in sub-task A. At the same time, the advanced multi-label sub-task B categorizes the misogyny of misogynous memes into one of four types: stereotype, shaming, objectification, and violence. The Ensemble technique has been used for three multi-modal deep learning models: two MMBT models and VisualBERT. Our proposed system ranked 17 place out of 83 participant teams with an F1-score of 0.722 in sub-task A, which shows a significant performance improvement over the baseline model’s F1-score of 0.65.
Misogynistic memes are rampant on social media, and often convey their messages using multimodal signals (e.g., images paired with derogatory text or captions). However, to date very few multimodal systems have been leveraged for the detection of misogynistic memes. Recently, researchers have turned to contrastive learning, and most notably OpenAI’s CLIP model, is an innovative solution to a variety of multimodal tasks. In this work, we experiment with contrastive learning to address the detection of misogynistic memes within the context of SemEval 2022 Task 5. Although our model does not achieve top results, these experiments provide important exploratory findings for this task. We conduct a detailed error analysis, revealing promising clues and offering a foundation for follow-up work.
In recent years, there has been an upsurge in a new form of entertainment medium called memes. These memes although seemingly innocuous have transcended the boundary of online harassment against women and created an unwanted bias against them. To help alleviate this problem, we propose an early fusion model for the prediction and identification of misogynistic memes and their type in this paper for which we participated in SemEval-2022 Task 5. The model receives as input meme image with its text transcription with a target vector. Given that a key challenge with this task is the combination of different modalities to predict misogyny, our model relies on pre-trained contextual representations from different state-of-the-art transformer-based language models and pre-trained image models to get an effective image representation. Our model achieved competitive results on both SubTask-A and SubTask-B with the other competingteams and significantly outperforms the baselines.
iSarcasmEval is the first shared task to target intended sarcasm detection: the data for this task was provided and labelled by the authors of the texts themselves. Such an approach minimises the downfalls of other methods to collect sarcasm data, which rely on distant supervision or third-party annotations. The shared task contains two languages, English and Arabic, and three subtasks: sarcasm detection, sarcasm category classification, and pairwise sarcasm identification given a sarcastic sentence and its non-sarcastic rephrase. The task received submissions from 60 different teams, with the sarcasm detection task being the most popular. Most of the participating teams utilised pre-trained language models. In this paper, we provide an overview of the task, data, and participating teams.
This paper describes the method we utilized in the SemEval-2022 Task 6 iSarcasmEval: Intended Sarcasm Detection In English and Arabic. Our system has achieved 1st in SubtaskB, which is to identify the categories of intended sarcasm. The proposed system integrates multiple BERT-based, RoBERTa-based and BERTweet-based models with finetuning. In this task, we contributed the following: 1) we reveal several large pre-trained models’ performance on tasks coping with the tweet-like text. 2) Our methods prove that we can still achieve excellent results in this particular task without a complex classifier adopting some proper training method. 3) we found there is a hierarchical relationship of sarcasm types in this task.
This paper describes the systematic approach applied in “SemEval-2022 Task 6 (iSarcasmEval) : Intended Sarcasm Detection in English and Arabic”. In particular, we illustrate the proposed system in detail for SubTask-A about determining a given text as sarcastic or non-sarcastic in English. We start with the training data from the officially released data and then experiment with different combinations of public datasets to improve the model generalization. Additional experiments conducted on the task demonstrate our strategies are effective in completing the task. Different transformer-based language models, as well as some popular plug-and-play proirs, are mixed into our system to enhance the model’s robustness. Furthermore, statistical and lexical-based text features are mined to improve the accuracy of the sarcasm detection. Our final submission achieves an F1-score for the sarcastic class of 0.6052 on the official test set (the top 1 of the 43 teams in “SubTask-A-English” on the leaderboard).
Sarcasm refers to the use of words that have different literal and intended meanings. It represents the usage of words that are opposite of what is literally said, especially in order to insult, mock, criticise or irritate someone. These types of statements may be funny or amusing to others but may hurt or annoy the person towards whom it is intended. Identification of sarcastic phrases from social media posts finds its application in different domains like sentiment analysis, opinion mining, author profiling, and harassment detection. We have proposed a model for the shared task iSarcasmEval - Intended Sarcasm Detection in English and Arabic (CITATION) by SemEval-2022 considering the language English based on ELmo embeddings for Subtasks A and C and TF-IDF vectors and Gaussian Naive bayes classifier for Subtask B. The proposed model resulted in a F1 score 0.2012 for sarcastic texts in Subtask A, macro-F1 score of 0.0387 and 0.2794 for Subtasks B and C respectively.
This paper describes the submission of the team Amrita_CEN to the shared task on iSarcasm Eval: Intended Sarcasm Detection in English and Arabic at SemEval 2022. We employed machine learning algorithms towards sarcasm detection. Here, we used K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Naïve Bayes, Logistic Regression, and Decision Tree along with the Random Forest ensemble method. Additionally, feature engineering techniques were applied to deal with the problems of class imbalance during training. Among the models considered, our study shows that the SVM, logistic regression and ensemble model Random Forest exhibited the best performance, which was submitted to the shared task.
This paper presents our proposed methods for the iSarcasmEval shared task. The shared task consists of three different subtasks. We participate in both subtask A and subtask C. The purpose of subtask A was to predict if a text is sarcastic while the aim of subtask C is to determine which text is sarcastic given a sarcastic text and its non-sarcastic rephrase. Both of the developed solutions used BERT pre-trained models. The proposed models are optimized on simple objectives and are easy to grasp. However, despite their simplicity, our methods ranked 4 and 2 in iSarcasmEval subtask A and subtask C for Arabic texts.
Sarcasm is a form of figurative language where the intended meaning of a sentence differs from its literal meaning. This poses a serious challenge to several Natural Language Processing (NLP) applications such as Sentiment Analysis, Opinion Mining, and Author Profiling. In this paper, we present our participating system to the intended sarcasm detection task in English and Arabic languages. Our system consists of three deep learning-based models leveraging two existing pre-trained language models for Arabic and English. We have participated in all sub-tasks. Our official submissions achieve the best performance on sub-task A for Arabic language and rank second in sub-task B. For sub-task C, our system is ranked 7th and 11th on Arabic and English datasets, respectively.
Irony detection in the social media is an upcoming research which places a main role in sentiment analysis and offensive language identification. Sarcasm is one form of irony that is used to provide intended comments against realism. This paper describes a method to detect intended sarcasm in text (SemEval-2022 Task 6). The TECHSSN team used Bidirectional Encoder Representations from Transformers (BERT) models and its variants to classify the text as sarcastic or non-sarcastic in English and Arabic languages. The data is preprocessed and fed to the model for training. The transformer models learn the weights during the training phase from the given dataset and predicts the output class labels for the unseen test data.
Sarcasm is often expressed through several verbal and non-verbal cues, e.g., a change of tone, overemphasis in a word, a drawn-out syllable, or a straight looking face. Most of the recent work in sarcasm detection has been carried out on textual data. This paper describes how the problem proposed in Task 6: Intended Sarcasm Detection in English (Abu Arfa et al. 2022) has been solved. Specifically, we participated in Subtask B: a binary multi-label classification task, where it is necessary to determine whether a tweet belongs to an ironic speech category, if any. Several approaches (classic machine learning and deep learning algorithms) were developed. The final submission consisted of a BERT based model and a macro-F1 score of 0.0699 was obtained.
The intended sarcasm cannot be understood until the listener observes that the text’s literal meaning violates truthfulness. Consequently, words and meanings play an essential role in specifying sarcasm. Enriched feature extraction techniques were proposed to capture both words and meanings in the contexts. Due to the overlapping features in sarcastic and non-sarcastic texts, a CNN model extracts local features from the combined class-dependent statistical embedding of sarcastic texts with contextualized embedding. Another component BiLSTM extracts long dependencies from combined non-sarcastic statistical and contextualized embeddings. This work combines a classifier that uses the combined high-level features of CNN and BiLSTM for sarcasm detection to produce the final predictions. The experimental analysis presented in this paper shows the effectiveness of the proposed method.
Sarcasm has gained notoriety for being difficult to detect by machine learning systems due to its figurative nature. In this paper, Bidirectional Encoder Representations from Transformers (BERT) model has been used with ensemble loss made of cross-entropy loss and negative log-likelihood loss to classify whether a given sentence is in English and Arabic tweets are sarcastic or not. From the results obtained in the experiments, our proposed BERT with ensemble loss achieved superior performance when applied to English and Arabic test datasets. For the validation dataset, our model performed better on the Arabic dataset but failed to outperform the baseline method (made of BERT with only a single loss function) when applied on the English validation set.
In this paper we present our approach and system description on iSarcasmEval: a SemEval task for intended sarcasm detection on social networks. This derives from our participation in SubTask A: Given a text, determine whether it is sarcastic or non-sarcastic. In our approach to complete the task, a comparison of several machine learning and deep learning algorithms using two datasets was conducted. The model which obtained the highest values of F1-score was a BERT-base-cased model. With this one, an F1-score of 0.2451 for the sarcastic class in the evaluation process was achieved. Finally, our team reached the 30th position.
This paper describes the systems submitted to iSarcasm shared task. The aim of iSarcasm is to identify the sarcastic contents in Arabic and English text. Our team participated in iSarcasm for the Arabic language. A multi-Layer machine learning based model has been submitted for Arabic sarcasm detection. In this model, a vector space TF-IDF has been used as for feature representation. The submitted system is simple and does not need any external resources. The test results show encouraging results.
Due to the widespread usage of social media sites and the enormous number of users who utilize irony implicit words in most of their tweets and posts, it has become necessary to detect sarcasm, which strongly influences understanding and analyzing the crowd’s opinions. Detecting sarcasm is difficult due to the nature of sarcastic tweets, which vary based on the topic, region, the user’s attitude, culture, terminologies, and other criteria. In addition to these difficulties, detecting sarcasm in Arabic has its challenges due to its complexities, such as being morphologically rich, having many different dialects, and having low resources.In this research, we present our submission of (iSarcasmEval) sub-task A of the shared task on SemEval 2022. In Sub-task A; we determine whether the tweets are sarcastic or non-sarcastic. We implemented different approaches based on Transformers. First, we fine-tuned the AraBERT, MARABERT, and AraELECTRA. One of the challenges that faced us was that the data was not balanced. Non-sarcastic data is much more than sarcastic. We used data augmentation techniques to balance the two classes, significantly affecting the performance. The performance F1 score of the three models was 87%, 90%, and 91%, respectively. Then we boosted the three models by developing an ensemble model based on hard voting. The final performance F1 Score was 93%.
Sarcasm detection is an important task in Natural Language Understanding. Sarcasm is a form of verbal irony that occurs when there is a discrepancy between the literal and intended meanings of an expression. In this paper, we use the tweets of the Arabic dataset provided by SemEval-2022 task 6 to train deep learning classifiers to solve the sub-tasks A and C associated with the dataset. Sub-task A is to determine if the tweet is sarcastic or not. For sub-task C, given a sarcastic text and its non-sarcastic rephrase, i.e. two texts that convey the same meaning, determine which is the sarcastic one. In our solution, we utilize fine-tuned MARBERT (Abdul-Mageed et al., 2021) model with an added single linear layer on top for classification. The proposed solution achieved 0.5076 F1-sarcastic in Arabic sub-task A, accuracy of 0.7450 and F-score of 0.7442 in Arabic sub-task C. We achieved the 2nd and the 9th places for Arabic sub-tasks A and C respectively.
This paper describes the system used in SemEval-2022 Task 6: Intended Sarcasm Detection in English and Arabic. Achieving 20th,3rd places with 34& 47 F1-Sarcastic score for task A, 16th place for task B with 0.0560 F1-macro score, and 10, 6th places for task C with72% and 80% accuracy on the leaderboard. A voting classifier between either multiple different BERT-based models or machine learningmodels is proposed, as our final model. Multiple key points has been extensively examined to overcome the problem of the unbalance ofthe dataset as: type of models, suitable architecture, augmentation, loss function, etc. In addition to that, we present an analysis of ourresults in this work, highlighting its strengths and shortcomings.
This paper presents the 10th and 11th place system for Subtask A -English and Subtask A Arabic respectively of the SemEval 2022 -Task 6. The purpose of the Subtask A was to classify a given text sequence into sarcastic and nonsarcastic. We also breifly cover our method for Subtask B which performed subpar when compared with most of the submissions on the official leaderboard . All of the developed solutions used a transformers based language model for encoding the text sequences with necessary changes of the pretrained weights and classifier according to the language and subtask at hand .
This paper introduces the result of Team Dartmouth’s experiments on each of the five subtasks for the detection of sarcasm in English and Arabic tweets. This detection was framed as a classification problem, and our contributions are threefold: we developed an English binary classifier system with RoBERTa, an Arabic binary classifier with XLM-RoBERTa, and an English multilabel classifier with BERT. Preprocessing steps are taken with labeled input data prior to tokenization, such as extracting and appending verbs/adjectives or representative/significant keywords to the end of an input tweet to help the models better understand and generalize sarcasm detection. We also discuss the results of simple data augmentation techniques to improve the quality of the given training dataset as well as an alternative approach to the question of multilabel sequence classification. Ultimately, our systems place us in the top 14 participants for each of the five subtasks.
A robust comprehension of sarcasm detection iscritical for creating artificial systems that can ef-fectively perform sentiment analysis in writtentext. In this work, we investigate AI approachesto identifying whether a text is sarcastic or notas part of SemEval-2022 Task 6. We focus oncreating systems for Task A, where we experi-ment with lightweight statistical classificationapproaches trained on both GloVe features andmanually-selected features. Additionally, weinvestigate fine-tuning the transformer modelBERT. Our final system for Task A is an Ex-treme Gradient Boosting Classifier trained onmanually-engineered features. Our final sys-tem achieved an F1-score of 0.2403 on SubtaskA and was ranked 32 of 43.
The paper describes our submission to SemEval-2022 Task 6 on sarcasm detection and its five subtasks for English and Arabic. Sarcasm conveys a meaning which contradicts the literal meaning, and it is mainly found on social networks. It has a significant role in understanding the intention of the user. For detecting sarcasm, we used deep learning techniques based on transformers due to its success in the field of Natural Language Processing (NLP) without the need for feature engineering. The datasets were taken from tweets. We created new datasets by augmenting with external data or by using word embeddings and repetition of instances. Experiments were done on the datasets with different types of preprocessing because it is crucial in this task. The rank of our team was consistent across four subtasks (fourth rank in three subtasks and sixth rank in one subtask); whereas other teams might be in the top ranks for some subtasks but rank drastically less in other subtasks. This implies the robustness and stability of the models and the techniques we used.
This paper describes the system architectures and the models submitted by our team “IISERB Brains” to SemEval 2022 Task 6 competition. We contested for all three sub-tasks floated for the English dataset. On the leader-board, we got 19th rank out of 43 teams for sub-task A, 8th rank out of 22 teams for sub-task B, and 13th rank out of 16 teams for sub-task C. Apart from the submitted results and models, we also report the other models and results that we obtained through our experiments after organizers published the gold labels of their evaluation data. All of our code and links to additional resources are present in GitHub for reproducibility.
We investigated the influence of contradictory connotations of words or phrases occurring in sarcastic statements, causing those statements to convey the opposite of their literal meaning. Our approach was to perform a sentiment analysis in order to capture potential opposite sentiments within one sentence and use its results as additional information for a further classifier extracting general text features, testing this for a Convolutional Neural Network, as well as for a Support Vector Machine classifier, respectively.We found that a more complex and sophisticated implementation of the sentiment analysis than just classifying the sentences as positive or negative is necessary, since our implementation showed a worse performance in both approaches than the respective classifier without using any sentiment analysis.
We present our systems and findings for the iSarcasmEval: Intended Sarcasm Detection In English and Arabic at SEMEVAL 2022. Specifically we take part in Subtask A for the English language. The task aims to determine whether a text from social media (a tweet) is sarcastic or not. We model the problem using knowledge sources, a pre-trained language model on sentiment/emotion data and a dataset focused on intended sarcasm. Our submission ranked third place among 43 teams. In addition, we show a brief error analysis of our best model to investigate challenging examples for detecting sarcasm.
In this paper, we (a YNU-HPCC team) describe the system we built in the SemEval-2022 competition. As participants in Task 6 (titled “iSarcasmEval: Intended Sarcasm Detection In English and Arabic”), we implement the sentiment system for all three subtasks in English and Arabic. All subtasks involve the detection of sarcasm (binary and multilabel classification) and the determination of the sarcastic text location (sentence pair classification). Our system primarily applies the sequence classification model of a bidirectional encoder representation from a transformer (BERT). The BERT is used to extract sentence information from both directions for downstream classification tasks. A single basic model is used for single-sentence and sentence-pair binary classification tasks. For the multilabel task, the Label-Powerset method and binary cross-entropy loss function with weights are used. Our system exhibits competitive performance, obtaining 12/43 (21/32), 11/22, and 3/16 (8/13) rankings in the three official rankings for English (Arabic).
Sarcasm is a term that refers to the use of words to mock, irritate, or amuse someone. It is commonly used on social media. The metaphorical and creative nature of sarcasm presents a significant difficulty for sentiment analysis systems based on affective computing. The methodology and results of our team, UTNLP, in the SemEval-2022 shared task 6 on sarcasm detection are presented in this paper. We put different models, and data augmentation approaches to the test and report on which one works best. The tests begin with traditional machine learning models and progress to transformer-based and attention-based models. We employed data augmentation based on data mutation and data generation. Using RoBERTa and mutation-based data augmentation, our best approach achieved an F1-score of 0.38 in the competition’s evaluation phase. After the competition, we fixed our model’s flaws and achieved anF1-score of 0.414.
The “iSarcasmEval - Intended Sarcasm Detection in English and Arabic” task at the SemEval 2022 competition focuses on detectingand rating the distinction between intendedand perceived sarcasm in the context of textual sarcasm detection, as well as the level ofirony contained in these texts. In the contextof SemEval, we present a binary classificationmethod which classifies the text as sarcasticor non-sarcastic (task A, for English) based onfive classical machine learning approaches bytrying to train the models based on this datasetsolely (i.e., no other datasets have been used).This process indicates low performance compared to previously studied datasets, which in2dicates that the previous ones might be biased.
The paper describes SemEval-2022’s shared task “Intended Sarcasm Detection in English and Arabic.” This task includes English and Arabic tweets with sarcasm and non-sarcasm samples and irony speech labels.The first two subtasks predict whether a text is sarcastic and the ironic category the sarcasm sample belongs to. The third one is to find the sarcastic sample from its non-sarcastic paraphrase. Deep neural networks have recently achieved highly competitive performance in many tasks.Combining deep learning with language models has also resulted in acceptable accuracy. Inspired by this, we propose a novel deep learning model on top of language models. On top of T5, this architecture uses an encoder module of the transformer, followed by LSTM and attention to utilizing past and future information, concentrating on informative tokens. Due to the success of the proposed model, we used the same architecture with a few modifications to the output layer in all three subtasks.
This paper describes the approach developed by the LT3 team in the Intended Sarcasm Detection task at SemEval-2022 Task 6. We considered the binary classification subtask A for English data. The presented system is based on the fuzzy-rough nearest neighbor classification method using various text embedding techniques. Our solution reached 9th place in the official leader-board for English subtask A.
In this paper, we present our system and findings for SemEval-2022 Task 6 - iSarcasmEval: Intended Sarcasm Detection in English. The main objective of this task was to identify sarcastic tweets. This task was challenging mainly due to (1) the small training dataset that contains only 3468 tweets and (2) the imbalanced class distribution (25% sarcastic and 75% non-sarcastic). Our submitted model (ranked eighth on Sub-Task A and fifth on Sub-Task C) consists of a Transformer-based approach (BERTweet model).
Detecting sarcasm and verbal irony from people’s subjective statements is crucial to understanding their intended meanings and real sentiments and positions in social scenarios. This paper describes the X-PuDu system that participated in SemEval-2022 Task 6, iSarcasmEval - Intended Sarcasm Detection in English and Arabic, which aims at detecting intended sarcasm in various settings of natural language understanding. Our solution finetunes pre-trained language models, such as ERNIE-M and DeBERTa, under the multilingual settings to recognize the irony from Arabic and English texts. Our system ranked second out of 43, and ninth out of 32 in Task A: one-sentence detection in English and Arabic; fifth out of 22 in Task B: binary multi-label classification in English; first out of 16, and fifth out of 13 in Task C: sentence-pair detection in English and Arabic.
This paper describes the participation of team DUCS at SemEval 2022 Task 6: iSarcasmEval - Intended Sarcasm Detection in English and Arabic. Team DUCS participated in SubTask A of iSarcasmEval which was to determine if the given English text was sarcastic or not. In this work, emojis were utilized to capture how they contributed to the sarcastic nature of a text. It is observed that emojis can augment or reverse the polarity of a given statement. Thus sentiment polarities and intensities of emojis, as well as those of text, were computed to determine sarcasm. Use of capitalization, word repetition, and use of punctuation marks like '!' were factored in as sentiment intensifiers. An NLP augmenter was used to tackle the imbalanced nature of the sarcasm dataset. Several architectures comprising of various ML and DL classifiers, and transformer models like BERT and Multimodal BERT were experimented with. It was observed that Multimodal BERT outperformed other architectures tested and achieved an F1-score of 30.71%. The key takeaway of this study was that sarcastic texts are usually positive sentences. In general emojis with positive polarity are used more than those with negative polarities in sarcastic texts.
In this manuscript we detail the participation of the UMUTeam in the iSarcasm shared task (SemEval-2022). This shared task is related to the identification of sarcasm in English and Arabic documents. Our team achieve in the first challenge, a binary classification task, a F1 score of the sarcastic class of 17.97 for English and 31.75 for Arabic. For the second challenge, a multi-label classification, our results are not recorded due to an unknown problem. Therefore, we report the results of each sarcastic mechanism with the validation split. For our proposal, several neural networks that combine language-independent linguistic features with pre-trained embeddings are trained. The embeddings are based on different schemes, such as word and sentence embeddings, and contextual and non-contextual embeddings. Besides, we evaluate different techniques for the integration of the feature sets, such as ensemble learning and knowledge integration. In general, our best results are achieved using the knowledge integration strategy.
This paper describes our system used for SemEval 2022 Task 6: iSarcasmEval: Intended Sarcasm Detection in English and Arabic. We participated in all subtasks based on only English datasets. Pre-trained Language Models (PLMs) have become a de-facto approach for most natural language processing tasks. In our work, we evaluate the performance of these models for identifying sarcasm. For Subtask A and Subtask B, we used simple finetuning on PLMs. For Subtask C, we propose a Siamese network architecture trained using a combination of cross-entropy and distance-maximisation loss. Our model was ranked 7th in Subtask B, 8th in Subtask C (English), and performed well in Subtask A (English). In our work, we also present the comparative performance of different PLMs for each Subtask.
This paper presents solution systems for task 6 at SemEval2022, iSarcasmEval: Intended Sarcasm Detection In English and Arabic. The shared task 6 consists of three sub-task. We participated in subtask A for both languages, Arabic and English. The goal of subtask A is to predict if a tweet would be considered sarcastic or not. The proposed solution SarcasmDet has been developed using the state-of-the-art Arabic and English pre-trained models AraBERT, MARBERT, BERT, and RoBERTa with ensemble techniques. The paper describes the SarcasmDet architecture with the fine-tuning of the best hyperparameter that led to this superior system. Our model ranked seventh out of 32 teams in subtask A- Arabic with an f1-sarcastic of 0.4305 and Seventeen out of 42 teams with f1-sarcastic 0.3561. However, we built another model to score f-1 sarcastic with 0.43 in English after the deadline. Both Models (Arabic and English scored 0.43 as f-1 sarcastic with ranking seventh).
In this paper, we describe our submissions to SemEval-2022 contest. We tackled subtask 6-A - “iSarcasmEval: Intended Sarcasm Detection In English and Arabic – Binary Classification”. We developed different models for two languages: English and Arabic. We applied 4 supervised machine learning methods, 6 preprocessing methods for English and 3 for Arabic, and 3 oversampling methods. Our best submitted model for the English test dataset was a SVC model that balanced the dataset using SMOTE and removed stop words. For the Arabic test dataset our best submitted model was a SVC model that preprocessed removed longation.
We describe SemEval-2022 Task 7, a shared task on rating the plausibility of clarifications in instructional texts. The dataset for this task consists of manually clarified how-to guides for which we generated alternative clarifications and collected human plausibility judgements. The task of participating systems was to automatically determine the plausibility of a clarification in the respective context. In total, 21 participants took part in this task, with the best system achieving an accuracy of 68.9%. This report summarizes the results and findings from 8 teams and their system descriptions. Finally, we show in an additional evaluation that predictions by the top participating team make it possible to identify contexts with multiple plausible clarifications with an accuracy of 75.2%.
In this study, we examine the ability of contextualized representations of pretrained language model to distinguish whether sequences from instructional articles are plausible or implausible. Towards this end, we compare the BERT, RoBERTa, and DeBERTa models using simple classifiers based on the sentence representations of the [CLS] tokens and perform a detailed analysis by visualizing the representations of the [CLS] tokens of the models. In the experimental results of Subtask A: Multi-Class Classification, DeBERTa exhibits the best performance and produces a more distinguishable representation across different labels. Submitting an ensemble of 10 DeBERTa-based models, our final system achieves an accuracy of 61.4% and is ranked fifth out of models submitted by eight teams. Further in-depth results suggest that the abilities of pretrained language models for the plausibility detection task are more strongly affected by their model structures or attention designs than by their model sizes.
This paper describes the system for the identifying Plausible Clarifications of Implicit and Underspecified Phrases. This task was set up as an English cloze task, in which clarifications are presented as possible fillers and systems have to score how well each filler plausibly fits in a given context. For this shared task, we propose our own solutions, including supervised proaches, unsupervised approaches with pretrained models, and then we use these models to build an ensemble model. Finally we get the 2nd best result in the subtask1 which is a classification task, and the 3rd best result in the subtask2 which is a regression task.
This paper describes the DuluthNLP system that participated in Task 7 of SemEval-2022 on Identifying Plausible Clarifications of Implicit and Underspecified Phrases in Instructional Texts. Given an instructional text with an omitted token, the task requires models to classify or rank the plausibility of potential fillers. To solve the task, we fine–tuned the models BERT, RoBERTa, and ELECTRA on training data where potential fillers are rated for plausibility. This is a challenging problem, as shown by BERT-based models achieving accuracy less than 45%. However, our ELECTRA model with tuned class weights on CrossEntropyLoss achieves an accuracy of 53.3% on the official evaluation test data, which ranks 6 out of the 8 total submissions for Subtask A.
In this paper, we detail the methods we used to determine the idiomaticity and plausibility of candidate words or phrases into an instructional text as part of the SemEval Task 7: Identifying Plausible Clarifications of Implicit and Underspecified Phrases in Instructional Texts. Given a set of steps in an instructional text, there are certain phrases that most plausibly fill that spot. We explored various possible architectures, including tree-based methods over GloVe embeddings, ensembled BERT and ELECTRA models, and GPT 2-based infilling methods.
This paper describes our winning system on SemEval 2022 Task 7: \textit{Identifying Plausible Clarifications ofImplicit and Underspecified Phrases in Instructional Texts}. A replaced token detection pre-trained model is utilized with minorly different task-specific heads for SubTask-A: \textit{Multi-class Classification} and SubTask-B: \textit{Ranking}. Incorporating a pattern-aware ensemble method, our system achieves a 68.90% accuracy score and 0.8070 spearman’s rank correlation score surpassing the 2nd place with a large margin by 2.7 and 2.2 percent points for SubTask-A and SubTask-B, respectively. Our approach is simple and easy to implement, and we conducted ablation studies and qualitative and quantitative analyses for the working strategies used in our system.
This paper describes the 9th place system description for SemEval-2022 Task 7. The goal of this shared task was to develop computational models to predict how plausible a clarification made on an instructional text is. This shared task was divided into two Subtasks A and B. We attempted to solve these using various transformers-based architecture under different regime. We initially treated this as a text2text generation problem but comparing it with our recent approach we dropped it and treated this as a text-sequence classification and regression depending on the Subtask.
Thousands of new news articles appear daily in outlets in different languages. Understanding which articles refer to the same story can not only improve applications like news aggregation but enable cross-linguistic analysis of media consumption and attention. However, assessing the similarity of stories in news articles is challenging due to the different dimensions in which a story might vary, e.g., two articles may have substantial textual overlap but describe similar events that happened years apart. To address this challenge, we introduce a new dataset of nearly 10,000 news article pairs spanning 18 language combinations annotated for seven dimensions of similarity as SemEval 2022 Task 8. Here, we present an overview of the task, the best performing submissions, and the frontiers and challenges for measuring multilingual news article similarity. While the participants of this SemEval task contributed very strong models, achieving up to 0.818 correlation with gold standard labels across languages, human annotators are capable of reaching higher correlations, suggesting space for further progress.
In this paper, we present the participation of the EMBEDDIA team in the SemEval-2022 Task 8 (Multilingual News Article Similarity). We cover several techniques and propose different methods for finding the multilingual news article similarity by exploring the dataset in its entirety. We take advantage of the textual content of the articles, the provided metadata (e.g., titles, keywords, topics), the translated articles, the images (those that were available), and knowledge graph-based representations for entities and relations present in the articles. We, then, compute the semantic similarity between the different features and predict through regression the similarity scores. Our findings show that, while our proposed methods obtained promising results, exploiting the semantic textual similarity with sentence representations is unbeatable. Finally, in the official SemEval-2022 Task 8, we ranked fifth in the overall team ranking cross-lingual results, and second in the English-only results.
This paper describes our system designed for SemEval-2022 Task 8: Multilingual News Article Similarity. We proposed a linguistics-inspired model trained with a few task-specific strategies. The main techniques of our system are: 1) data augmentation, 2) multi-label loss, 3) adapted R-Drop, 4) samples reconstruction with the head-tail combination. We also present a brief analysis of some negative methods like two-tower architecture. Our system ranked 1st on the leaderboard while achieving a Pearson’s Correlation Coefficient of 0.818 on the official evaluation set.
This paper describes the second-placed system on the leaderboard of SemEval-2022 Task 8: Multilingual News Article Similarity. We propose an entity-enriched Siamese Transformer which computes news article similarity based on different sub-dimensions, such as the shared narrative, entities, location and time of the event discussed in the news article. Our system exploits a Siamese network architecture using a Transformer encoder to learn document-level representations for the purpose of capturing the narrative together with the auxiliary entity-based features extracted from the news articles. The intuition behind using all these features together is to capture the similarity between news articles at different granularity levels and to assess the extent to which different news outlets write about “the same events”. Our experimental results and detailed ablation study demonstrate the effectiveness and the validity of our proposed method.
This work is about finding the similarity between a pair of news articles. There are seven different objective similarity metrics provided in the dataset for each pair and the news articles are in multiple different languages. On top of the pre-trained embedding model, we calculated cosine similarity for baseline results and feed-forward neural network was then trained on top of it to improve the results. We also built separate pipelines for each similarity metric for feature extraction. We could see significant improvement from baseline results using feature extraction and feed-forward neural network.
This paper describes our contribution to SemEval 2022 Task 8: Multilingual News Article Similarity. The aim was to test completely different approaches and distinguish the best performing. That is why we’ve considered systems based on Transformer-based encoders, NER-based, and NLI-based methods (and their combination with SVO dependency triplets representation). The results prove that Transformer models produce the best scores. However, there is space for research and approaches that give not yet comparable but more interpretable results.
The task of multilingual news article similarity entails determining the degree of similarity of a given pair of news articles in a language-agnostic setting. This task aims to determine the extent to which the articles deal with the entities and events in question without much consideration of the subjective aspects of the discourse. Considering the superior representations being given by these models as validated on other tasks in NLP across an array of high and low-resource languages and this task not having any restricted set of languages to focus on, we adopted using the encoder representations from these models as our choice throughout our experiments. For modeling the similarity task by using the representations given by these models, a Siamese architecture was used as the underlying architecture. In experimentation, we investigated on several fronts including features passed to the encoder model, data augmentation and ensembling among our major experiments. We found data augmentation to be the most effective working strategy among our experiments.
This paper describes our submission to SemEval-2022 Multilingual News Article Similarity task. We experiment with different approaches that utilize a pre-trained language model fitted with a regression head to predict similarity scores for a given pair of news articles. Our best performing systems include 2 key steps: 1) pre-training with in-domain data 2) training data enrichment through machine translation. Our final submission is an ensemble of predictions from our top systems. While we show the significance of pre-training and augmentation, we believe the issue of language coverage calls for more attention.
This paper presents our approach for tackling SemEval-2022 Task 8: Multilingual News Article Similarity. Our experiments show that even by using multi-lingual pre-trained language models (LMs), translating the text into the same language yields the best evaluation performance. We also find that stylometric features of the text and meta-information of the news articles can be predicted based on the text with low error rates, and these predictions could be used to improve the predictions of the overall similarity scores. These findings suggest substantial correlations between authorship information and topical similarity estimation, which sheds light on future stylometric and topic modeling research.
This work represents the system proposed by team Innovators for SemEval 2022 Task 8: Multilingual News Article Similarity. Similar multilingual news articles should match irrespective of the style of writing, the language of conveyance, and subjective decisions and biases induced by medium/outlet. The proposed architecture includes a machine translation system that translates multilingual news articles into English and presents a multitask learning model trained simultaneously on three distinct datasets. The system leverages the PageRank algorithm for Long-form text alignment. Multitask learning approach allows simultaneous training of multiple tasks while sharing the same encoder during training, facilitating knowledge transfer between tasks. Our best model is ranked 16 with a Pearson score of 0.733.
We investigate the capabilities of pre-trained models, without any fine-tuning, for a document-level multilingual news similarity task of SemEval-2022. We utilize title and news content with appropriate pre-processing techniques. Our system derives 14 different similarity features using a combination of state-of-the-art methods (MPNet) with well-known statistical methods (i.e. TF-IDF, Word Mover’s distance). We formulate multilingual news similarity task as a regression task and approximate the overall similarity between two news articles using these features. Our best-performing system achieved a correlation score of 70.1% and was ranked 20th among the 34 participating teams. In this paper, in addition to a system description, we also provide further analysis of our results and an ablation study highlighting the strengths and limitations of our features. We make our code publicly available at https://github.com/cicl-iscl/multinewssimilarity
We present our contribution to the SemEval 22 Share Task 8: Multilingual news article similarity. The approach is lightweight and language-agnostic, it is based on the computation of several lexicographic and embedding-based features, and the use of a simple ML approach: random forests. In a notable departure from the task formulation, which is a ranking task, we tackled this task as a classification one. We present a detailed analysis of the behaviour of our system under different settings.
This article introduces a system to solve the SemEval 2022 Task 8: Multilingual News Article Similarity. The task focuses on the consistency of events reported in two news articles. The system consists of a pre-trained model(e.g., INFOXLM and XLM-RoBERTa) to extract multilingual news features, following fully-connected networks to measure the similarity. In addition, data augmentation and Ten Fold Voting are used to enhance the model. Our final submitted model is an ensemble of three base models, with a Pearson value of 0.784 on the test dataset.
This paper introduces our submission for the SemEval 2022 Task 8: Multilingual News Article Similarity. The task of the competition consisted of the development of a model, capable of determining the similarity between pairs of multilingual news articles. To address this challenge, we evaluated the Word Mover’s Distance in conjunction with word embeddings from ConceptNet Numberbatch and term frequencies of WorldLex, as well the Sentence Mover’s Distance based on sentence embeddings generated by pretrained transformer models of Sentence-BERT. To facilitate the comparison of multilingual articles with Sentence-BERT models, we deployed a Neural Machine Translation system. All our models achieve stable results in multilingual similarity estimation without learning parameters.
This paper describes the participation of the team “dina” in the Multilingual News Similarity task at SemEval 2022. To build our system for the task, we experimented with several multilingual language models which were originally pre-trained for semantic similarity but were not further fine-tuned. We use these models in combination with state-of-the-art packages for machine translation and named entity recognition with the expectation of providing valuable input to the model. Our work assesses the applicability of such “pure” models to solve the multilingual semantic similarity task in the case of news articles. Our best model achieved a score of 0.511, but shows that there is room for improvement.
Previous studies focus on measuring the degree of similarity of textsby using traditional machine learning methods, such as Support Vector Regression (SVR). Based on Transformers, this paper describes our contribution to SemEval-2022 Task 8 Multilingual News Article Similarity. The similarity of multilingual news articles requires a regression prediction on the similarity of multilingual articles, rather than a classification for judging text similarity. This paper mainly describes the architecture of the model and how to adjust the parameters in the experiment and strengthen the generalization ability. In this paper, we implement and construct different models through transformer-based models. We applied different transformer-based models, as well as ensemble them together by using ensemble learning. To avoid the overfit, we focus on the adjustment of parameters and the increase of generalization ability in our experiments. In the last submitted contest, we achieve a score of 0.715 and rank the 21st place.
This paper describes our system in SemEval-2022 Task 8, where participants were required to predict the similarity of two multilingual news articles. In the task of pairwise sentence and document scoring, there are two main approaches: Cross-Encoder, which inputs pairs of texts into a single encoder, and Bi-Encoder, which encodes each input independently. The former method often achieves higher performance, but the latter gave us a better result in SemEval-2022 Task 8. This paper presents our exploration of BERT-based Bi-Encoder approach for this task, and there are several findings such as pretrained models, pooling methods, translation, data separation, and the number of tokens. The weighted average ensemble of the four models achieved the competitive result and ranked in the top 12.
This paper describes the system submitted by our team (YNU-HPCC) to SemEval-2022 Task 8: Multilingual news article similarity. This task requires participants to develop a system which could evaluate the similarity between multilingual news article pairs. We propose an approach that relies on Transformers to compute the similarity between pairs of news. We tried different models namely BERT, ALBERT, ELECTRA, RoBERTa, M-BERT and Compared their results. At last, we chose M-BERT as our System, which has achieved the best Pearson Correlation Coefficient score of 0.738.
This paper presents our system for document-level semantic textual similarity (STS) evaluation at SemEval-2022 Task 8: “Multilingual News Article Similarity”. The semantic information used is obtained by using different semantic models ranging from the extraction of key terms and named entities to the document classification and obtaining similarity from automatic summarization of documents. All these semantic information’s are then used as features to feed a supervised system in order to evaluate the degree of similarity of a pair of documents. We obtained a Pearson correlation score of 0.706 compared to the best score of 0.818 from teams that participated in this task.
In this paper, we describe the approach we designed to solve SemEval-2022 Task 8: Multilingual News Article Similarity. We collect and use exclusively textual features (title, description and body) of articles. Our best model is a stacking of 14 Transformer-based Language models fine-tuned on single or multiple fields, using data in the original language or translated to English. It placed fourth on the original leaderboard, sixth on the complete official one and fourth on the English-subset official one. We observe the data collection as our principal source of error due to a relevant fraction of missing or wrong fields.
We present a system that creates pair-wise cosine and arccosine sentence similarity matrices using multilingual sentence embeddings obtained from pre-trained SBERT and Universal Sentence Encoder (USE) models respectively. For each news article sentence, it searches the most similar sentence from the other article and computes an average score. Further, a convolutional neural network calculates a total similarity score for the article pairs on these matrices. Finally, a random forest regressor merges the previous results to a final score that can optionally be extended with a publishing date score.
In this task, we identify a challenge that is reflective of linguistic and cognitive competencies that humans have when speaking and reasoning. Particularly, given the intuition that textual and visual information mutually inform each other for semantic reasoning, we formulate a Competence-based Question Answering challenge, designed to involve rich semantic annotation and aligned text-video objects. The task is to answer questions from a collection of cooking recipes and videos, where each question belongs to a “question family” reflecting a specific reasoning competence. The data and task result is publicly available.
This paper presents the second place system for the R2VQ: competence-based multimodal question answering shared task. The purpose of this task is to involve semantic&cooking roles and text-images objects when querying how well a system understands the procedure of a recipe. This task is approached with text-to-text generative model based on transformer architecture. As a result, the model can well generalise to soft constrained and other competence-based question answering problem. We propose label enclosed input method which help the model achieve significant improvement from 65.34 (baseline) to 91.3. In addition to describing the submitted system, the impact of model architecture and label selection are investigated along with remarks regarding error analysis. Finally, future works are presented.
In this work we present an overview of our winning system for the R2VQ - Competence-based Multimodal Question Answering task, with the final exact match score of 92.53%.The task is structured as question-answer pairs, querying how well a system is capable of competence-based comprehension of recipes.We propose a hybrid of a rule-based system, Question Answering Transformer, and a neural classifier for N/A answers recognition.The rule-based system focuses on intent identification, data extraction and response generation.
This paper describes our system used in the SemEval-2022 Task 09: R2VQ - Competence-based Multimodal Question Answering. We propose a knowledge-enhanced model for predicting answer in QA task, this model use BERT as the backbone. We adopted two knowledge-enhanced methods in this model: the knowledge auxiliary text method and the knowledge embedding method. We also design an answer extraction task pipeline, which contains an extraction-based model, an automatic keyword labeling module, and an answer generation module. Our system ranked 3rd in task 9 and achieved an exact match score of 78.21 and a word-level F1 score of 82.62.
In this paper, we introduce the first SemEval shared task on Structured Sentiment Analysis, for which participants are required to predict all sentiment graphs in a text, where a single sentiment graph is composed of a sentiment holder, target, expression and polarity. This new shared task includes two subtracks (monolingual and cross-lingual) with seven datasets available in five languages, namely Norwegian, Catalan, Basque, Spanish and English. Participants submitted their predictions on a held-out test set and were evaluated on Sentiment Graph F1 . Overall, the task received over 200 submissions from 32 participating teams. We present the results of the 15 teams that provided system descriptions and our own expanded analysis of the test predictions.
We describe the work carried out by AMEX AI Labs on the structured sentiment analysis task at SemEval-2022. This task focuses on extracting fine grained information w.r.t. to source, target and polar expressions in a given text. We propose a BERT based encoder, which utilizes a novel concatenation mechanism for combining syntactic and pretrained embeddings with BERT embeddings. Our system achieved an average rank of 14/32 systems, based on the average scores across seven datasets for five languages provided for the monolingual task. The proposed BERT based approaches outperformed BiLSTM based approaches used for structured sentiment extraction problem. We provide an in-depth analysis based on our post submission analysis.
ISCAS participated in both sub-tasks in SemEval-2022 Task 10: Structured Sentiment competition. We design an extraction-validation pipeline architecture to tackle both monolingual and cross-lingual sub-tasks. Experimental results show the multilingual effectiveness and cross-lingual robustness of our system. Our system is openly released on: https://github.com/luxinyu1/SemEval2022-Task10/.
Structured Sentiment Analysis is the task of extracting sentiment tuples in a graph structure commonly from review texts. We adapt the Aspect-Based Sentiment Analysis pointer network BARTABSA to model this tuple extraction as a sequence prediction task and extend their output grammar to account for the increased complexity of Structured Sentiment Analysis. To predict structured sentiment tuples in languages other than English we swap BART for a multilingual mT5 and introduce a novel Output Length Regularization to mitigate overfitting to common target sequence lengths, thereby improving the performance of the model by up to 70%. We evaluate our approach on seven datasets in five languages including a zero shot crosslingual setting.
Task 10 in SemEval 2022 is a composite task which entails analysis of opinion tuples, and recognition and demarcation of their nature. In this paper, we will elaborate on how such a methodology is implemented, how it is undertaken for a Structured Sentiment Analysis, and the results obtained thereof. To achieve this objective, we have adopted a bi-layered BiLSTM approach. In our research, a variation on the norm has been effected towards enhancement of accuracy, by basing the categorization meted out to an individual member as a by-product of its adjacent members, using specialized algorithms to ensure the veracity of the output, which has been modelled to be the holistically most accurate label for the entire sequence.Such a strategy is superior in terms of its parsing accuracy and requires less time. This manner of action has yielded an SF1 of 0.33 in the highest-performing configuration.
Sentiment analysis is a fundamental task, and structure sentiment analysis (SSA) is an important component of sentiment analysis. However, traditional SSA is suffering from some important issues: (1) lack of interactive knowledge of different languages; (2) small amount of annotation data or even no annotation data. To address the above problems, we incorporate data augment and auxiliary tasks within a cross-lingual pretrained language model into SSA. Specifically, we employ XLM-Roberta to enhance mutually interactive information when parallel data is available in the pretraining stage. Furthermore, we leverage two data augment strategies and auxiliary tasks to improve the performance on few-label data and zero-shot cross-lingual settings. Experiments demonstrate the effectiveness of our models. Our models rank first on the cross-lingual sub-task and rank second on the monolingual sub-task of SemEval-2022 task 10.
This paper presents our submission to task 10, Structured Sentiment Analysis of the SemEval 2022 competition. The task aims to extract all elements of the fine-grained sentiment in a text. We cast structured sentiment analysis to the prediction of the sentiment graphs following (Barnes et al., 2021), where nodes are spans of sentiment holders, targets and expressions, and directed edges denote the relation types between them. Our approach closely follows that of semantic dependency parsing (Dozat and Manning, 2018). The difference is that we use pre-trained language models (e.g., BERT and RoBERTa) as text encoder to solve the problem of limited annotated data. Additionally, we make improvements on the computation of cross attention and present the suffix masking technique to make further performance improvement. Substantially, our model achieved the Top-1 average Sentiment Graph F1 score on seven datasets in five different languages in the monolingual subtask.
This paper describes our participation in SemEval-2022 Task 10, a structured sentiment analysis. In this task, we have to parse opinions considering both structure- and context-dependent subjective aspects, which is different from typical dependency parsing. Some of the major parser types have recently been used for semantic and syntactic parsing, while it is still unknown which type can capture structured sentiments well due to their subjective aspects. To this end, we compared two different types of state-of-the-art parser, namely graph-based and seq2seq-based. Our in-depth analyses suggest that, even though graph-based parser generally outperforms the seq2seq-based one, with strong pre-trained language models both parsers can essentially output acceptable and reasonable predictions. The analyses highlight that the difficulty derived from subjective aspects in structured sentiment analysis remains an essential challenge.
This paper describes the system submitted by our team (UFRGSent) to SemEval-2022 Task 10: Structured Sentiment Analysis. We propose a multilingual approach that relies on a Question Answering model to find tuples consisting of aspect, opinion, and holder. The approach starts from general questions and uses the extracted tuple elements to find the remaining components. Finally, we employ an aspect sentiment classification model to classify the polarity of the entire tuple. Despite our method being in a mid-rank position on SemEval competition, we show that the question-answering approach can achieve good coverage retrieving sentiment tuples, allowing room for improvements in the technique.
This paper presents our solution for SemEval-2022 Task 10: Structured Sentiment Analysis. The solution consisted of two modules: the first for sequence tagging and the second for relation classification. In both modules we used transformer-based language models. In addition to utilizing language models specific to each of the five competition languages, we also adopted multilingual models. This approach allowed us to apply the solution to both monolingual and cross-lingual sub-tasks, where we obtained average Sentiment Graph F1 of 54.5% and 53.1%, respectively. The source code of the prepared solution is available at https://github.com/rafalposwiata/structured-sentiment-analysis.
Structured Sentiment Analysis (SSA) deals with extracting opinion tuples in a text, where each tuple (h, e, t, p) consists of h, the holder, who expresses a sentiment polarity p towards a target t through a sentiment expression e. While prior works explore graph-based or sequence labeling-based approaches for the task, we in this paper present a novel unified generative method to solve SSA, a SemEval2022 shared task. We leverage a BART-based encoder-decoder architecture and suitably modify it to generate, given a sentence, a sequence of opinion tuples. Each generated tuple consists of seven integers respectively representing the indices corresponding to the start and end positions of the holder, target, and expression spans, followed by the sentiment polarity class associated between the target and the sentiment expression. We perform rigorous experiments for both Monolingual and Cross-lingual subtasks, and achieve competitive Sentiment F1 scores on the leaderboard in both settings.
Sentiment analysis is a useful problem which could serve a variety of fields from business intelligence to social studies and even health studies. Using SemEval 2022 Task 10 formulation of this problem and taking sequence labeling as our approach, we propose a model which learns the task by finetuning a pretrained transformer, introducing as few parameters (~150k) as possible and making use of precomputed attention values in the transformer. Our model improves shared task baselines on all task datasets.
This paper addressed the problem of structured sentiment analysis using a bi-affine semantic dependency parser, large pre-trained language models, and publicly available translation models. For the monolingual setup, we considered: (i) training on a single treebank, and (ii) relaxing the setup by training on treebanks coming from different languages that can be adequately processed by cross-lingual language models. For the zero-shot setup and a given target treebank, we relied on: (i) a word-level translation of available treebanks in other languages to get noisy, unlikely-grammatical, but annotated data (we release as much of it as licenses allow), and (ii) merging those translated treebanks to obtain training data. In the post-evaluation phase, we also trained cross-lingual models that simply merged all the English treebanks and did not use word-level translations, and yet obtained better results. According to the official results, we ranked 8th and 9th in the monolingual and cross-lingual setups.
Sentiment analysis is a classical problem of natural language processing. SemEval 2022 sets a problem on the structured sentiment analysis in task 10, which is also a study-worthy topic in research area. In this paper, we propose a method which can predict structured sentiment information on multiple languages with limited data. The ERNIE-M pretrained language model is employed as a lingual feature extractor which works well on multiple language processing, followed by a graph parser as a opinion extractor. The method can predict structured sentiment information with high interpretability. We apply data augmentation as the given datasets are so small. Furthermore, we use K-fold cross-validation and DeBERTaV3 pretrained model as extra English embedding generator to train multiple models as our ensemble strategies. Experimental results show that the proposed model has considerable performance on both monolingual and cross-lingual tasks.
This paper describes our system that participated in the SemEval-2022 Task 10: Structured Sentiment Analysis, which aims to extract opinion tuples from texts.A full opinion tuple generally contains an opinion holder, an opinion target, the sentiment expression, and the corresponding polarity.The complex structure of the opinion tuple makes the task challenging.To address this task, we formalize it as a span-relation extraction problem and propose a two-stage extraction framework accordingly.In the first stage, we employ the span module to enumerate spans and then recognize the type of every span.In the second stage, we employ the relation module to determine the relation between spans.Our system achieves competitive results and ranks among the top-10 systems in almost subtasks.
We present the findings of SemEval-2022 Task 11 on Multilingual Complex Named Entity Recognition MULTICONER. Divided into 13 tracks, the task focused on methods to identify complex named entities (like names of movies, products and groups) in 11 languages in both monolingual and multi-lingual scenarios. Eleven tracks required building monolingual NER models for individual languages, one track focused on multilingual models able to work on all languages, and the last track featured code-mixed texts within any of these languages. The task is based on the MULTICONER dataset comprising of 2.3 millions instances in Bangla, Chinese, Dutch, English, Farsi, German, Hindi, Korean, Russian, Spanish, and Turkish. Results showed that methods fusing external knowledge into transformer models achieved the best results. However, identifying entities like creative works is still challenging even with external knowledge. MULTICONER was one of the most popular tasks in SemEval-2022 and it attracted 377 participants during the practice phase. 236 participants signed up for the final test phase and 55 teams submitted their systems.
Processing complex and ambiguous named entities is a challenging research problem, but it has not received sufficient attention from the natural language processing community. In this short paper, we present our participation in the English track of SemEval-2022 Task 11: Multilingual Complex Named Entity Recognition. Inspired by the recent advances in pretrained Transformer language models, we propose a simple yet effective Transformer-based baseline for the task. Despite its simplicity, our proposed approach shows competitive results in the leaderboard as we ranked 12 over 30 teams. Our system achieved a macro F1 score of 72.50% on the held-out test set. We have also explored a data augmentation approach using entity linking. While the approach does not improve the final performance, we also discuss it in this paper.
From pretrained contextual embedding to document-level embedding, the selection and construction of embedding have drawn more and more attention in the NER domain in recent research. This paper aims to discuss the performance of ensemble embeddings on complex NER tasks. Enlightened by Wang’s methodology, we try to replicate the dominating power of ensemble models with reinforcement learning optimizor on plain NER tasks to complex ones. Based on the composition of semeval dataset, the performance of the applied model is tested on lower-context, QA, and search query scenarios together with its zero-shot learning ability. Results show that with abundant training data, the model can achieve similar performance on lower-context cases compared to plain NER cases, but can barely transfer the performance to other scenarios in the test phase.
This study introduces the system submitted to the SemEval 2022 Task 11: MultiCoNER (Multilingual Complex Named Entity Recognition) by the UC3M-PUCPR team. We proposed an ensemble of transformer-based models for entity recognition in cross-domain texts. Our deep learning method benefits from the transformer architecture, which adopts the attention mechanism to handle the long-range dependencies of the input text. Also, the ensemble approach for named entity recognition (NER) improved the results over baselines based on individual models on two of the three tracks we participated in. The ensemble model for the code-mixed task achieves an overall performance of 76.36% F1-score, a 2.85 percentage point increase upon our individually best model for this task, XLM-RoBERTa-large (73.51%), outperforming the baseline provided for the shared task by 18.26 points. Our preliminary results suggest that contextualized language models ensembles can, even if modestly, improve the results in extracting information from unstructured data.
The MultiCoNER shared task aims at detecting semantically ambiguous and complex named entities in short and low-context settings for multiple languages. The lack of contexts makes the recognition of ambiguous named entities challenging. To alleviate this issue, our team DAMO-NLP proposes a knowledge-based system, where we build a multilingual knowledge base based on Wikipedia to provide related context information to the named entity recognition (NER) model. Given an input sentence, our system effectively retrieves related contexts from the knowledge base. The original input sentences are then augmented with such context information, allowing significantly better contextualized token representations to be captured. Our system wins 10 out of 13 tracks in the MultiCoNER shared task.
We leverage pre-trained language models to solve the task of complex NER for two low-resource languages: Chinese and Spanish. We use the technique of Whole Word Masking (WWM) to boost the performance of masked language modeling objective on large and unsupervised corpora. We experiment with multiple neural network architectures, incorporating CRF, BiLSTMs, and Linear Classifiers on top of a fine-tuned BERT layer. All our models outperform the baseline by a significant margin and our best performing model obtains a competitive position on the evaluation leaderboard for the blind test set.
This paper presents the system description of team AaltoNLP for SemEval-2022 shared task 11: MultiCoNER. Transformer-based models have produced high scores on standard Named Entity Recognition (NER) tasks. However, accuracy on complex named entities is still low. Complex and ambiguous named entities have been identified as a major error source in NER tasks. The shared task is about multilingual complex named entity recognition. In this paper, we describe an ensemble approach, which increases accuracy across all tested languages. The system ensembles output from multiple same architecture task-adaptive pretrained transformers trained with different random seeds. We notice a large discrepancy between performance on development and test data. Model selection based on limited development data may not yield optimal results on large test data sets.
In this paper, we describe a system that we built to participate in the SemEval 2022 Task 11: MultiCoNER Multilingual Complex Named Entity Recognition, specifically the track Mono-lingual in English. To construct this system, we used Pre-trained Language Models (PLMs). Especially, the Pre-trained Model base on BERT is applied for the task of recognizing named entities by fine-tuning method. We performed the evaluation on two test datasets of the shared task: the Practice Phase and the Evaluation Phase of the competition.
This article describes the OPDAI submission to SemEval-2022 Task 11 on Chinese complex NER. First, we explore the performance of model-based approaches and their ensemble, finding that fine-tuning the pre-trained Chinese RoBERTa-wwm model with word semantic representation and contextual gazetteer representation performs best among single models. However, the model-based approach performs poorly on test data because of low-context and unseen-entity cases. Then, we extend our system into two stages: (1) generating entity candidates by using neural model, soft-templates and Wikipedia lexicon. (2) predicting the final entity results within a feature-based rank model.For the evaluation, our best submission achieves an F1 score of 0.7954 and attains the third-best score in the Chinese sub-track.
Massively multilingual language models (MMLMs) have become a widely-used representation method, and multiple large MMLMs were proposed in recent years. A trend is to train MMLMs on larger text corpora or with more layers. In this paper we set out to test recent popular MMLMs on detecting semantically ambiguous and complex named entities with an academic GPU budget. Our submission of a single model for 11 languages on the SemEval Task 11 MultiCoNER shows that a vanilla transformer-CRF with XLM-Rlarge outperforms the more recent RemBERT, ranking 9th from 26 submissions in the multilingual track. Compared to RemBERT, the XLM-R model has the additional advantage to fit on a slice of a multi-instance GPU. As contrary to expectations and recent findings, we found RemBERT to not be the best MMLM, we further set out to investigate this discrepancy with additional experiments on multilingual Wikipedia NER data. While we expected RemBERT to have an edge on that dataset as it is closer to its pre-training data, surprisingly, our results show that this is not the case, suggesting that text domain match does not explain the discrepancy.
In low-resource languages, the amount of training data is limited. Hence, the model has to perform well in unseen sentences and syntax on which the model has not trained. We propose a method that addresses the problem through an encoder and an ensemble of language models. A language-specific language model performed poorly when compared to a multilingual language model. So, the multilingual language model checkpoint is fine-tuned to a specific language. A novel approach of one hot encoder is introduced between the model outputs and the CRF to combine the results in an ensemble format. Our team, Infrrd.ai, competed in the MultiCoNER competition. The results are encouraging where the team is positioned within the top 10 positions. There is less than a 4% percent difference from the third position in most of the tracks that we participated in. The proposed method shows that the ensemble of models with a multilingual language model as the base with the help of an encoder performs better than a single language-specific model.
Building real-world complex Named Entity Recognition (NER) systems is a challenging task. This is due to the complexity and ambiguity of named entities that appear in various contexts such as short input sentences, emerging entities, and complex entities. Besides, real-world queries are mostly malformed, as they can be code-mixed or multilingual, among other scenarios. In this paper, we introduce our submitted system to the Multilingual Complex Named Entity Recognition (MultiCoNER) shared task. We approach the complex NER for multilingual and code-mixed queries, by relying on the contextualized representation provided by the multilingual Transformer XLM-RoBERTa. In addition to the CRF-based token classification layer, we incorporate a span classification loss to recognize named entities spans. Furthermore, we use a self-training mechanism to generate weakly-annotated data from a large unlabeled dataset. Our proposed system is ranked 6th and 8th in the multilingual and code-mixed MultiCoNER’s tracks respectively.
This paper describes our approach to develop a complex named entity recognition system in SemEval 2022 Task 11: MultiCoNER Multilingual Complex Named Entity Recognition,Track 9 - Chinese. In this task, we need to identify the entity boundaries and categorylabels for the six identified categories of CW,LOC, PER, GRP, CORP, and PORD.The task focuses on detecting semantically ambiguous and complex entities in short and low-context settings. We constructed a hybrid system based on Roberta-large model with three training mechanisms and a series of data gugmentation.Three training mechanisms include adversarial training, Child-Tuning training, and continued pre-training. The core idea of the hybrid system is to improve the performance of the model in complex environments by introducing more domain knowledge through data augmentation and continuing pre-training domain adaptation of the model. Our proposed method in this paper achieves a macro-F1 of 0.797 on the final test set, ranking second.
Many areas, such as the biological and healthcare domain, artistic works, and organization names, have nested, overlapping, discontinuous entity mentions that may even be syntactically or semantically ambiguous in practice. Traditional sequence tagging algorithms are unable to recognize these complex mentions because they may violate the assumptions upon which sequence tagging schemes are founded. In this paper, we describe our contribution to SemEval 2022 Task 11 on identifying such complex Named Entities. We have leveraged the ensemble of multiple ELECTRA-based models that were exclusively pretrained on the Bangla language with the performance of ELECTRA-based models pretrained on English to achieve competitive performance on the Track-11. Besides providing a system description, we will also present the outcomes of our experiments on architectural decisions, dataset augmentations, and post-competition findings.
In this work, we introduce our system to the SemEval 2022 Task 11: Multilingual Complex Named Entity Recognition (MultiCoNER) competition. Our team (KDDIE) attempted the sub-task of Named Entity Recognition (NER) for the language of English in the challenge and reported our results. For this task, we use transfer learning method: fine-tuning the pre-trained language models (PLMs) on the competition dataset. Our two approaches are the BERT-based PLMs and PLMs with additional layer such as Condition Random Field. We report our finding and results in this report.
We present Transformer based pretrained models, which are fine-tuned for Named Entity Recognition (NER) task. Our team participated in SemEval-2022 Task 11 MultiCoNER: Multilingual Complex Named Entity Recognition task for Hindi and Bangla. Result comparison of six models (mBERT, IndicBERT, MuRIL (Base), MuRIL (Large), XLM-RoBERTa (Base) and XLM-RoBERTa (Large) ) has been performed. It is found that among these models MuRIL (Large) model performs better for both the Hindi and Bangla languages. Its F1-Scores for Hindi and Bangla are 0.69 and 0.59 respectively.
In this paper, we describe our proposed method for the SemEval 2022 Task 11: Multilingual Complex Named Entity Recognition (MultiCoNER). The goal of this task is to locate and classify named entities in unstructured short complex texts in 11 different languages.After training a variety of contextual language models on the NER dataset, we used an ensemble strategy based on a majority vote to finalize our model. We evaluated our proposed approach on the multilingual NER dataset at SemEval-2022. The ensemble model provided consistent improvements against the individual models on the multilingual track, achieving a macro F1 performance of 65.2%. However, our results were significantly outperformed by the top ranking systems, achieving thus a baseline performance.
Recognizing complex and ambiguous named entities (NEs) is one of the formidable tasks in the NLP domain. However, the diversity of linguistic constituents, syntactic structure, semantic ambiguity as well as differences from traditional NEs make it challenging to identify the complex NEs. To address these challenges, SemEval-2022 Task 11 introduced a shared task MultiCoNER focusing on complex named entity recognition in multilingual settings. This paper presents our participation in this task where we propose two different approaches including a BiLSTM-CRF model with stacked-embedding strategy and a transformer-based approach. Our proposed method achieved competitive performance among the participants’ methods in a few languages.
Identifying named entities is, in general, a practical and challenging task in the field of Natural Language Processing. Named Entity Recognition on the code-mixed text is further challenging due to the linguistic complexity resulting from the nature of the mixing. This paper addresses the submission of team CMNEROne to the SEMEVAL 2022 shared task 11 MultiCoNER. The Code-mixed NER task aimed to identify named entities on the code-mixed dataset. Our work consists of Named Entity Recognition (NER) on the code-mixed dataset by leveraging the multilingual data. We achieved a weighted average F1 score of 0.7044, i.e., 6% greater than the NER baseline.
This paper presents RACAI’s system used for the shared task of “Multilingual Complex Named Entity Recognition (MultiCoNER)”, organized as part of the “The 16th International Workshop on Semantic Evaluation (SemEval 2022)”. The system employs a novel layer inspired by the biological mechanism of lateral inhibition. This allowed the system to achieve good results without any additional resources apart from the provided training data. In addition to the system’s architecture, results are provided as well as observations regarding the provided dataset.
This paper presents the two submissions of NamedEntityRangers Team to the MultiCoNER Shared Task, hosted at SemEval-2022. We evaluate two state-of-the-art approaches, of which both utilize pre-trained multi-lingual language models differently. The first approach follows the token classification schema, in which each token is assigned with a tag. The second approach follows a recent template-free paradigm, in which an encoder-decoder model translates the input sequence of words to a special output, encoding named entities with predefined labels. We utilize RemBERT and mT5 as backbone models for these two approaches, respectively. Our results show that the oldie but goodie token classification outperforms the template-free method by a wide margin. Our code is available at: https://github.com/Abiks/MultiCoNER.
Named Entity Recognition (NER), an essential subtask in NLP that identifies text belonging to predefined semantics such as a person, location, organization, drug, time, clinical procedure, biological protein, etc. NER plays a vital role in various fields such as informationextraction, question answering, and machine translation. This paper describes our participating system run to the Named entity recognitionand classification shared task SemEval-2022. The task is motivated towards detecting semantically ambiguous and complex entities in shortand low-context settings. Our team focused on improving entity recognition by improving the word embeddings. We concatenated the word representations from State-of-the-art language models and passed them to find the best representation through a reinforcement trainer. Our results highlight the improvements achieved by various embedding concatenations.
This paper presents our system used to participate in task 11 (MultiCONER) of the SemEval 2022 competition. Our system ranked fourth place in track 12 (Multilingual) and fifth place in track 13 (Code-Mixed). The goal of track 12 is to detect complex named entities in a multilingual setting, while track 13 is dedicated to detecting complex named entities in a code-mixed setting. Both systems were developed using transformer-based language models. We used an ensemble of XLM-RoBERTa-large and Microsoft/infoxlm-large with a Conditional Random Field (CRF) layer. In addition, we describe the algorithms employed to train our models and our hyper-parameter selection. We furthermore study the impact of different methods to aggregate the outputs of the individual models that compose our ensemble. Finally, we present an extensive analysis of the results and errors.
Large scale pre-training models have been widely used in named entity recognition (NER) tasks. However, model ensemble through parameter averaging or voting can not give full play to the differentiation advantages of different models, especially in the open domain. This paper describes our NER system in the SemEval 2022 task11: MultiCoNER. We proposed an effective system to adaptively ensemble pre-trained language models by a Transformer layer. By assigning different weights to each model for different inputs, we adopted the Transformer layer to integrate the advantages of diverse models effectively. Experimental results show that our method achieves superior performances in Farsi and Dutch.
This study describes the model design of the NCUEE-NLP system for the Chinese track of the SemEval-2022 MultiCoNER task. We use the BERT embedding for character representation and train the BiLSTM-CRF model to recognize complex named entities. A total of 21 teams participated in this track, with each team allowed a maximum of six submissions. Our best submission, with a macro-averaging F1-score of 0.7418, ranked the seventh position out of 21 teams.
This paper presents a solution for the SemEval-2022 Task 11 Multilingual Complex Named Entity Recognition. What is challenging in this task is detecting semantically ambiguous and complex entities in short and low-context settings. Our team (CMB AI Lab) propose a two-stage method to recognize the named entities: first, a model based on biaffine layer is built to predict span boundaries, and then a span classification model based on pooling layer is built to predict semantic tags of the spans. The basic pre-trained models we choose are XLM-RoBERTa and mT5. The evaluation result of our approach achieves an F1 score of 84.62 on sub-task 13, which ranks the third on the learder board.
This paper presents the approaches and systems of the UA-KO team for the Korean portion of SemEval-2022 Task 11 on Multilingual Complex Named Entity Recognition.We fine-tuned Korean and multilingual BERT and RoBERTA models, conducted experiments on data augmentation, ensembles, and task-adaptive pretraining. Our final system ranked 8th out of 17 teams with an F1 score of 0.6749 F1.
This paper describes the system developed by the USTC-NELSLIP team for SemEval-2022 Task 11 Multilingual Complex Named Entity Recognition (MultiCoNER). We propose a gazetteer-adapted integration network (GAIN) to improve the performance of language models for recognizing complex named entities. The method first adapts the representations of gazetteer networks to those of language models by minimizing the KL divergence between them. After adaptation, these two networks are then integrated for backend supervised named entity recognition (NER) training. The proposed method is applied to several state-of-the-art Transformer-based NER models with a gazetteer built from Wikidata, and shows great generalization ability across them. The final predictions are derived from an ensemble of these trained models. Experimental results and detailed analysis verify the effectiveness of the proposed method. The official results show that our system ranked 1st on three tracks (Chinese, Code-mixed and Bangla) and 2nd on the other ten tracks in this task.
We investigate the task of complex NER for the English language. The task is non-trivial due to the semantic ambiguity of the textual structure and the rarity of occurrence of such entities in the prevalent literature. Using pre-trained language models such as BERT, we obtain a competitive performance on this task. We qualitatively analyze the performance of multiple architectures for this task. All our models are able to outperform the baseline by a significant margin. Our best performing model beats the baseline F1-score by over 9%.
This paper summarizes the participation of the L3i laboratory of the University of La Rochelle in the SemEval-2022 Task 11, Multilingual Complex Named Entity Recognition (MultiCoNER). The task focuses on detecting semantically ambiguous and complex entities in short and low-context monolingual and multilingual settings. We argue that using a language-specific and a multilingual language model could improve the performance of multilingual and mixed NER. Also, we consider that using additional contexts from the training set could improve the performance of a NER on short texts. Thus, we propose a straightforward technique for generating additional contexts with and without the presence of entities. Our findings suggest that, in our internal experimental setup, this approach is promising. However, we ranked above average for the high-resource languages and lower than average for low-resource and multilingual models.
The multilingual complex named entity recognition task of SemEval2020 required participants to detect semantically ambiguous and complex entities in 11 languages. In order to participate in this competition, a deep learning model is being used with the T5 text-to-text language model and its multilingual version, MT5, along with the transformer’s encoder module. The subtoken check has also been introduced, resulting in a 4% increase in the model F1-score in English. We also examined the use of the BPEmb model for converting input tokens to representation vectors in this research. A performance evaluation of the proposed entity detection model is presented at the end of this paper. Six different scenarios were defined, and the proposed model was evaluated in each scenario within the English development set. Our model is also evaluated in other languages.
This paper describes the system proposed by Sabancı University Natural Language Processing Group in the SemEval-2022 MultiCoNER task. We developed an unsupervised entity linking pipeline that detects potential entity mentions with the help of Wikipedia and also uses the corresponding Wikipedia context to help the classifier in finding the named entity type of that mention. The proposed pipeline significantly improved the performance, especially for complex entities in low-context settings.