This is an internal preview of the ACL Anthology that may be incomplete and contain mistakes. Do not treat this content as an official publication.
Marta R.Costa-jussà
Also published as:
Marta R. Costa-Jussa,
Marta R. Costa-Jussà,
Marta R. Costa-jussa,
Marta R. Costa-jussà,
Marta R. Costa-jussá,
Marta Ruiz Costa-jussà,
Marta Ruiz Costa-jussà
Speech-to-Speech and Speech-to-Text translation are currently dynamic areas of research. In our commitment to advance these fields, we present SpeechAlign, a framework designed to evaluate the underexplored field of source-target alignment in speech models. The SpeechAlign framework has two core components. First, to tackle the absence of suitable evaluation datasets, we introduce the Speech Gold Alignment dataset, built upon a English-German text translation gold alignment dataset. Secondly, we introduce two novel metrics, Speech Alignment Error Rate (SAER) and Time-weighted Speech Alignment Error Rate (TW-SAER), which enable the evaluation of alignment quality within speech models. While the former gives equal importance to each word, the latter assigns weights based on the length of the words in the speech signal. By publishing SpeechAlign we provide an accessible evaluation framework for model assessment, and we employ it to benchmark open-source Speech Translation models. In doing so, we contribute to the ongoing research progress within the fields of Speech-to-Speech and Speech-to-Text translation.
While the problem of hallucinations in neural machine translation has long been recognized, so far the progress on its alleviation is very little. Indeed, recently it turned out that without artificially encouraging models to hallucinate, previously existing methods fall short and even the standard sequence log-probability is more informative. It means that internal characteristics of the model can give much more information than we expect, and before using external models and measures, we first need to ask: how far can we go if we use nothing but the translation model itself ? We propose to use a method that evaluates the percentage of the source contribution to a generated translation. Intuitively, hallucinations are translations “detached” from the source, hence they can be identified by low source contribution. This method improves detection accuracy for the most severe hallucinations by a factor of 2 and is able to alleviate hallucinations at test time on par with the previous best approach that relies on external models. Next, if we move away from internal model characteristics and allow external tools, we show that using sentence similarity from cross-lingual embeddings further improves these results. We release the code of our experiments.
Language Generation Models produce words based on the previous context. Although existing methods offer input attributions as explanations for a model’s prediction, it is still unclear how prior words affect the model’s decision throughout the layers. In this work, we leverage recent advances in explainability of the Transformer and present a procedure to analyze models for language generation. Using contrastive examples, we compare the alignment of our explanations with evidence of the linguistic phenomena, and show that our method consistently aligns better than gradient-based and perturbation-based baselines. Then, we investigate the role of MLPs inside the Transformer and show that they learn features that help the model predict words that are grammatically acceptable. Lastly, we apply our method to Neural Machine Translation models, and demonstrate that they generate human-like source-target alignments for building predictions.
End-to-End speech-to-speech translation (S2ST) is generally evaluated with text-based metrics. This means that generated speech has to be automatically transcribed, making the evaluation dependent on the availability and quality of automatic speech recognition (ASR) systems. In this paper, we propose a text-free evaluation metric for end-to-end S2ST, named BLASER, to avoid the dependency on ASR systems. BLASER leverages a multilingual multimodal encoder to directly encode the speech segments for source input, translation output and reference into a shared embedding space and computes a score of the translation quality that can be used as a proxy to human evaluation. To evaluate our approach, we construct training and evaluation sets from more than 40k human annotations covering seven language directions. The best results of BLASER are achieved by training with supervision from human rating scores. We show that when evaluated at the sentence level, BLASER correlates significantly better with human judgment compared to ASR dependent metrics including ASR-SENTBLEU in all translation directions and ASR-COMET in five of them. Our analysis shows combining speech and text as inputs to BLASER does not increase the correlation with human scores, but best correlations are achieved when using speech, which motivates the goal of our research. Moreover, we show that using ASR for references is detrimental for text-based metrics.
This paper describes the submission of the UPC Machine Translation group to the IWSLT 2023 Offline Speech Translation task. Our Speech Translation systems utilize foundation models for speech (wav2vec 2.0) and text (mBART50). We incorporate a Siamese pretraining step of the speech and text encoders with CTC and Optimal Transport, to adapt the speech representations to the space of the text model, thus maximizing transfer learning from MT. After this pretraining, we fine-tune our system end-to-end on ST, with Cross Entropy and Knowledge Distillation. Apart from the available ST corpora, we create synthetic data with SegAugment to better adapt our models to the custom segmentations of the IWSLT test sets. Our best single model obtains 31.2 BLEU points on MuST-C tst-COMMON, 29.8 points on IWLST.tst2020 and 33.4 points on the newly released IWSLT.ACLdev2023.
Gender biases in language generation systems are challenging to mitigate. One possible source for these biases is gender representation disparities in the training and evaluation data. Despite recent progress in documenting this problem and many attempts at mitigating it, we still lack shared methodology and tooling to report gender representation in large datasets. Such quantitative reporting will enable further mitigation, e.g., via data augmentation. This paper describes the Gender-Gap Pipeline (for Gender-Aware Polyglot Pipeline), an automatic pipeline to characterize gender representation in large-scale datasets for 55 languages. The pipeline uses a multilingual lexicon of gendered person-nouns to quantify the gender representation in text. We showcase it to report gender representation in WMT training data and development data for the News task, confirming that current data is skewed towards masculine representation. Having unbalanced datasets may indirectly optimize our systems towards outperforming one gender over the others. We suggest introducing our gender quantification pipeline in current datasets and, ideally, modifying them toward a balanced representation.
We present the results of the WMT’22 SharedTask on Large-Scale Machine Translation Evaluation for African Languages. The shared taskincluded both a data and a systems track, alongwith additional innovations, such as a focus onAfrican languages and extensive human evaluation of submitted systems. We received 14system submissions from 8 teams, as well as6 data track contributions. We report a largeprogress in the quality of translation for Africanlanguages since the last iteration of this sharedtask: there is an increase of about 7.5 BLEUpoints across 72 language pairs, and the average BLEU scores went from 15.09 to 22.60.
The Transformer architecture aggregates input information through the self-attention mechanism, but there is no clear understanding of how this information is mixed across the entire model. Additionally, recent works have demonstrated that attention weights alone are not enough to describe the flow of information. In this paper, we consider the whole attention block –multi-head attention, residual connection, and layer normalization– and define a metric to measure token-to-token interactions within each layer. Then, we aggregate layer-wise interpretations to provide input attribution scores for model predictions. Experimentally, we show that our method, ALTI (Aggregation of Layer-wise Token-to-token Interactions), provides more faithful explanations and increased robustness than gradient-based methods.
In Neural Machine Translation (NMT), each token prediction is conditioned on the source sentence and the target prefix (what has been previously translated at a decoding step). However, previous work on interpretability in NMT has mainly focused solely on source sentence tokens’ attributions. Therefore, we lack a full understanding of the influences of every input token (source sentence and target prefix) in the model predictions. In this work, we propose an interpretability method that tracks input tokens’ attributions for both contexts. Our method, which can be extended to any encoder-decoder Transformer-based model, allows us to better comprehend the inner workings of current NMT models. We apply the proposed method to both bilingual and multilingual Transformers and present insights into their behaviour.
Transformer-based models have been achieving state-of-the-art results in several fields of Natural Language Processing. However, its direct application to speech tasks is not trivial. The nature of this sequences carries problems such as long sequence lengths and redundancy between adjacent tokens. Therefore, we believe that regular self-attention mechanism might not be well suited for it. Different approaches have been proposed to overcome these problems, such as the use of efficient attention mechanisms. However, the use of these methods usually comes with a cost, which is a performance reduction caused by information loss. In this study, we present the Multiformer, a Transformer-based model which allows the use of different attention mechanisms on each head. By doing this, the model is able to bias the self-attention towards the extraction of more diverse token interactions, and the information loss is reduced. Finally, we perform an analysis of the head contributions, and we observe that those architectures where all heads relevance is uniformly distributed obtain better results. Our results show that mixing attention patterns along the different heads and layers outperforms our baseline by up to 0.7 BLEU.
The scientific community is increasingly aware of the necessity to embrace pluralism and consistently represent major and minor social groups. Currently, there are no standard evaluation techniques for different types of biases. Accordingly, there is an urgent need to provide evaluation sets and protocols to measure existing biases in our automatic systems. Evaluating the biases should be an essential step towards mitigating them in the systems. This paper introduces WinoST, a new freely available challenge set for evaluating gender bias in speech translation. WinoST is the speech version of WinoMT, an MT challenge set, and both follow an evaluation protocol to measure gender accuracy. Using an S-Transformer end-to-end speech translation system, we report the gender bias evaluation on four language pairs, and we reveal the inaccuracies in translations generating gender-stereotyped translations.
Transformers have achieved state-of-the-art results across multiple NLP tasks. However, the self-attention mechanism complexity scales quadratically with the sequence length, creating an obstacle for tasks involving long sequences, like in the speech domain. In this paper, we discuss the usefulness of self-attention for Direct Speech Translation. First, we analyze the layer-wise token contributions in the self-attention of the encoder, unveiling local diagonal patterns. To prove that some attention weights are avoidable, we propose to substitute the standard self-attention with a local efficient one, setting the amount of context used based on the results of the analysis. With this approach, our model matches the baseline performance, and improves the efficiency by skipping the computation of those weights that standard attention discards.
This paper describes the submissions of the UPC Machine Translation group to the IWSLT 2022 Offline Speech Translation and Speech-to-Speech Translation tracks. The offline task involves translating English speech to German, Japanese and Chinese text. Our Speech Translation systems are trained end-to-end and are based on large pretrained speech and text models. We use an efficient fine-tuning technique that trains only specific layers of our system, and explore the use of adapter modules for the non-trainable layers. We further investigate the suitability of different speech encoders (wav2vec 2.0, HuBERT) for our models and the impact of knowledge distillation from the Machine Translation model that we use for the decoder (mBART). For segmenting the IWSLT test sets we fine-tune a pretrained audio segmentation model and achieve improvements of 5 BLEU compared to the given segmentation. Our best single model uses HuBERT and parallel adapters and achieves 29.42 BLEU at English-German MuST-C tst-COMMON and 26.77 at IWSLT 2020 test. By ensembling many models, we further increase translation quality to 30.83 BLEU and 27.78 accordingly. Furthermore, our submission for English-Japanese achieves 15.85 and English-Chinese obtains 25.63 BLEU on the MuST-C tst-COMMON sets. Finally, we extend our system to perform English-German Speech-to-Speech Translation with a pretrained Text-to-Speech model.
This paper describes the submission to the IWSLT 2021 offline speech translation task by the UPC Machine Translation group. The task consists of building a system capable of translating English audio recordings extracted from TED talks into German text. Submitted systems can be either cascade or end-to-end and use a custom or given segmentation. Our submission is an end-to-end speech translation system, which combines pre-trained models (Wav2Vec 2.0 and mBART) with coupling modules between the encoder and decoder, and uses an efficient fine-tuning technique, which trains only 20% of its total parameters. We show that adding an Adapter to the system and pre-training it, can increase the convergence speed and the final result, with which we achieve a BLEU score of 27.3 on the MuST-C test set. Our final model is an ensemble that obtains 28.22 BLEU score on the same set. Our submission also uses a custom segmentation algorithm that employs pre-trained Wav2Vec 2.0 for identifying periods of untranscribable text and can bring improvements of 2.5 to 3 BLEU score on the IWSLT 2019 test set, as compared to the result with the given segmentation.
State-of-the-art multilingual machine translation relies on a universal encoder-decoder, which requires retraining the entire system to add new languages. In this paper, we propose an alternative approach that is based on language-specific encoder-decoders, and can thus be more easily extended to new languages by learning their corresponding modules. So as to encourage a common interlingua representation, we simultaneously train the N initial languages. Our experiments show that the proposed approach outperforms the universal encoder-decoder by 3.28 BLEU points on average, while allowing to add new languages without the need to retrain the rest of the modules. All in all, our work closes the gap between shared and language-specific encoderdecoders, advancing toward modular multilingual machine translation systems that can be flexibly extended in lifelong learning settings.
This work proposes an extensive analysis of the Transformer architecture in the Neural Machine Translation (NMT) setting. Focusing on the encoder-decoder attention mechanism, we prove that attention weights systematically make alignment errors by relying mainly on uninformative tokens from the source sequence. However, we observe that NMT models assign attention to these tokens to regulate the contribution in the prediction of the two contexts, the source and the prefix of the target sequence. We provide evidence about the influence of wrong alignments on the model behavior, demonstrating that the encoder-decoder attention mechanism is well suited as an interpretability method for NMT. Finally, based on our analysis, we propose methods that largely reduce the word alignment error rate compared to standard induced alignments from attention weights.
This paper presents the results of the newstranslation task, the multilingual low-resourcetranslation for Indo-European languages, thetriangular translation task, and the automaticpost-editing task organised as part of the Con-ference on Machine Translation (WMT) 2021.In the news task, participants were asked tobuild machine translation systems for any of10 language pairs, to be evaluated on test setsconsisting mainly of news stories. The taskwas also opened up to additional test suites toprobe specific aspects of translation.
This paper describes the submission to the WMT 2021 news translation shared task by the UPC Machine Translation group. The goal of the task is to translate German to French (De-Fr) and French to German (Fr-De). Our submission focuses on fine-tuning a pre-trained model to take advantage of monolingual data. We fine-tune mBART50 using the filtered data, and additionally, we train a Transformer model on the same data from scratch. In the experiments, we show that fine-tuning mBART50 results in 31.69 BLEU for De-Fr and 23.63 BLEU for Fr-De, which increases 2.71 and 1.90 BLEU accordingly, as compared to the model we train from scratch. Our final submission is an ensemble of these two models, further increasing 0.3 BLEU for Fr-De.
This paper reports the optimization of using the out-of-domain data in the Biomedical translation task. We firstly optimized our parallel training dataset using the BabelNet in-domain terminology words. Afterward, to increase the training set, we studied the effects of the out-of-domain data on biomedical translation tasks, and we created a mixture of in-domain and out-of-domain training sets and added more in-domain data using forward translation in the English-Spanish task. Finally, with a simple bpe optimization method, we increased the number of in-domain sub-words in our mixed training set and trained the Transformer model on the generated data. Results show improvements using our proposed method.
Introducing factors, that is to say, word features such as linguistic information referring to the source tokens, is known to improve the results of neural machine translation systems in certain settings, typically in recurrent architectures. This study proposes enhancing the current state-of-the-art neural machine translation architecture, the Transformer, so that it allows to introduce external knowledge. In particular, our proposed modification, the Factored Transformer, uses linguistic factors that insert additional knowledge into the machine translation system. Apart from using different kinds of features, we study the effect of different architectural configurations. Specifically, we analyze the performance of combining words and features at the embedding level or at the encoder level, and we experiment with two different combination strategies. With the best-found configuration, we show improvements of 0.8 BLEU over the baseline Transformer in the IWSLT German-to-English task. Moreover, we experiment with the more challenging FLoRes English-to-Nepali benchmark, which includes both extremely low-resourced and very distant languages, and obtain an improvement of 1.2 BLEU
Machine Translation is highly impacted by social biases present in data sets, indicating that it reflects and amplifies stereotypes. In this work, we study mitigating gender bias by jointly learning the translation, the part-of-speech, and the gender of the target language with different morphological complexity. This approach has shown improvements up to 6.8 points in gender accuracy without significantly impacting the translation quality.
This study sheds light on the effects of COVID-19 in the particular field of Computational Linguistics and Natural Language Processing within Artificial Intelligence. We provide an inter-sectional study on gender, contribution, and experience that considers one school year (from August 2019 to August 2020) as a pandemic year. August is included twice for the purpose of an inter-annual comparison. While the trend in publications increased with the crisis, the results show that the ratio between female and male publications decreased. This only helps to reduce the importance of the female role in the scientific contributions of computational linguistics (it is now far below its peak of 0.24). The pandemic has a particularly negative effect on the production of female senior researchers in the first position of authors (maximum work), followed by the female junior researchers in the last position of authors (supervision or collaborative work).
Misrepresentation of certain communities in datasets is causing big disruptions in artificial intelligence applications. In this paper, we propose using an automatically extracted gender-balanced dataset parallel corpus from Wikipedia. This balanced set is used to perform fine-tuning techniques from a bigger model trained on unbalanced datasets to mitigate gender biases in neural machine translation.
The dominant language modeling paradigm handles text as a sequence of discrete tokens. While that approach can capture the latent structure of the text, it is inherently constrained to sequential dynamics for text generation. We propose a new paradigm for introducing a syntactic inductive bias into neural text generation, where the dependency parse tree is used to drive the Transformer model to generate sentences iteratively. Our experiments show that this paradigm is effective at text generation, with quality between LSTMs and Transformers, and comparable diversity, requiring less than half their decoding steps, and its generation process allows direct control over the syntactic constructions of the generated text, enabling the induction of stylistic variations.
We introduce the Computational Linguistics special issue on Multilingual and Interlingual Semantic Representations for Natural Language Processing. We situate the special issue’s five articles in the context of our fast-changing field, explaining our motivation for this project. We offer a brief summary of the work in the issue, which includes developments on lexical and sentential semantic representations, from symbolic and neural perspectives.
Abusive texts are reaching the interests of the scientific and social community. How to automatically detect them is onequestion that is gaining interest in the natural language processing community. The main contribution of this paper is toevaluate the quality of the recently developed ”Spanish Database for cyberbullying prevention” for the purpose of trainingclassifiers on detecting abusive short texts. We compare classical machine learning techniques to the use of a more ad-vanced model: the contextual word embeddings in the particular case of classification of abusive short-texts for the Spanishlanguage. As contextual word embeddings, we use Bidirectional Encoder Representation from Transformers (BERT), pro-posed at the end of 2018. We show that BERT mostly outperforms classical techniques. Far beyond the experimentalimpact of our research, this project aims at planting the seeds for an innovative technological tool with a high potentialsocial impact and aiming at being part of the initiatives in artificial intelligence for social good.
We introduce GeBioToolkit, a tool for extracting multilingual parallel corpora at sentence level, with document and gender information from Wikipedia biographies. Despite the gender inequalities present in Wikipedia, the toolkit has been designed to extract corpus balanced in gender. While our toolkit is customizable to any number of languages (and different domains), in this work we present a corpus of 2,000 sentences in English, Spanish and Catalan, which has been post-edited by native speakers to become a high-quality dataset for machine translation evaluation. While GeBioCorpus aims at being one of the first non-synthetic gender-balanced test datasets, GeBioToolkit aims at paving the path to standardize procedures to produce gender-balanced datasets.
Recently, multilingual question answering became a crucial research topic, and it is receiving increased interest in the NLP community. However, the unavailability of large-scale datasets makes it challenging to train multilingual QA systems with performance comparable to the English ones. In this work, we develop the Translate Align Retrieve (TAR) method to automatically translate the Stanford Question Answering Dataset (SQuAD) v1.1 to Spanish. We then used this dataset to train Spanish QA systems by fine-tuning a Multilingual-BERT model. Finally, we evaluated our QA models with the recently proposed MLQA and XQuAD benchmarks for cross-lingual Extractive QA. Experimental results show that our models outperform the previous Multilingual-BERT baselines achieving the new state-of-the-art values of 68.1 F1 on the Spanish MLQA corpus and 77.6 F1 on the Spanish XQuAD corpus. The resulting, synthetically generated SQuAD-es v1.1 corpora, with almost 100% of data contained in the original English version, to the best of our knowledge, is the first large-scale QA training resource for Spanish.
Gender bias negatively impacts many natural language processing applications, including machine translation (MT). The motivation behind this work is to study whether recent proposed MT techniques are significantly contributing to attenuate biases in document-level and gender-balanced data. For the study, we consider approaches of adding the previous sentence and the speaker information, implemented in a decoder-based neural MT system. We show improvements both in translation quality (+1 BLEU point) as well as in gender bias mitigation on WinoMT (+5% accuracy).
In Neural Machine Translation, using word-level tokens leads to degradation in translation quality. The dominant approaches use subword-level tokens, but this increases the length of the sequences and makes it difficult to profit from word-level information such as POS tags or semantic dependencies. We propose a modification to the Transformer model to combine subword-level representations into word-level ones in the first layers of the encoder, reducing the effective length of the sequences in the following layers and providing a natural point to incorporate extra word-level information. Our experiments show that this approach maintains the translation quality with respect to the normal Transformer model when no extra word-level information is injected and that it is superior to the currently dominant method for incorporating word-level source language information to models based on subword-level vocabularies.
In this work, we present an effective method for semantic specialization of word vector representations. To this end, we use traditional word embeddings and apply specialization methods to better capture semantic relations between words. In our approach, we leverage external knowledge from rich lexical resources such as BabelNet. We also show that our proposed post-specialization method based on an adversarial neural network with the Wasserstein distance allows to gain improvements over state-of-the-art methods on two tasks: word similarity and dialog state tracking.
Continual learning (CL) aims to enable information systems to learn from a continuous data stream across time. However, it is difficult for existing deep learning architectures to learn a new task without largely forgetting previously acquired knowledge. Furthermore, CL is particularly challenging for language learning, as natural language is ambiguous: it is discrete, compositional, and its meaning is context-dependent. In this work, we look at the problem of CL through the lens of various NLP tasks. Our survey discusses major challenges in CL and current methods applied in neural network models. We also provide a critical review of the existing CL evaluation methods and datasets in NLP. Finally, we present our outlook on future research directions.
This paper presents the results of the news translation task and the similar language translation task, both organised alongside the Conference on Machine Translation (WMT) 2020. In the news task, participants were asked to build machine translation systems for any of 11 language pairs, to be evaluated on test sets consisting mainly of news stories. The task was also opened up to additional test suites to probe specific aspects of translation. In the similar language translation task, participants built machine translation systems for translating between closely related pairs of languages.
A lifelong learning system can adapt to new data without forgetting previously acquired knowledge. In this paper, we introduce the first benchmark for lifelong learning machine translation. For this purpose, we provide training, lifelong and test data sets for two language pairs: English-German and English-French. Additionally, we report the results of our baseline systems, which we make available to the public. The goal of this shared task is to encourage research on the emerging topic of lifelong learning machine translation.
In this article, we describe the TALP-UPC participation in the WMT20 news translation shared task for Tamil-English. Given the low amount of parallel training data, we resort to adapt the task to a multilingual system to benefit from the positive transfer from high resource languages. We use iterative backtranslation to fine-tune the system and benefit from the monolingual data available. In order to measure the effectivity of such methods, we compare our results to a bilingual baseline system.
This paper describes the participation of the NLP research team of the IPN Computer Research center in the WMT 2020 Similar Language Translation Task. We have submitted systems for the Spanish-Portuguese language pair (in both directions). The three submitted systems are based on the Transformer architecture and used fine tuning for domain Adaptation.
In this paper, we describe the TALP-UPC participation in the WMT Similar Language Translation task between Catalan, Spanish, and Portuguese, all of them, Romance languages. We made use of different techniques to improve the translation between these languages. The multilingual shared encoder/decoder has been used for all of them. Additionally, we applied back-translation to take advantage of the monolingual data. Finally, we have applied fine-tuning to improve the in-domain data. Each of these techniques brings improvements over the previous one. In the official evaluation, our system was ranked 1st in the Portuguese-to-Spanish direction, 2nd in the opposite direction, and 3rd in the Catalan-Spanish pair.
In this paper, we present two productive and functional recommender methods to improve the ac- curacy of predicting the right product for the user. One proposal is a survey-based recommender system that uses k-nearest neighbors. It recommends products by asking questions from the user, efficiently applying a binary product vector to the product attributes, and processing the request with a minimum error. The second proposal uses an enriched collaborative-based recommender system using enriched weighted vectors. Thanks to the style rules, the enriched collaborative- based method recommends outfits with competitive recommendation quality. We evaluated both of the proposals on a Kaggle fashion-dataset along with iMaterialist and, results show equivalent performance on binary gender and product attributes.
The 1st ACL workshop on Gender Bias in Natural Language Processing included a shared task on gendered ambiguous pronoun (GAP) resolution. This task was based on the coreference challenge defined in Webster et al. (2018), designed to benchmark the ability of systems to resolve pronouns in real-world contexts in a gender-fair way. 263 teams competed via a Kaggle competition, with the winning system achieving logloss of 0.13667 and near gender parity. We review the approaches of eleven systems with accepted description papers, noting their effective use of BERT (Devlin et al., 2018), both via fine-tuning and for feature extraction, as well as ensembling.
Gender bias is highly impacting natural language processing applications. Word embeddings have clearly been proven both to keep and amplify gender biases that are present in current data sources. Recently, contextualized word embeddings have enhanced previous word embedding techniques by computing word vector representations dependent on the sentence they appear in. In this paper, we study the impact of this conceptual change in the word embedding computation in relation with gender bias. Our analysis includes different measures previously applied in the literature to standard word embeddings. Our findings suggest that contextualized word embeddings are less biased than standard ones even when the latter are debiased.
This paper explains the TALP-UPC participation for the Gendered Pronoun Resolution shared-task of the 1st ACL Workshop on Gender Bias for Natural Language Processing. We have implemented two models for mask language modeling using pre-trained BERT adjusted to work for a classification problem. The proposed solutions are based on the word probabilities of the original BERT model, but using common English names to replace the original test names.
Neural machine translation has significantly pushed forward the quality of the field. However, there are remaining big issues with the output translations and one of them is fairness. Neural models are trained on large text corpora which contain biases and stereotypes. As a consequence, models inherit these social biases. Recent methods have shown results in reducing gender bias in other natural language processing tools such as word embeddings. We take advantage of the fact that word embeddings are used in neural machine translation to propose a method to equalize gender biases in neural machine translation using these representations. Specifically, we propose, experiment and analyze the integration of two debiasing techniques over GloVe embeddings in the Transformer translation architecture. We evaluate our proposed system on the WMT English-Spanish benchmark task, showing gains up to one BLEU point. As for the gender bias evaluation, we generate a test set of occupations and we show that our proposed system learns to equalize existing biases from the baseline system.
This paper presents the results of the premier shared task organized alongside the Conference on Machine Translation (WMT) 2019. Participants were asked to build machine translation systems for any of 18 language pairs, to be evaluated on a test set of news stories. The main metric for this task is human judgment of translation quality. The task was also opened up to additional test suites to probe specific aspects of translation.
In this article, we describe the TALP-UPC research group participation in the WMT19 news translation shared task for Kazakh-English. Given the low amount of parallel training data, we resort to using Russian as pivot language, training subword-based statistical translation systems for Russian-Kazakh and Russian-English that were then used to create two synthetic pseudo-parallel corpora for Kazakh-English and English-Kazakh respectively. Finally, a self-attention model based on the decoder part of the Transformer architecture was trained on the two pseudo-parallel corpora.
In this work, we give a description of the TALP-UPC systems submitted for the WMT19 Biomedical Translation Task. Our proposed strategy is NMT model-independent and relies only on one ingredient, a biomedical terminology list. We first extracted such a terminology list by labelling biomedical words in our training dataset using the BabelNet API. Then, we designed a data preparation strategy to insert the terms information at a token level. Finally, we trained the Transformer model with this terms-informed data. Our best-submitted system ranked 2nd and 3rd for Spanish-English and English-Spanish translation directions, respectively.
Although the problem of similar language translation has been an area of research interest for many years, yet it is still far from being solved. In this paper, we study the performance of two popular approaches: statistical and neural. We conclude that both methods yield similar results; however, the performance varies depending on the language pair. While the statistical approach outperforms the neural one by a difference of 6 BLEU points for the Spanish-Portuguese language pair, the proposed neural model surpasses the statistical one by a difference of 2 BLEU points for Czech-Polish. In the former case, the language similarity (based on perplexity) is much higher than in the latter case. Additionally, we report negative results for the system combination with back-translation. Our TALP-UPC system submission won 1st place for Czech->Polish and 2nd place for Spanish->Portuguese in the official evaluation of the 1st WMT Similar Language Translation task.
The main alternatives nowadays to deal with sequences are Recurrent Neural Networks (RNN) architectures and the Transformer. In this context, Both RNN’s and Transformer have been used as an encoder-decoder architecture with multiple layers in each module. Far beyond this, these architectures are the basis for the contextual word embeddings which are revolutionizing most natural language downstream applications. However, intermediate representations in either the RNN or Transformer architectures can be difficult to interpret. To make these layer representations more accessible and meaningful, we introduce a web-based tool that visualizes them both at the sentence and token level. We present three use cases. The first analyses gender issues in contextual word embeddings. The second and third are showing multilingual intermediate representations for sentences and tokens and the evolution of these intermediate representations along with the multiple layers of the decoder and in the context of multilingual machine translation.
Multilingual Neural Machine Translation approaches are based on the use of task specific models and the addition of one more language can only be done by retraining the whole system. In this work, we propose a new training schedule that allows the system to scale to more languages without modification of the previous components based on joint training and language-independent encoder/decoder modules allowing for zero-shot translation. This work in progress shows close results to state-of-the-art in the WMT task.
In this paper we present the first neural-based machine translation system trained to translate between standard national varieties of the same language. We take the pair Brazilian - European Portuguese as an example and compare the performance of this method to a phrase-based statistical machine translation system. We report a performance improvement of 0.9 BLEU points in translating from European to Brazilian Portuguese and 0.2 BLEU points when translating in the opposite direction. We also carried out a human evaluation experiment with native speakers of Brazilian Portuguese which indicates that humans prefer the output produced by the neural-based system in comparison to the statistical system.
In this article we describe the TALP-UPC research group participation in the WMT18 news shared translation task for Finnish-English and Estonian-English within the multi-lingual subtrack. All of our primary submissions implement an attention-based Neural Machine Translation architecture. Given that Finnish and Estonian belong to the same language family and are similar, we use as training data the combination of the datasets of both language pairs to paliate the data scarceness of each individual pair. We also report the translation quality of systems trained on individual language pair data to serve as baseline and comparison reference.
The Transformer architecture has become the state-of-the-art in Machine Translation. This model, which relies on attention-based mechanisms, has outperformed previous neural machine translation architectures in several tasks. In this system description paper, we report details of training neural machine translation with multi-source Romance languages with the Transformer model and in the evaluation frame of the biomedical WMT 2018 task. Using multi-source languages from the same family allows improvements of over 6 BLEU points.
Catalan and Spanish are two related languages given that both derive from Latin. They share similarities in several linguistic levels including morphology, syntax and semantics. This makes them particularly interesting for the MT task. Given the recent appearance and popularity of neural MT, this paper analyzes the performance of this new approach compared to the well-established rule-based and phrase-based MT systems. Experiments are reported on a large database of 180 million words. Results, in terms of standard automatic measures, show that neural MT clearly outperforms the rule-based and phrase-based MT system on in-domain test set, but it is worst in the out-of-domain test set. A naive system combination specially works for the latter. In-domain manual analysis shows that neural MT tends to improve both adequacy and fluency, for example, by being able to generate more natural translations instead of literal ones, choosing to the adequate target word when the source word has several translations and improving gender agreement. However, out-of-domain manual analysis shows how neural MT is more affected by unknown words or contexts.
This paper presents experiments comparing character-based and byte-based neural machine translation systems. The main motivation of the byte-based neural machine translation system is to build multi-lingual neural machine translation systems that can share the same vocabulary. We compare the performance of both systems in several language pairs and we see that the performance in test is similar for most language pairs while the training time is slightly reduced in the case of byte-based neural machine translation.
Natural language inference (NLI) is a central problem in language understanding. End-to-end artificial neural networks have reached state-of-the-art performance in NLI field recently. In this paper, we propose Character-level Intra Attention Network (CIAN) for the NLI task. In our model, we use the character-level convolutional network to replace the standard word embedding layer, and we use the intra attention to capture the intra-sentence semantics. The proposed CIAN model provides improved results based on a newly published MNLI corpus.
In recent years, machine translation (MT) research has focused on investigating how hybrid machine translation as well as system combination approaches can be designed so that the resulting hybrid translations show an improvement over the individual component translations. As a first step towards achieving this objective we have developed a parallel corpus with source text and the corresponding translation output from a number of machine translation engines, annotated with metadata information, capturing aspects of the translation process performed by the different MT systems. This corpus aims to serve as a basic resource for further research on whether hybrid machine translation algorithms and system combination techniques can benefit from additional (linguistically motivated, decoding, and runtime) information provided by the different systems involved. In this paper, we describe the annotated corpus we have created. We provide an overview on the component MT systems and the XLIFF-based annotation format we have developed. We also report on first experiments with the ML4HMT corpus data.
We present work in progress aiming to build tools for the normalization of User-Generated Content (UGC). As we will see, the task requires the revisiting of the initial steps of NLP processing, since UGC (micro-blog, blog, and, generally, Web 2.0 user texts) presents a number of non-standard communicative and linguistic characteristics, and is in fact much closer to oral and colloquial language than to edited text. We present and characterize a corpus of UGC text in Spanish from three different sources: Twitter, consumer reviews and blogs. We motivate the need for UGC text normalization by analyzing the problems found when processing this type of text through a conventional language processing pipeline, particularly in the tasks of lemmatization and morphosyntactic tagging, and finally we propose a strategy for automatically normalizing UGC using a selector of correct forms on top of a pre-existing spell-checker.
This paper presents a web-based multimedia search engine built within the Buceador (www.buceador.org) research project. A proof-of-concept tool has been implemented which is able to retrieve information from a digital library made of multimedia documents in the 4 official languages in Spain (Spanish, Basque, Catalan and Galician). The retrieved documents are presented in the user language after translation and dubbing (the four previous languages + English). The paper presents the tool functionality, the architecture, the digital library and provide some information about the technology involved in the fields of automatic speech recognition, statistical machine translation, text-to-speech synthesis and information retrieval. Each technology has been adapted to the purposes of the presented tool as well as to interact with the rest of the technologies involved.
We describe the Shared Task on Applying Machine Learning Techniques to Optimise the Division of Labour in Hybrid Machine Translation (ML4HMT) which aims to foster research on improved system combination approaches for machine translation (MT). Participants of the challenge are requested to build hybrid translations by combining the output of several MT systems of different types. We first describe the ML4HMT corpus used in the shared task, then explain the XLIFF-based annotation format we have designed for it, and briefly summarize the participating systems. Using both automated metrics scores and extensive manual evaluation, we discuss the individual performance of the various systems. An interesting result from the shared task is the fact that we were able to observe different systems winning according to the automated metrics scores when compared to the results from the manual evaluation. We conclude by summarising the first edition of the challenge and by giving an outlook to future work.
This paper proposes to introduce a novel reordering model in the open-source Moses toolkit. The main idea is to provide weighted reordering hypotheses to the SMT decoder. These hypotheses are built using a first-step Ngram-based SMT translation from a source language into a third representation that is called reordered source language. Each hypothesis has its own weight provided by the Ngram-based decoder. This proposed reordering technique offers a better and more efficient translation when compared to both the distance-based and the lexicalized reordering. In addition to this reordering approach, this paper describes a domain adaptation technique which is based on a linear combination of an specific in-domain and an extra out-domain translation models. Results for both approaches are reported in the Arabic-to-English 2008 IWSLT task. When implementing the weighted reordering hypotheses and the domain adaptation technique in the final translation system, translation results reach improvements up to 2.5 BLEU compared to a standard state-of-the-art Moses baseline system.
Machine translation systems can be classified into rule-based and corpus-based approaches, in terms of their core technology. Since both paradigms have largely been used during the last years, one of the aims in the research community is to know how these systems differ in terms of translation quality. To this end, this paper reports a study and comparison of a rule-based and a corpus-based (particularly, statistical) Catalan-Spanish machine translation systems, both of them freely available in the web. The translation quality analysis is performed under two different domains: journalistic and medical. The systems are evaluated by using standard automatic measures, as well as by native human evaluators. Automatic results show that the statistical system performs better than the rule-based system. Human judgements show that in the Spanish-to-Catalan direction the statistical system also performs better than the rule-based system, while in the Catalan-to-Spanish direction is the other way round. Although the statistical system obtains the best automatic scores, its errors tend to be more penalized by human judgements than the errors of the rule-based system. This can be explained because statistical errors are usually unexpected and they do not follow any pattern.
This paper describes the UPC-BMIC-VMU participation in the IWSLT 2010 evaluation campaign. The SMT system is a standard phrase-based enriched with novel segmentations. These novel segmentations are computed using statistical measures such as Log-likelihood, T-score, Chi-squared, Dice, Mutual Information or Gravity-Counts. The analysis of translation results allows to divide measures into three groups. First, Log-likelihood, Chi-squared and T-score tend to combine high frequency words and collocation segments are very short. They improve the SMT system by adding new translation units. Second, Mutual Information and Dice tend to combine low frequency words and collocation segments are short. They improve the SMT system by smoothing the translation units. And third, GravityCounts tends to combine high and low frequency words and collocation segments are long. However, in this case, the SMT system is not improved. Thus, the road-map for translation system improvement is to introduce new phrases with either low frequency or high frequency words. It is hard to introduce new phrases with low and high frequency words in order to improve translation quality. Experimental results are reported in the French-to-English IWSLT 2010 evaluation where our system was ranked 3rd out of nine systems.
This paper describes the Barcelona Media SMT system in the IWSLT 2009 evaluation campaign. The Barcelona Media system is an statistical phrase-based system enriched with source context information. Adding source context in an SMT system is interesting to enhance the translation in order to solve lexical and structural choice errors. The novel technique uses a similarity metric among each test sentence and each training sentence. First experimental results of this technique are reported in the Arabic and Chinese Basic Traveling Expression Corpus (BTEC) task. Although working in a single domain, there are ambiguities in SMT translation units and slight improvements in BLEU are shown in both tasks (Zh2En and Ar2En).
Statistical Machine Translation (SMT) is based on alignment models which learn from bilingual corpora the word correspondences between source and target language. These models are assumed to be capable of learning reorderings of sequences of words. However, the difference in word order between two languages is one of the most important sources of errors in SMT. This paper proposes a Recursive Alignment Block Classification algorithm (RABCA) that can take advantage of inductive learning in order to solve reordering problems. This algorithm should be able to cope with swapping examples seen during training; it should infer properties that might permit to reorder pairs of blocks (sequences of words) which did not appear during training; and finally it should be robust with respect to training errors and ambiguities. Experiments are reported on the EuroParl task and RABCA is tested using two state-of-the-art SMT systems: a phrased-based and an Ngram-based. In both cases, RABCA improves results.
Reordering is one source of error in statistical machine translation (SMT). This paper extends the study of the statistical machine reordering (SMR) approach, which uses the powerful techniques of the SMT systems to solve reordering problems. Here, the novelties yield in: (1) using the SMR approach in a SMT phrase-based system, (2) adding a feature function in the SMR step, and (3) analyzing the reordering hypotheses at several stages. Coherent improvements are reported in the TC-STAR task (Es/En) at a relatively low computational cost.
This paper describes TALPtuples, the 2007 N-gram-based statistical machine translation system developed at the TALP Research Center of the UPC (Universitat Polite`cnica de Catalunya) in Barcelona. Emphasis is put on improvements and extensions of the system of previous years. Mainly, these include optimizing alignment parameters in function of translation metric scores and rescoring with a neural network language model. Results on two translation directions are reported, namely from Arabic and Chinese into English, thoroughly explaining all language-related preprocessing and translation schemes.