In this study, we propose a self-supervised learning method that distils representations of word meaning in context from a pre-trained masked language model. Word representations are the basis for context-aware lexical semantics and unsupervised semantic textual similarity (STS) estimation. A previous study transforms contextualised representations employing static word embeddings to weaken excessive effects of contextual information. In contrast, the proposed method derives representations of word meaning in context while preserving useful context information intact. Specifically, our method learns to combine outputs of different hidden layers using self-attention through self-supervised learning with an automatically generated training corpus. To evaluate the performance of the proposed approach, we performed comparative experiments using a range of benchmark tasks. The results confirm that our representations exhibited a competitive performance compared to that of the state-of-the-art method transforming contextualised representations for the context-aware lexical semantic tasks and outperformed it for STS estimation.

We create a large-scale dialogue corpus that provides pragmatic paraphrases to advance technology for understanding the underlying intentions of users. While neural conversation models acquire the ability to generate fluent responses through training on a dialogue corpus, previous corpora have mainly focused on the literal meanings of utterances. However, in reality, people do not always present their intentions directly. For example, if a person said to the operator of a reservation service “I don’t have enough budget.”, they, in fact, mean “please find a cheaper option for me.” Our corpus provides a total of 71,498 indirect–direct utterance pairs accompanied by a multi-turn dialogue history extracted from the MultiWoZ dataset. In addition, we propose three tasks to benchmark the ability of models to recognize and generate indirect and direct utterances. We also investigated the performance of state-of-the-art pre-trained models as baselines.

Few-shot text classification aims to classify inputs whose label has only a few examples. Previous studies overlooked the semantic relevance between label representations. Therefore, they are easily confused by labels that are relevant. To address this problem, we propose a method that generates distinct label representations that embed information specific to each label. Our method is applicable to conventional few-shot classification models. Experimental results show that our method significantly improved the performance of few-shot text classification across models and datasets.

Curriculum learning has improved the quality of neural machine translation, where only source-side features are considered in the metrics to determine the difficulty of translation. In this study, we apply curriculum learning to paraphrase generation for the first time. Different from machine translation, paraphrase generation allows a certain level of discrepancy in semantics between source and target, which results in diverse transformations from lexical substitution to reordering of clauses. Hence, the difficulty of transformations requires considering both source and target contexts. Experiments on formality transfer using GYAFC showed that our curriculum learning with edit distance improves the quality of paraphrase generation. Additionally, the proposed method improves the quality of difficult samples, which was not possible for previous methods.

Definition generation techniques aim to generate a definition of a target word or phrase given a context. In previous studies, researchers have faced various issues such as the out-of-vocabulary problem and over/under-specificity problems. Over-specific definitions present narrow word meanings, whereas under-specific definitions present general and context-insensitive meanings. Herein, we propose a method for definition generation with appropriate specificity. The proposed method addresses the aforementioned problems by leveraging a pre-trained encoder-decoder model, namely Text-to-Text Transfer Transformer, and introducing a re-ranking mechanism to model specificity in definitions. Experimental results on standard evaluation datasets indicate that our method significantly outperforms the previous state-of-the-art method. Moreover, manual evaluation confirms that our method effectively addresses the over/under-specificity problems.

We propose a method to distill a language-agnostic meaning embedding from a multilingual sentence encoder. By removing language-specific information from the original embedding, we retrieve an embedding that fully represents the sentence’s meaning. The proposed method relies only on parallel corpora without any human annotations. Our meaning embedding allows efficient cross-lingual sentence similarity estimation by simple cosine similarity calculation. Experimental results on both quality estimation of machine translation and cross-lingual semantic textual similarity tasks reveal that our method consistently outperforms the strong baselines using the original multilingual embedding. Our method consistently improves the performance of any pre-trained multilingual sentence encoder, even in low-resource language pairs where only tens of thousands of parallel sentence pairs are available.

We optimize rewards of reinforcement learning in text simplification using metrics that are highly correlated with human-perspectives. To address problems of exposure bias and loss-evaluation mismatch, text-to-text generation tasks employ reinforcement learning that rewards task-specific metrics. Previous studies in text simplification employ the weighted sum of sub-rewards from three perspectives: grammaticality, meaning preservation, and simplicity. However, the previous rewards do not align with human-perspectives for these perspectives. In this study, we propose to use BERT regressors fine-tuned for grammaticality, meaning preservation, and simplicity as reward estimators to achieve text simplification conforming to human-perspectives. Experimental results show that reinforcement learning with our rewards balances meaning preservation and simplicity. Additionally, human evaluation confirmed that simplified texts by our method are preferred by humans compared to previous studies.

Phrase alignment is the basis for modelling sentence pair interactions, such as paraphrase and textual entailment recognition. Most phrase alignments are compositional processes such that an alignment of a phrase pair is constructed based on the alignments of their child phrases. Nonetheless, studies have revealed that non-compositional alignments involving long-distance phrase reordering are prevalent in practice. We address the phrase alignment problem by combining an unordered tree mapping algorithm and phrase representation modelling that explicitly embeds the similarity distribution in the sentences onto powerful contextualized representations. Experimental results demonstrate that our method effectively handles compositional and non-compositional global phrase alignments. Our method significantly outperforms that used in a previous study and achieves a performance competitive with that of experienced human annotators.

We performed a detailed error analysis in domain-specific neural machine translation (NMT) for the English and Japanese language pair with fine-grained manual annotation. Despite its importance for advancing NMT technologies, research on the performance of domain-specific NMT and non-European languages has been limited. In this study, we designed an error typology based on the error types that were typically generated by NMT systems and might cause significant impact in technical translations: “Addition,” “Omission,” “Mistranslation,” “Grammar,” and “Terminology.” The error annotation was targeted to the medical domain and was performed by experienced professional translators specialized in medicine under careful quality control. The annotation detected 4,912 errors on 2,480 sentences, and the frequency and distribution of errors were analyzed. We found that the major errors in NMT were “Mistranslation” and “Terminology” rather than “Addition” and “Omission,” which have been reported as typical problems of NMT. Interestingly, more errors occurred in documents for professionals compared with those for the general public. The results of our annotation work will be published as a parallel corpus with error labels, which are expected to contribute to developing better NMT models, automatic evaluation metrics, and quality estimation models.

We propose a method to control the specificity of responses while maintaining the consistency with the utterances. We first design a metric based on pointwise mutual information, which measures the co-occurrence degree between an utterance and a response. To control the specificity of generated responses, we add the distant supervision based on the co-occurrence degree and a PMI-based word prediction mechanism to a sequence-to-sequence model. With these mechanisms, our model outputs the words with optimal specificity for a given specificity control variable. In experiments with open-domain dialogue corpora, automatic and human evaluation results confirm that our model controls the specificity of the response more sensitively than the conventional model and can generate highly consistent responses.

We reduce the model size of pre-trained word embeddings by a factor of 200 while preserving its quality. Previous studies in this direction created a smaller word embedding model by reconstructing pre-trained word representations from those of subwords, which allows to store only a smaller number of subword embeddings in the memory. However, previous studies that train the reconstruction models using only target words cannot reduce the model size extremely while preserving its quality. Inspired by the observation of words with similar meanings having similar embeddings, our reconstruction training learns the global relationships among words, which can be employed in various models for word embedding reconstruction. Experimental results on word similarity benchmarks show that the proposed method improves the performance of the all subword-based reconstruction models.

Adverse drug reactions are a severe problem that significantly degrade quality of life, or even threaten the life of patients. Patient-generated texts available on the web have been gaining attention as a promising source of information in this regard. While previous studies annotated such patient-generated content, they only reported on limited information, such as whether a text described an adverse drug reaction or not. Further, they only annotated short texts of a few sentences crawled from online forums and social networking services. The dataset we present in this paper is unique for the richness of annotated information, including detailed descriptions of drug reactions with full context. We crawled patient’s weblog articles shared on an online patient-networking platform and annotated the effects of drugs therein reported. We identified spans describing drug reactions and assigned labels for related drug names, standard codes for the symptoms of the reactions, and types of effects. As a first dataset, we annotated 677 drug reactions with these detailed labels based on 169 weblog articles by Japanese lung cancer patients. Our annotation dataset is made publicly available at our web site (https://yukiar.github.io/adr-jp/) for further research on the detection of adverse drug reactions and more broadly, on patient-generated text processing.

We present SAPPHIRE, a Simple Aligner for Phrasal Paraphrase with HIerarchical REpresentation. Monolingual phrase alignment is a fundamental problem in natural language understanding and also a crucial technique in various applications such as natural language inference and semantic textual similarity assessment. Previous methods for monolingual phrase alignment are language-resource intensive; they require large-scale synonym/paraphrase lexica and high-quality parsers. Different from them, SAPPHIRE depends only on a monolingual corpus to train word embeddings. Therefore, it is easily transferable to specific domains and different languages. Specifically, SAPPHIRE first obtains word alignments using pre-trained word embeddings and then expands them to phrase alignments by bilingual phrase extraction methods. To estimate the likelihood of phrase alignments, SAPPHIRE uses phrase embeddings that are hierarchically composed of word embeddings. Finally, SAPPHIRE searches for a set of consistent phrase alignments on a lattice of phrase alignment candidates. It achieves search-efficiency by constraining the lattice so that all the paths go through a phrase alignment pair with the highest alignment score. Experimental results using the standard dataset for phrase alignment evaluation show that SAPPHIRE outperforms the previous method and establishes the state-of-the-art performance.

Advanced pre-trained models for text representation have achieved state-of-the-art performance on various text classification tasks. However, the discrepancy between the semantic similarity of texts and labelling standards affects classifiers, i.e. leading to lower performance in cases where classifiers should assign different labels to semantically similar texts. To address this problem, we propose a simple multitask learning model that uses negative supervision. Specifically, our model encourages texts with different labels to have distinct representations. Comprehensive experiments show that our model outperforms the state-of-the-art pre-trained model on both single- and multi-label classifications, sentence and document classifications, and classifications in three different languages.

Sequence-to-sequence models are a common approach to develop a chatbot. They can train a conversational model in an end-to-end manner. One significant drawback of such a neural network based approach is that the response generation process is a black-box, and how a specific response is generated is unclear. To tackle this problem, an interpretable response generation mechanism is desired. As a step toward this direction, we focus on dialogue-acts (DAs) that may provide insight to understand the response generation process. In particular, we propose a method to predict a DA of the next response based on the history of previous utterances and their DAs. Experiments using a Switch Board Dialogue Act corpus show that compared to the baseline considering only a single utterance, our model achieves 10.8% higher F1-score and 3.0% higher accuracy on DA prediction.

We propose a method to control the level of a sentence in a text simplification task. Text simplification is a monolingual translation task translating a complex sentence into a simpler and easier to understand the alternative. In this study, we use the grade level of the US education system as the level of the sentence. Our text simplification method succeeds in translating an input into a specific grade level by considering levels of both sentences and words. Sentence level is considered by adding the target grade level as input. By contrast, the word level is considered by adding weights to the training loss based on words that frequently appear in sentences of the desired grade level. Although existing models that consider only the sentence level may control the syntactic complexity, they tend to generate words beyond the target level. Our approach can control both the lexical and syntactic complexity and achieve an aggressive rewriting. Experiment results indicate that the proposed method improves the metrics of both BLEU and SARI.

A sequence-to-sequence model tends to generate generic responses with little information for input utterances. To solve this problem, we propose a neural model that generates relevant and informative responses. Our model has simple architecture to enable easy application to existing neural dialogue models. Specifically, using positive pointwise mutual information, it first identifies keywords that frequently co-occur in responses given an utterance. Then, the model encourages the decoder to use the keywords for response generation. Experiment results demonstrate that our model successfully diversifies responses relative to previous models.

A neural conversation model is a promising approach to develop dialogue systems with the ability of chit-chat. It allows training a model in an end-to-end manner without complex rule design nor feature engineering. However, as a side effect, the neural model tends to generate safe but uninformative and insensitive responses like “OK” and “I don’t know.” Such replies are called generic responses and regarded as a critical problem for user-engagement of dialogue systems. For a more engaging chit-chat experience, we propose a neural conversation model that generates responsive and self-expressive replies. Specifically, our model generates domain-aware and sentiment-rich responses. Experiments empirically confirmed that our model outperformed the sequence-to-sequence model; 68.1% of our responses were domain-aware with sentiment polarities, which was only 2.7% for responses generated by the sequence-to-sequence model.

A semantic equivalence assessment is defined as a task that assesses semantic equivalence in a sentence pair by binary judgment (i.e., paraphrase identification) or grading (i.e., semantic textual similarity measurement). It constitutes a set of tasks crucial for research on natural language understanding. Recently, BERT realized a breakthrough in sentence representation learning (Devlin et al., 2019), which is broadly transferable to various NLP tasks. While BERT’s performance improves by increasing its model size, the required computational power is an obstacle preventing practical applications from adopting the technology. Herein, we propose to inject phrasal paraphrase relations into BERT in order to generate suitable representations for semantic equivalence assessment instead of increasing the model size. Experiments on standard natural language understanding tasks confirm that our method effectively improves a smaller BERT model while maintaining the model size. The generated model exhibits superior performance compared to a larger BERT model on semantic equivalence assessment tasks. Furthermore, it achieves larger performance gains on tasks with limited training datasets for fine-tuning, which is a property desirable for transfer learning.

Lexical substitution ranks substitution candidates from the viewpoint of paraphrasability for a target word in a given sentence. There are two major approaches for lexical substitution: (1) generating contextualized word embeddings by assigning multiple embeddings to one word and (2) generating context embeddings using the sentence. Herein we propose a method that combines these two approaches to contextualize word embeddings for lexical substitution. Experiments demonstrate that our method outperforms the current state-of-the-art method. We also create CEFR-LP, a new evaluation dataset for the lexical substitution task. It has a wider coverage of substitution candidates than previous datasets and assigns English proficiency levels to all target words and substitution candidates.

The word order between source and target languages significantly influences the translation quality. Preordering can effectively address this problem. Previous preordering methods require a manual feature design, making language dependent design difficult. In this paper, we propose a preordering method with recursive neural networks that learn features from raw inputs. Experiments show the proposed method is comparable to the state-of-the-art method but without a manual feature design.

To improve the translation adequacy in neural machine translation (NMT), we propose a rewarding model with target word prediction using bilingual dictionaries inspired by the success of decoder constraints in statistical machine translation. In particular, the model first predicts a set of target words promising for translation; then boosts the probabilities of the predicted words to give them better chances to be output. Our rewarding model minimally interacts with the decoder so that it can be easily applied to the decoder of an existing NMT system. Extensive evaluation under both resource-rich and resource-poor settings shows that (1) BLEU score improves more than 10 points with oracle prediction, (2) BLEU score improves about 1.0 point with target word prediction using bilingual dictionaries created either manually or automatically, (3) hyper-parameters of our model are relatively easy to optimize, and (4) undergeneration problem can be alleviated in exchange for increasing over-generated words.

We propose an efficient method to conduct phrase alignment on parse forests for paraphrase detection. Unlike previous studies, our method identifies syntactic paraphrases under linguistically motivated grammar. In addition, it allows phrases to non-compositionally align to handle paraphrases with non-homographic phrase correspondences. A dataset that provides gold parse trees and their phrase alignments is created. The experimental results confirm that the proposed method conducts highly accurate phrase alignment compared to human performance.