COLING 2020 Papers

Recent work has shown that deeper character-based neural machine translation (NMT) models can outperform subword-based models. However, it is still unclear what makes deeper character-based models successful. In this paper, we conduct an investigation into pure character-based models in the case of translating Finnish into English, including exploring the ability to learn word senses and morphological inflections and the attention mechanism. We demonstrate that word-level information is distributed over the entire character sequence rather than over a single character, and characters at different positions play different roles in learning linguistic knowledge. In addition, character-based models need more layers to encode word senses which explains why only deeper models outperform subword-based models. The attention distribution pattern shows that separators attract a lot of attention and we explore a sparse word-level attention to enforce character hidden states to capture the full word-level information. Experimental results show that the word-level attention with a single head results in 1.2 BLEU points drop.

This paper analyzes the pre-trained hidden representations learned from reviews on BERT for tasks in aspect-based sentiment analysis (ABSA). Our work is motivated by the recent progress in BERT-based language models for ABSA. However, it is not clear how the general proxy task of (masked) language model trained on unlabeled corpus without annotations of aspects or opinions can provide important features for downstream tasks in ABSA. By leveraging the annotated datasets in ABSA, we investigate both the attentions and the learned representations of BERT pre-trained on reviews. We found that BERT uses very few self-attention heads to encode context words (such as prepositions or pronouns that indicating an aspect) and opinion words for an aspect. Most features in the representation of an aspect are dedicated to the fine-grained semantics of the domain (or product category) and the aspect itself, instead of carrying summarized opinions from its context. We hope this investigation can help future research in improving self-supervised learning, unsupervised learning and fine-tuning for ABSA. The pre-trained model and code can be found at https://github.com/howardhsu/BERT-for-RRC-ABSA.

While neural approaches have achieved significant improvement in machine comprehension tasks, models often work as a black-box, resulting in lower interpretability, which requires special attention in domains such as healthcare or education. Quantifying uncertainty helps pave the way towards more interpretable neural networks. In classification and regression tasks, Bayesian neural networks have been effective in estimating model uncertainty. However, inference time increases linearly due to the required sampling process in Bayesian neural networks. Thus speed becomes a bottleneck in tasks with high system complexity such as question-answering or dialogue generation. In this work, we propose a hybrid neural architecture to quantify model uncertainty using Bayesian weight approximation but boosts up the inference speed by 80% relative at test time, and apply it for a clinical dialogue comprehension task. The proposed approach is also used to enable active learning so that an updated model can be trained more optimally with new incoming data by selecting samples that are not well-represented in the current training scheme.

Dialogue act classification becomes a complex task when dealing with fine-grain labels. Many applications require such level of labelling, typically automatic dialogue systems. We present in this paper a 2-level classification technique, distinguishing between generic and specific dialogue acts (DA). This approach makes it possible to benefit from the very good accuracy of generic DA classification at the first level and proposes an efficient approach for specific DA, based on high-level linguistic features. Our results show the interest of involving such features into the classifiers, outperforming all other feature sets, in particular those classically used in DA classification.

Word sense disambiguation (WSD) is a fundamental natural language processing task. Unsupervised knowledge-based WSD only relies on a lexical knowledge base as the sense inventory and has wider practical use than supervised WSD that requires a mass of sense-annotated data. HowNet is the most widely used lexical knowledge base in Chinese WSD. Because of its uniqueness, however, most of existing unsupervised WSD methods cannot work for HowNet-based WSD, and the tailor-made methods have not obtained satisfying results. In this paper, we propose a new unsupervised method for HowNet-based Chinese WSD, which exploits the masked language model task of pre-trained language models. In experiments, considering existing evaluation dataset is small and out-of-date, we build a new and larger HowNet-based WSD dataset. Experimental results demonstrate that our model achieves significantly better performance than all the baseline methods. All the code and data of this paper are available at https://github.com/thunlp/SememeWSD.

The transition-based systems in the past studies propose a series of actions, to build a right-heavy binarized tree for the RST parsing. However, the nodes of the binary-nuclear relations (e.g., Contrast) have the same nuclear type with those of the multi-nuclear relations (e.g., Joint) in the binary tree structure. In addition, the reduce action only construct binary trees instead of multi-branch trees, which is the original RST tree structure. In our paper, we design a new nuclear type for the multi-nuclear relations, and a new action to construct a multi-branch tree. We enrich the feature set by extracting additional refined dependency feature of texts from the Bi-Affine model. We also compare the performance of two approaches for RST parsing in the transition-based system: a joint action of reduce-shift and nuclear type (i.e., Reduce-SN) vs a separate one that applies Reduce action first and then assigns nuclear type. We find that the new devised nuclear type and action are more capable of capturing the multi-nuclear relation and the joint action is more suitable than the separate one. Our multi-branch tree structure obtains the state-of-the-art performance for all the 18 coarse relations.

Understanding neural models is a major topic of interest in the deep learning community. In this paper, we propose to interpret a general neural model comparatively. Specifically, we study the sequence-to-sequence (Seq2Seq) model in the contexts of two mainstream NLP tasks–machine translation and dialogue response generation–as they both use the seq2seq model. We investigate how the two tasks are different and how their task difference results in major differences in the behaviors of the resulting translation and dialogue generation systems. This study allows us to make several interesting observations and gain valuable insights, which can be used to help develop better translation and dialogue generation models. To our knowledge, no such comparative study has been done so far.

Sparsity is regarded as a desirable property of representations, especially in terms of explanation. However, its usage has been limited due to the gap with dense representations. Most research progresses in NLP in recent years are based on dense representations. Thus the desirable property of sparsity cannot be leveraged. Inspired by Fourier Transformation, in this paper, we propose a novel Semantic Transformation method to bridge the dense and sparse spaces, which can facilitate the NLP research to shift from dense spaces to sparse spaces or to jointly use both spaces. Experiments using classification tasks and natural language inference task show that the proposed Semantic Transformation is effective.

Recent years have seen big advances in the field of sentence-level quality estimation (QE), largely as a result of using neural-based architectures. However, the majority of these methods work only on the language pair they are trained on and need retraining for new language pairs. This process can prove difficult from a technical point of view and is usually computationally expensive. In this paper we propose a simple QE framework based on cross-lingual transformers, and we use it to implement and evaluate two different neural architectures. Our evaluation shows that the proposed methods achieve state-of-the-art results outperforming current open-source quality estimation frameworks when trained on datasets from WMT. In addition, the framework proves very useful in transfer learning settings, especially when dealing with low-resourced languages, allowing us to obtain very competitive results.

We demonstrate TrainX, a system for Named Entity Linking for medical experts. It combines state-of-the-art entity recognition and linking architectures, such as Flair and fine-tuned Bi-Encoders based on BERT, with an easy-to-use interface for healthcare professionals. We support medical experts in annotating training data by using active sampling strategies to forward informative samples to the annotator. We demonstrate that our model is capable of linking against large knowledge bases, such as UMLS (3.6 million entities), and supporting zero-shot cases, where the linker has never seen the entity before. Those zero-shot capabilities help to mitigate the problem of rare and expensive training data that is a common issue in the medical domain.

In this paper we propose a unified approach for supporting different generation manners of machine translation, including autoregressive, semi-autoregressive, and refinement-based non-autoregressive models. Our approach works by repeatedly selecting positions and generating tokens at these selected positions. After being trained once, our approach achieves better or competitive translation performance compared with some strong task-specific baseline models in all the settings. This generalization ability benefits mainly from the new training objective that we propose. We validate our approach on the WMT’14 English-German and IWSLT’14 German-English translation tasks. The experimental results are encouraging.

The Sumerian cuneiform script was invented more than 5,000 years ago and represents one of the oldest in history. We present the first attempt to translate Sumerian texts into English automatically. We publicly release high-quality corpora for standardized training and evaluation and report results on experiments with supervised, phrase-based, and transfer learning techniques for machine translation. Quantitative and qualitative evaluations indicate the usefulness of the translations. Our proposed methodology provides a broader audience of researchers with novel access to the data, accelerates the costly and time-consuming manual translation process, and helps them better explore the relationships between Sumerian cuneiform and Mesopotamian culture.

Industry-scale NLP systems necessitate two features. 1. Robustness: “zero-shot transfer learning” (ZSTL) performance has to be commendable and 2. Efficiency: systems have to train efficiently and respond instantaneously. In this paper, we introduce the development of a production model called GAAMA (Go Ahead Ask Me Anything) which possess the above two characteristics. For robustness, it trains on the recently introduced Natural Questions (NQ) dataset. NQ poses additional challenges over older datasets like SQuAD: (a) QA systems need to read and comprehend an entire Wikipedia article rather than a small passage, and (b) NQ does not suffer from observation bias during construction, resulting in less lexical overlap between the question and the article. GAAMA consists of Attention-over-Attention, diversity among attention heads, hierarchical transfer learning, and synthetic data augmentation while being computationally inexpensive. Building on top of the powerful BERTQA model, GAAMA provides a ∼2.0% absolute boost in F1 over the industry-scale state-of-the-art (SOTA) system on NQ. Further, we show that GAAMA transfers zero-shot to unseen real life and important domains as it yields respectable performance on two benchmarks: the BioASQ and the newly introduced CovidQA datasets.

Questions under Discussion (QUD; Roberts, 2012) are emerging as a conceptually fruitful approach to spelling out the connection between the information structure of a sentence and the nature of the discourse in which the sentence can function. To make this approach useful for analyzing authentic data, Riester, Brunetti & De Kuthy (2018) presented a discourse annotation framework based on explicit pragmatic principles for determining a QUD for every assertion in a text. De Kuthy et al. (2018) demonstrate that this supports more reliable discourse structure annotation, and Ziai and Meurers (2018) show that based on explicit questions, automatic focus annotation becomes feasible. But both approaches are based on manually specified questions. In this paper, we present an automatic question generation approach to partially automate QUD annotation by generating all potentially relevant questions for a given sentence. While transformation rules can concisely capture the typical question formation process, a rule-based approach is not sufficiently robust for authentic data. We therefore employ the transformation rules to generate a large set of sentence-question-answer triples and train a neural question generation model on them to obtain both systematic question type coverage and robustness.

Conversational recommender systems (CRS) aim to recommend high-quality items to users through interactive conversations. To develop an effective CRS, the support of high-quality datasets is essential. Existing CRS datasets mainly focus on immediate requests from users, while lack proactive guidance to the recommendation scenario. In this paper, we contribute a new CRS dataset named TG-ReDial (Recommendation through Topic-Guided Dialog). Our dataset has two major features. First, it incorporates topic threads to enforce natural semantic transitions towards the recommendation scenario. Second, it is created in a semi-automatic way, hence human annotation is more reasonable and controllable. Based on TG-ReDial, we present the task of topic-guided conversational recommendation, and propose an effective approach to this task. Extensive experiments have demonstrated the effectiveness of our approach on three sub-tasks, namely topic prediction, item recommendation and response generation. TG-ReDial is available at bluehttps://github.com/RUCAIBox/TG-ReDial.

This article reports on an ongoing project aiming at automatization of pseudonymization of learner essays. The process includes three steps: identification of personal information in an unstructured text, labeling for a category, and pseudonymization. We experiment with rule-based methods for detection of 15 categories out of the suggested 19 (Megyesi et al., 2018) that we deem important and/or doable with automatic approaches. For the detection and labeling steps,we use resources covering personal names, geographic names, company and university names and others. For the pseudonymization step, we replace the item using another item of the same type from the above-mentioned resources. Evaluation of the detection and labeling steps are made on a set of manually anonymized essays. The results are promising and show that 89% of the personal information can be successfully identified in learner data, and annotated correctly with an inter-annotator agreement of 86% measured as Fleiss kappa and Krippendorff’s alpha.

Visual Question Answering (VQA) remains algorithmically challenging while it is effortless for humans. Humans combine visual observations with general and commonsense knowledge to answer questions about a given image. In this paper, we address the problem of incorporating general knowledge into VQA models while leveraging the visual information. We propose a model that captures the interactions between objects in a visual scene and entities in an external knowledge source. Our model is a graph-based approach that combines scene graphs with concept graphs, which learns a question-adaptive graph representation of related knowledge instances. We use Graph Attention Networks to set higher importance to key knowledge instances that are mostly relevant to each question. We exploit ConceptNet as the source of general knowledge and evaluate the performance of our model on the challenging OK-VQA dataset.

Current methods of cross-lingual parser transfer focus on predicting the best parser for a low-resource target language globally, that is, “at treebank level”. In this work, we propose and argue for a novel cross-lingual transfer paradigm: instance-level parser selection (ILPS), and present a proof-of-concept study focused on instance-level selection in the framework of delexicalized parser transfer. Our work is motivated by an empirical observation that different source parsers are the best choice for different Universal POS-sequences (i.e., UPOS sentences) in the target language. We then propose to predict the best parser at the instance level. To this end, we train a supervised regression model, based on the Transformer architecture, to predict parser accuracies for individual POS-sequences. We compare ILPS against two strong single-best parser selection baselines (SBPS): (1) a model that compares POS n-gram distributions between the source and target languages (KL) and (2) a model that selects the source based on the similarity between manually created language vectors encoding syntactic properties of languages (L2V). The results from our extensive evaluation, coupling 42 source parsers and 20 diverse low-resource test languages, show that ILPS outperforms KL and L2V on 13/20 and 14/20 test languages, respectively. Further, we show that by predicting the best parser “at treebank level” (SBPS), using the aggregation of predictions from our instance-level model, we outperform the same baselines on 17/20 and 16/20 test languages.

The ambiguous annotation criteria lead to divergence of Chinese Word Segmentation (CWS) datasets in various granularities. Multi-criteria Chinese word segmentation aims to capture various annotation criteria among datasets and leverage their common underlying knowledge. In this paper, we propose a domain adaptive segmenter to exploit diverse criteria of various datasets. Our model is based on Bidirectional Encoder Representations from Transformers (BERT), which is responsible for introducing open-domain knowledge. Private and shared projection layers are proposed to capture domain-specific knowledge and common knowledge, respectively. We also optimize computational efficiency via distillation, quantization, and compiler optimization. Experiments show that our segmenter outperforms the previous state of the art (SOTA) models on 10 CWS datasets with superior efficiency.