EMNLP 2020 Papers

Structured prediction is often approached by training a locally normalized model with maximum likelihood and decoding approximately with beam search. This approach leads to mismatches as, during training, the model is not exposed to its mistakes and does not use beam search. Beam-aware training aims to address these problems, but unfortunately, it is not yet widely used due to a lack of understanding about how it impacts performance, when it is most useful, and whether it is stable. Recently, Negrinho et al. (2018) proposed a meta-algorithm that captures beam-aware training algorithms and suggests new ones, but unfortunately did not provide empirical results. In this paper, we begin an empirical investigation: we train the supertagging model of Vaswani et al. (2018) and a simpler model with instantiations of the meta-algorithm. We explore the influence of various design choices and make recommendations for choosing them. We observe that beam-aware training improves performance for both models, with large improvements for the simpler model which must effectively manage uncertainty during decoding. Our results suggest that a model must be learned with search to maximize its effectiveness.

The success of deep learning methods hinges on the availability of large training datasets annotated for the task of interest. In contrast to human intelligence, these methods lack versatility and struggle to learn and adapt quickly to new tasks, where labeled data is scarce. Meta-learning aims to solve this problem by training a model on a large number of few-shot tasks, with an objective to learn new tasks quickly from a small number of examples. In this paper, we propose a meta-learning framework for few-shot word sense disambiguation (WSD), where the goal is to learn to disambiguate unseen words from only a few labeled instances. Meta-learning approaches have so far been typically tested in an N-way, K-shot classification setting where each task has N classes with K examples per class. Owing to its nature, WSD deviates from this controlled setup and requires the models to handle a large number of highly unbalanced classes. We extend several popular meta-learning approaches to this scenario, and analyze their strengths and weaknesses in this new challenging setting.

Transfer learning is an effective technique to improve a target recommender system with the knowledge from a source domain. Existing research focuses on the recommendation performance of the target domain while ignores the privacy leakage of the source domain. The transferred knowledge, however, may unintendedly leak private information of the source domain. For example, an attacker can accurately infer user demographics from their historical purchase provided by a source domain data owner. This paper addresses the above privacy-preserving issue by learning a privacy-aware neural representation by improving target performance while protecting source privacy. The key idea is to simulate the attacks during the training for protecting unseen users’ privacy in the future, modeled by an adversarial game, so that the transfer learning model becomes robust to attacks. Experiments show that the proposed PrivNet model can successfully disentangle the knowledge benefitting the transfer from leaking the privacy.

Non-task-oriented dialog models suffer from poor quality and non-diverse responses. To overcome limited conversational data, we apply Simulated Multiple Reference Training (SMRT; Khayrallah et al., 2020), and use a paraphraser to simulate multiple responses per training prompt. We find SMRT improves over a strong Transformer baseline as measured by human and automatic quality scores and lexical diversity. We also find SMRT is comparable to pretraining in human evaluation quality, and outperforms pretraining on automatic quality and lexical diversity, without requiring related-domain dialog data.

There has been an increased interest in modelling political discourse within the natural language processing (NLP) community, in tasks such as political bias and misinformation detection, among others. Metaphor-rich and emotion-eliciting communication strategies are ubiquitous in political rhetoric, according to social science research. Yet, none of the existing computational models of political discourse has incorporated these phenomena. In this paper, we present the first joint models of metaphor, emotion and political rhetoric, and demonstrate that they advance performance in three tasks: predicting political perspective of news articles, party affiliation of politicians and framing of policy issues.

Many efforts have been devoted to extracting constituency trees from pre-trained language models, often proceeding in two stages: feature definition and parsing. However, this kind of methods may suffer from the branching bias issue, which will inflate the performances on languages with the same branch it biases to. In this work, we propose quantitatively measuring the branching bias by comparing the performance gap on a language and its reversed language, which is agnostic to both language models and extracting methods. Furthermore, we analyze the impacts of three factors on the branching bias, namely feature definitions, parsing algorithms, and language models. Experiments show that several existing works exhibit branching biases, and some implementations of these three factors can introduce the branching bias.

Existing dialogue state tracking (DST) models require plenty of labeled data. However, collecting high-quality labels is costly, especially when the number of domains increases. In this paper, we address a practical DST problem that is rarely discussed, i.e., learning efficiently with limited labeled data. We present and investigate two self-supervised objectives: preserving latent consistency and modeling conversational behavior. We encourage a DST model to have consistent latent distributions given a perturbed input, making it more robust to an unseen scenario. We also add an auxiliary utterance generation task, modeling a potential correlation between conversational behavior and dialogue states. The experimental results show that our proposed self-supervised signals can improve joint goal accuracy by 8.95% when only 1% labeled data is used on the MultiWOZ dataset. We can achieve an additional 1.76% improvement if some unlabeled data is jointly trained as semi-supervised learning. We analyze and visualize how our proposed self-supervised signals help the DST task and hope to stimulate future data-efficient DST research.

In this paper, we present Linguistics Informed Multi-Task BERT (LIMIT-BERT) for learning language representations across multiple linguistics tasks by Multi-Task Learning. LIMIT-BERT includes five key linguistics tasks: Part-Of-Speech (POS) tags, constituent and dependency syntactic parsing, span and dependency semantic role labeling (SRL). Different from recent Multi-Task Deep Neural Networks (MT-DNN), our LIMIT-BERT is fully linguistics motivated and thus is capable of adopting an improved masked training objective according to syntactic and semantic constituents. Besides, LIMIT-BERT takes a semi-supervised learning strategy to offer the same large amount of linguistics task data as that for the language model training. As a result, LIMIT-BERT not only improves linguistics tasks performance but also benefits from a regularization effect and linguistics information that leads to more general representations to help adapt to new tasks and domains. LIMIT-BERT outperforms the strong baseline Whole Word Masking BERT on both dependency and constituent syntactic/semantic parsing, GLUE benchmark, and SNLI task. Our practice on the proposed LIMIT-BERT also enables us to release a well pre-trained model for multi-purpose of natural language processing tasks once for all.

Both syntactic and semantic structures are key linguistic contextual clues, in which parsing the latter has been well shown beneficial from parsing the former. However, few works ever made an attempt to let semantic parsing help syntactic parsing. As linguistic representation formalisms, both syntax and semantics may be represented in either span (constituent/phrase) or dependency, on both of which joint learning was also seldom explored. In this paper, we propose a novel joint model of syntactic and semantic parsing on both span and dependency representations, which incorporates syntactic information effectively in the encoder of neural network and benefits from two representation formalisms in a uniform way. The experiments show that semantics and syntax can benefit each other by optimizing joint objectives. Our single model achieves new state-of-the-art or competitive results on both span and dependency semantic parsing on Propbank benchmarks and both dependency and constituent syntactic parsing on Penn Treebank.

In previous work, artificial agents were shown to achieve almost perfect accuracy in referential games where they have to communicate to identify images. Nevertheless, the resulting communication protocols rarely display salient features of natural languages, such as compositionality. In this paper, we propose some realistic sources of pressure on communication that avert this outcome. More specifically, we formalise the principle of least effort through an auxiliary objective. Moreover, we explore several game variants, inspired by the principle of object constancy, in which we alter the frequency, position, and luminosity of the objects in the images. We perform an extensive analysis on their effect through compositionality metrics, diagnostic classifiers, and zero-shot evaluation. Our findings reveal that the proposed sources of pressure result in emerging languages with less redundancy, more focus on high-level conceptual information, and better abilities of generalisation. Overall, our contributions reduce the gap between emergent and natural languages.

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.

The recently proposed ALFRED challenge task aims for a virtual robotic agent to complete complex multi-step everyday tasks in a virtual home environment from high-level natural language directives, such as “put a hot piece of bread on a plate”. Currently, the best-performing models are able to complete less than 1% of these tasks successfully. In this work we focus on modeling the translation problem of converting natural language directives into detailed multi-step sequences of actions that accomplish those goals in the virtual environment. We empirically demonstrate that it is possible to generate gold multi-step plans from language directives alone without any visual input in 26% of unseen cases. When a small amount of visual information, the starting location in the virtual environment, is incorporated, our best-performing GPT-2 model successfully generates gold command sequences in 58% of cases, suggesting contextualized language models may provide strong planning modules for grounded virtual agents.

Task-agnostic forms of data augmentation have proven widely effective in computer vision, even on pretrained models. In NLP similar results are reported most commonly for low data regimes, non-pretrained models, or situationally for pretrained models. In this paper we ask how effective these techniques really are when applied to pretrained transformers. Using two popular varieties of task-agnostic data augmentation (not tailored to any particular task), Easy Data Augmentation (Wei andZou, 2019) and Back-Translation (Sennrichet al., 2015), we conduct a systematic examination of their effects across 5 classification tasks, 6 datasets, and 3 variants of modern pretrained transformers, including BERT, XLNet, and RoBERTa. We observe a negative result, finding that techniques which previously reported strong improvements for non-pretrained models fail to consistently improve performance for pretrained transformers, even when training data is limited. We hope this empirical analysis helps inform practitioners where data augmentation techniques may confer improvements.

In this paper, we investigate the following two limitations for the existing distractor generation (DG) methods. First, the quality of the existing DG methods are still far from practical use. There are still room for DG quality improvement. Second, the existing DG designs are mainly for single distractor generation. However, for practical MCQ preparation, multiple distractors are desired. Aiming at these goals, in this paper, we present a new distractor generation scheme with multi-tasking and negative answer training strategies for effectively generating multiple distractors. The experimental results show that (1) our model advances the state-of-the-art result from 28.65 to 39.81 (BLEU 1 score) and (2) the generated multiple distractors are diverse and shows strong distracting power for multiple choice question.

This paper proposes a novel approach to learn commonsense from images, instead of limited raw texts or costly constructed knowledge bases, for the commonsense reasoning problem in NLP. Our motivation comes from the fact that an image is worth a thousand words, where richer scene information could be leveraged to help distill the commonsense knowledge, which is often hidden in languages. Our approach, namely Loire, consists of two stages. In the first stage, a bi-modal sequence-to-sequence approach is utilized to conduct the scene layout generation task, based on a text representation model ViBERT. In this way, the required visual scene knowledge, such as spatial relations, will be encoded in ViBERT by the supervised learning process with some bi-modal data like COCO. Then ViBERT is concatenated with a pre-trained language model to perform the downstream commonsense reasoning tasks. Experimental results on two commonsense reasoning problems, i.e.commonsense question answering and pronoun resolution, demonstrate that Loire outperforms traditional language-based methods. We also give some case studies to show what knowledge is learned from images and explain how the generated scene layout helps the commonsense reasoning process.

Morphologically Rich Languages (MRLs) such as Arabic, Hebrew and Turkish often require Morphological Disambiguation (MD), i.e., the prediction of morphological decomposition of tokens into morphemes, early in the pipeline. Neural MD may be addressed as a simple pipeline, where segmentation is followed by sequence tagging, or as an end-to-end model, predicting morphemes from raw tokens. Both approaches are sub-optimal; the former is heavily prone to error propagation, and the latter does not enjoy explicit access to the basic processing units called morphemes. This paper offers MD architecture that combines the symbolic knowledge of morphemes with the learning capacity of neural end-to-end modeling. We propose a new, general and easy-to-implement Pointer Network model where the input is a morphological lattice and the output is a sequence of indices pointing at a single disambiguated path of morphemes. We demonstrate the efficacy of the model on segmentation and tagging, for Hebrew and Turkish texts, based on their respective Universal Dependencies (UD) treebanks. Our experiments show that with complete lattices, our model outperforms all shared-task results on segmenting and tagging these languages. On the SPMRL treebank, our model outperforms all previously reported results for Hebrew MD in realistic scenarios.

Data collection for natural language (NL) understanding tasks has increasingly included human explanations alongside data points, allowing past works to introduce models that both perform a task and generate NL explanations for their outputs. Yet to date, model-generated explanations have been evaluated on the basis of surface-level similarities to human explanations, both through automatic metrics like BLEU and human evaluations. We argue that these evaluations are insufficient, since they fail to indicate whether explanations support actual model behavior (faithfulness), rather than simply match what a human would say (plausibility). In this work, we address the problem of evaluating explanations from the the model simulatability perspective. Our contributions are as follows: (1) We introduce a leakage-adjusted simulatability (LAS) metric for evaluating NL explanations, which measures how well explanations help an observer predict a model’s output, while controlling for how explanations can directly leak the output. We use a model as a proxy for a human observer, and validate this choice with two human subject experiments. (2) Using the CoS-E and e-SNLI datasets, we evaluate two existing generative graphical models and two new approaches; one rationalizing method we introduce achieves roughly human-level LAS scores. (3) Lastly, we frame explanation generation as a multi-agent game and optimize explanations for simulatability while penalizing label leakage, which can improve LAS scores.

Peeking into the inner workings of BERT has shown that its layers resemble the classical NLP pipeline, with progressively more complex tasks being concentrated in later layers. To investigate to what extent these results also hold for a language other than English, we probe a Dutch BERT-based model and the multilingual BERT model for Dutch NLP tasks. In addition, through a deeper analysis of part-of-speech tagging, we show that also within a given task, information is spread over different parts of the network and the pipeline might not be as neat as it seems. Each layer has different specialisations, so that it may be more useful to combine information from different layers, instead of selecting a single one based on the best overall performance.

Reverse dictionary is the task to find the proper target word given the word description. In this paper, we tried to incorporate BERT into this task. However, since BERT is based on the byte-pair-encoding (BPE) subword encoding, it is nontrivial to make BERT generate a word given the description. We propose a simple but effective method to make BERT generate the target word for this specific task. Besides, the cross-lingual reverse dictionary is the task to find the proper target word described in another language. Previous models have to keep two different word embeddings and learn to align these embeddings. Nevertheless, by using the Multilingual BERT (mBERT), we can efficiently conduct the cross-lingual reverse dictionary with one subword embedding, and the alignment between languages is not necessary. More importantly, mBERT can achieve remarkable cross-lingual reverse dictionary performance even without the parallel corpus, which means it can conduct the cross-lingual reverse dictionary with only corresponding monolingual data. Code is publicly available at https://github.com/yhcc/BertForRD.git.