EMNLP 2019 Papers

This paper studies the problem of non-factoid question answering, where the answer may span over multiple sentences. Existing solutions can be categorized into representation- and interaction-focused approaches. We combine their complementary strength, by a hybrid approach allowing multi-granular interactions, but represented at word level, enabling an easy integration with strong word-level signals. Specifically, we propose MICRON: Multigranular Interaction for Contextualizing RepresentatiON, a novel approach which derives contextualized uni-gram representation from n-grams. Our contributions are as follows: First, we enable multi-granular matches between question and answer n-grams. Second, by contextualizing word representation with surrounding n-grams, MICRON can naturally utilize word-based signals for query term weighting, known to be effective in information retrieval. We validate MICRON in two public non-factoid question answering datasets: WikiPassageQA and InsuranceQA, showing our model achieves the state of the art among baselines with reported performances on both datasets.

Machine Reading Comprehension (MRC) has become enormously popular recently and has attracted a lot of attention. However, the existing reading comprehension datasets are mostly in English. In this paper, we introduce a Span-Extraction dataset for Chinese machine reading comprehension to add language diversities in this area. The dataset is composed by near 20,000 real questions annotated on Wikipedia paragraphs by human experts. We also annotated a challenge set which contains the questions that need comprehensive understanding and multi-sentence inference throughout the context. We present several baseline systems as well as anonymous submissions for demonstrating the difficulties in this dataset. With the release of the dataset, we hosted the Second Evaluation Workshop on Chinese Machine Reading Comprehension (CMRC 2018). We hope the release of the dataset could further accelerate the Chinese machine reading comprehension research. Resources are available: https://github.com/ymcui/cmrc2018

BERT model has been successfully applied to open-domain QA tasks. However, previous work trains BERT by viewing passages corresponding to the same question as independent training instances, which may cause incomparable scores for answers from different passages. To tackle this issue, we propose a multi-passage BERT model to globally normalize answer scores across all passages of the same question, and this change enables our QA model find better answers by utilizing more passages. In addition, we find that splitting articles into passages with the length of 100 words by sliding window improves performance by 4%. By leveraging a passage ranker to select high-quality passages, multi-passage BERT gains additional 2%. Experiments on four standard benchmarks showed that our multi-passage BERT outperforms all state-of-the-art models on all benchmarks. In particular, on the OpenSQuAD dataset, our model gains 21.4% EM and 21.5% F1 over all non-BERT models, and 5.8% EM and 6.5% F1 over BERT-based models.

Theory of mind, i.e., the ability to reason about intents and beliefs of agents is an important task in artificial intelligence and central to resolving ambiguous references in natural language dialogue. In this work, we revisit the evaluation of theory of mind through question answering. We show that current evaluation methods are flawed and that existing benchmark tasks can be solved without theory of mind due to dataset biases. Based on prior work, we propose an improved evaluation protocol and dataset in which we explicitly control for data regularities via a careful examination of the answer space. We show that state-of-the-art methods which are successful on existing benchmarks fail to solve theory-of-mind tasks in our proposed approach.

Scenario-based question answering (SQA) has attracted increasing research attention. It typically requires retrieving and integrating knowledge from multiple sources, and applying general knowledge to a specific case described by a scenario. SQA widely exists in the medical, geography, and legal domains—both in practice and in the exams. In this paper, we introduce the GeoSQA dataset. It consists of 1,981 scenarios and 4,110 multiple-choice questions in the geography domain at high school level, where diagrams (e.g., maps, charts) have been manually annotated with natural language descriptions to benefit NLP research. Benchmark results on a variety of state-of-the-art methods for question answering, textual entailment, and reading comprehension demonstrate the unique challenges presented by SQA for future research.

While models for Visual Question Answering (VQA) have steadily improved over the years, interacting with one quickly reveals that these models lack consistency. For instance, if a model answers “red” to “What color is the balloon?”, it might answer “no” if asked, “Is the balloon red?”. These responses violate simple notions of entailment and raise questions about how effectively VQA models ground language. In this work, we introduce a dataset, ConVQA, and metrics that enable quantitative evaluation of consistency in VQA. For a given observable fact in an image (e.g. the balloon’s color), we generate a set of logically consistent question-answer (QA) pairs (e.g. Is the balloon red?) and also collect a human-annotated set of common-sense based consistent QA pairs (e.g. Is the balloon the same color as tomato sauce?). Further, we propose a consistency-improving data augmentation module, a Consistency Teacher Module (CTM). CTM automatically generates entailed (or similar-intent) questions for a source QA pair and fine-tunes the VQA model if the VQA’s answer to the entailed question is consistent with the source QA pair. We demonstrate that our CTM-based training improves the consistency of VQA models on the Con-VQA datasets and is a strong baseline for further research.

Feature norm datasets of human conceptual knowledge, collected in surveys of human volunteers, yield highly interpretable models of word meaning and play an important role in neurolinguistic research on semantic cognition. However, these datasets are limited in size due to practical obstacles associated with exhaustively listing properties for a large number of words. In contrast, the development of distributional modelling techniques and the availability of vast text corpora have allowed researchers to construct effective vector space models of word meaning over large lexicons. However, this comes at the cost of interpretable, human-like information about word meaning. We propose a method for mapping human property knowledge onto a distributional semantic space, which adapts the word2vec architecture to the task of modelling concept features. Our approach gives a measure of concept and feature affinity in a single semantic space, which makes for easy and efficient ranking of candidate human-derived semantic properties for arbitrary words. We compare our model with a previous approach, and show that it performs better on several evaluation tasks. Finally, we discuss how our method could be used to develop efficient sampling techniques to extend existing feature norm datasets in a reliable way.

Domain-specific training typically makes NLP systems work better. We show that this extends to cognitive modeling as well by relating the states of a neural phrase-structure parser to electrophysiological measures from human participants. These measures were recorded as participants listened to a spoken recitation of the same literary text that was supplied as input to the neural parser. Given more training data, the system derives a better cognitive model — but only when the training examples come from the same textual genre. This finding is consistent with the idea that humans adapt syntactic expectations to particular genres during language comprehension (Kaan and Chun, 2018; Branigan and Pickering, 2017).

Representational Similarity Analysis (RSA) is a technique developed by neuroscientists for comparing activity patterns of different measurement modalities (e.g., fMRI, electrophysiology, behavior). As a framework, RSA has several advantages over existing approaches to interpretation of language encoders based on probing or diagnostic classification: namely, it does not require large training samples, is not prone to overfitting, and it enables a more transparent comparison between the representational geometries of different models and modalities. We demonstrate the utility of RSA by establishing a previously unknown correspondence between widely-employed pretrained language encoders and human processing difficulty via eye-tracking data, showcasing its potential in the interpretability toolbox for neural models.

Recurrent neural networks can learn to predict upcoming words remarkably well on average; in syntactically complex contexts, however, they often assign unexpectedly high probabilities to ungrammatical words. We investigate to what extent these shortcomings can be mitigated by increasing the size of the network and the corpus on which it is trained. We find that gains from increasing network size are minimal beyond a certain point. Likewise, expanding the training corpus yields diminishing returns; we estimate that the training corpus would need to be unrealistically large for the models to match human performance. A comparison to GPT and BERT, Transformer-based models trained on billions of words, reveals that these models perform even more poorly than our LSTMs in some constructions. Our results make the case for more data efficient architectures.

Memory neurons of long short-term memory (LSTM) networks encode and process information in powerful yet mysterious ways. While there has been work to analyze their behavior in carrying low-level information such as linguistic properties, how they directly contribute to label prediction remains unclear. We find inspiration from biologists and study the affinity between individual neurons and labels, propose a novel metric to quantify the sensitivity of neurons to each label, and conduct experiments to show the validity of our proposed metric. We discover that some neurons are trained to specialize on a subset of labels, and while dropping an arbitrary neuron has little effect on the overall accuracy of the model, dropping label-specialized neurons predictably and significantly degrades prediction accuracy on the associated label. We further examine the consistency of neuron-label affinity across different models. These observations provide insight into the inner mechanisms of LSTMs.

We propose new BERT-based methods for recognizing event causality such as “smoke cigarettes” –> “die of lung cancer” written in web texts. In our methods, we grasp each annotator’s policy by training multiple classifiers, each of which predicts the labels given by a single annotator, and combine the resulting classifiers’ outputs to predict the final labels determined by majority vote. Furthermore, we investigate the effect of supplying background knowledge to our classifiers. Since BERT models are pre-trained with a large corpus, some sort of background knowledge for event causality may be learned during pre-training. Our experiments with a Japanese dataset suggest that this is actually the case: Performance improved when we pre-trained the BERT models with web texts containing a large number of event causalities instead of Wikipedia articles or randomly sampled web texts. However, this effect was limited. Therefore, we further improved performance by simply adding texts related to an input causality candidate as background knowledge to the input of the BERT models. We believe these findings indicate a promising future research direction.

Generating a long, coherent text such as a paragraph requires a high-level control of different levels of relations between sentences (e.g., tense, coreference). We call such a logical connection between sentences as a (paragraph) flow. In order to produce a coherent flow of text, we explore two forms of intersentential relations in a paragraph: one is a human-created linguistical relation that forms a structure (e.g., discourse tree) and the other is a relation from latent representation learned from the sentences themselves. Our two proposed models incorporate each form of relations into document-level language models: the former is a supervised model that jointly learns a language model as well as discourse relation prediction, and the latter is an unsupervised model that is hierarchically conditioned by a recurrent neural network (RNN) over the latent information. Our proposed models with both forms of relations outperform the baselines in partially conditioned paragraph generation task. Our codes and data are publicly available.

We apply BERT to coreference resolution, achieving a new state of the art on the GAP (+11.5 F1) and OntoNotes (+3.9 F1) benchmarks. A qualitative analysis of model predictions indicates that, compared to ELMo and BERT-base, BERT-large is particularly better at distinguishing between related but distinct entities (e.g., President and CEO), but that there is still room for improvement in modeling document-level context, conversations, and mention paraphrasing. We will release all code and trained models upon publication.

Rhetorical Structure Theory (RST) parsing is crucial for many downstream NLP tasks that require a discourse structure for a text. Most of the previous RST parsers have been based on supervised learning approaches. That is, they require an annotated corpus of sufficient size and quality, and heavily rely on the language and domain dependent corpus. In this paper, we present two language-independent unsupervised RST parsing methods based on dynamic programming. The first one builds the optimal tree in terms of a dissimilarity score function that is defined for splitting a text span into smaller ones. The second builds the optimal tree in terms of a similarity score function that is defined for merging two adjacent spans into a large one. Experimental results on English and German RST treebanks showed that our parser based on span merging achieved the best score, around 0.8 F_1 score, which is close to the scores of the previous supervised parsers.

Implicit discourse relation classification is one of the most difficult tasks in discourse parsing. Previous studies have generally focused on extracting better representations of the relational arguments. In order to solve the task, it is however additionally necessary to capture what events are expected to cause or follow each other. Current discourse relation classifiers fall short in this respect. We here show that this shortcoming can be effectively addressed by using the bidirectional encoder representation from transformers (BERT) proposed by Devlin et al. (2019), which were trained on a next-sentence prediction task, and thus encode a representation of likely next sentences. The BERT-based model outperforms the current state of the art in 11-way classification by 8% points on the standard PDTB dataset. Our experiments also demonstrate that the model can be successfully ported to other domains: on the BioDRB dataset, the model outperforms the state of the art system around 15% points.

We examine the capabilities of a unified, multi-task framework for three information extraction tasks: named entity recognition, relation extraction, and event extraction. Our framework (called DyGIE++) accomplishes all tasks by enumerating, refining, and scoring text spans designed to capture local (within-sentence) and global (cross-sentence) context. Our framework achieves state-of-the-art results across all tasks, on four datasets from a variety of domains. We perform experiments comparing different techniques to construct span representations. Contextualized embeddings like BERT perform well at capturing relationships among entities in the same or adjacent sentences, while dynamic span graph updates model long-range cross-sentence relationships. For instance, propagating span representations via predicted coreference links can enable the model to disambiguate challenging entity mentions. Our code is publicly available at https://github.com/dwadden/dygiepp and can be easily adapted for new tasks or datasets.

Existing event extraction methods classify each argument role independently, ignoring the conceptual correlations between different argument roles. In this paper, we propose a Hierarchical Modular Event Argument Extraction (HMEAE) model, to provide effective inductive bias from the concept hierarchy of event argument roles. Specifically, we design a neural module network for each basic unit of the concept hierarchy, and then hierarchically compose relevant unit modules with logical operations into a role-oriented modular network to classify a specific argument role. As many argument roles share the same high-level unit module, their correlation can be utilized to extract specific event arguments better. Experiments on real-world datasets show that HMEAE can effectively leverage useful knowledge from the concept hierarchy and significantly outperform the state-of-the-art baselines. The source code can be obtained from https://github.com/thunlp/HMEAE.

Scalar implicatures are language features that imply the negation of stronger statements, e.g., “She was married twice” typically implicates that she was not married thrice. In this paper we discuss the importance of scalar implicatures in the context of textual information extraction. We investigate how textual features can be used to predict whether a given text segment mentions all objects standing in a certain relationship with a certain subject. Preliminary results on Wikipedia indicate that this prediction is feasible, and yields informative assessments.