ACL 2019 Papers

The non-indexed parts of the Internet (the Darknet) have become a haven for both legal and illegal anonymous activity. Given the magnitude of these networks, scalably monitoring their activity necessarily relies on automated tools, and notably on NLP tools. However, little is known about what characteristics texts communicated through the Darknet have, and how well do off-the-shelf NLP tools do on this domain. This paper tackles this gap and performs an in-depth investigation of the characteristics of legal and illegal text in the Darknet, comparing it to a clear net website with similar content as a control condition. Taking drugs-related websites as a test case, we find that texts for selling legal and illegal drugs have several linguistic characteristics that distinguish them from one another, as well as from the control condition, among them the distribution of POS tags, and the coverage of their named entities in Wikipedia.

Neural semantic parsers utilize the encoder-decoder framework to learn an end-to-end model for semantic parsing that transduces a natural language sentence to the formal semantic representation. To keep the model aware of the underlying grammar in target sequences, many constrained decoders were devised in a multi-stage paradigm, which decode to the sketches or abstract syntax trees first, and then decode to target semantic tokens. We instead to propose an adaptive decoding method to avoid such intermediate representations. The decoder is guided by model uncertainty and automatically uses deeper computations when necessary. Thus it can predict tokens adaptively. Our model outperforms the state-of-the-art neural models and does not need any expertise like predefined grammar or sketches in the meantime.

We propose a new end-to-end question answering model, which learns to aggregate answer evidence from an incomplete knowledge base (KB) and a set of retrieved text snippets. Under the assumptions that structured data is easier to query and the acquired knowledge can help the understanding of unstructured text, our model first accumulates knowledge ofKB entities from a question-related KB sub-graph; then reformulates the question in the latent space and reads the text with the accumulated entity knowledge at hand. The evidence from KB and text are finally aggregated to predict answers. On the widely-used KBQA benchmark WebQSP, our model achieves consistent improvements across settings with different extents of KB incompleteness.

Multi-hop reading comprehension (RC) questions are challenging because they require reading and reasoning over multiple paragraphs. We argue that it can be difficult to construct large multi-hop RC datasets. For example, even highly compositional questions can be answered with a single hop if they target specific entity types, or the facts needed to answer them are redundant. Our analysis is centered on HotpotQA, where we show that single-hop reasoning can solve much more of the dataset than previously thought. We introduce a single-hop BERT-based RC model that achieves 67 F1—comparable to state-of-the-art multi-hop models. We also design an evaluation setting where humans are not shown all of the necessary paragraphs for the intended multi-hop reasoning but can still answer over 80% of questions. Together with detailed error analysis, these results suggest there should be an increasing focus on the role of evidence in multi-hop reasoning and possibly even a shift towards information retrieval style evaluations with large and diverse evidence collections.

Machine reading comprehension with unanswerable questions is a challenging task. In this work, we propose a data augmentation technique by automatically generating relevant unanswerable questions according to an answerable question paired with its corresponding paragraph that contains the answer. We introduce a pair-to-sequence model for unanswerable question generation, which effectively captures the interactions between the question and the paragraph. We also present a way to construct training data for our question generation models by leveraging the existing reading comprehension dataset. Experimental results show that the pair-to-sequence model performs consistently better compared with the sequence-to-sequence baseline. We further use the automatically generated unanswerable questions as a means of data augmentation on the SQuAD 2.0 dataset, yielding 1.9 absolute F1 improvement with BERT-base model and 1.7 absolute F1 improvement with BERT-large model.

In this paper, we propose a method for why-question answering (why-QA) that uses an adversarial learning framework. Existing why-QA methods retrieve “answer passages” that usually consist of several sentences. These multi-sentence passages contain not only the reason sought by a why-question and its connection to the why-question, but also redundant and/or unrelated parts. We use our proposed “Adversarial networks for Generating compact-answer Representation” (AGR) to generate from a passage a vector representation of the non-redundant reason sought by a why-question and exploit the representation for judging whether the passage actually answers the why-question. Through a series of experiments using Japanese why-QA datasets, we show that these representations improve the performance of our why-QA neural model as well as that of a BERT-based why-QA model. We show that they also improve a state-of-the-art distantly supervised open-domain QA (DS-QA) method on publicly available English datasets, even though the target task is not a why-QA.

Modeling script knowledge can be useful for a wide range of NLP tasks. Current statistical script learning approaches embed the events, such that their relationships are indicated by their similarity in the embedding. While intuitive, these approaches fall short of representing nuanced relations, needed for downstream tasks. In this paper, we suggest to view learning event embedding as a multi-relational problem, which allows us to capture different aspects of event pairs. We model a rich set of event relations, such as Cause and Contrast, derived from the Penn Discourse Tree Bank. We evaluate our model on three types of tasks, the popular Mutli-Choice Narrative Cloze and its variants, several multi-relational prediction tasks, and a related downstream task—implicit discourse sense classification.

When a speaker, Mary, asks “Do you know that Florence is packed with visitors?”, we take her to believe that Florence is packed with visitors, but not if she asks “Do you think that Florence is packed with visitors?”. Inferring speaker commitment (aka event factuality) is crucial for information extraction and question answering. Here, we explore the hypothesis that linguistic deficits drive the error patterns of existing speaker commitment models by analyzing the linguistic correlates of model error on a challenging naturalistic dataset. We evaluate two state-of-the-art speaker commitment models on the CommitmentBank, an English dataset of naturally occurring discourses. The CommitmentBank is annotated with speaker commitment towards the content of the complement (“Florence is packed with visitors” in our example) of clause-embedding verbs (“know”, “think”) under four entailment-canceling environments (negation, modal, question, conditional). A breakdown of items by linguistic features reveals asymmetrical error patterns: while the models achieve good performance on some classes (e.g., negation), they fail to generalize to the diverse linguistic constructions (e.g., conditionals) in natural language, highlighting directions for improvement.

It has been shown that implicit connectives can be exploited to improve the performance of the models for implicit discourse relation recognition (IDRR). An important property of the implicit connectives is that they can be accurately mapped into the discourse relations conveying their functions. In this work, we explore this property in a multi-task learning framework for IDRR in which the relations and the connectives are simultaneously predicted, and the mapping is leveraged to transfer knowledge between the two prediction tasks via the embeddings of relations and connectives. We propose several techniques to enable such knowledge transfer that yield the state-of-the-art performance for IDRR on several settings of the benchmark dataset (i.e., the Penn Discourse Treebank dataset).

We propose an efficient neural framework for sentence-level discourse analysis in accordance with Rhetorical Structure Theory (RST). Our framework comprises a discourse segmenter to identify the elementary discourse units (EDU) in a text, and a discourse parser that constructs a discourse tree in a top-down fashion. Both the segmenter and the parser are based on Pointer Networks and operate in linear time. Our segmenter yields an F1 score of 95.4%, and our parser achieves an F1 score of 81.7% on the aggregated labeled (relation) metric, surpassing previous approaches by a good margin and approaching human agreement on both tasks (98.3 and 83.0 F1).

Recognizing coreferring events and entities across multiple texts is crucial for many NLP applications. Despite the task’s importance, research focus was given mostly to within-document entity coreference, with rather little attention to the other variants. We propose a neural architecture for cross-document coreference resolution. Inspired by Lee et al. (2012), we jointly model entity and event coreference. We represent an event (entity) mention using its lexical span, surrounding context, and relation to entity (event) mentions via predicate-arguments structures. Our model outperforms the previous state-of-the-art event coreference model on ECB+, while providing the first entity coreference results on this corpus. Our analysis confirms that all our representation elements, including the mention span itself, its context, and the relation to other mentions contribute to the model’s success.

The common practice in coreference resolution is to identify and evaluate the maximum span of mentions. The use of maximum spans tangles coreference evaluation with the challenges of mention boundary detection like prepositional phrase attachment. To address this problem, minimum spans are manually annotated in smaller corpora. However, this additional annotation is costly and therefore, this solution does not scale to large corpora. In this paper, we propose the MINA algorithm for automatically extracting minimum spans to benefit from minimum span evaluation in all corpora. We show that the extracted minimum spans by MINA are consistent with those that are manually annotated by experts. Our experiments show that using minimum spans is in particular important in cross-dataset coreference evaluation, in which detected mention boundaries are noisier due to domain shift. We have integrated MINA into https://github.com/ns-moosavi/coval for reporting standard coreference scores based on both maximum and automatically detected minimum spans.

We introduce the Scratchpad Mechanism, a novel addition to the sequence-to-sequence (seq2seq) neural network architecture and demonstrate its effectiveness in improving the overall fluency of seq2seq models for natural language generation tasks. By enabling the decoder at each time step to write to all of the encoder output layers, Scratchpad can employ the encoder as a “scratchpad” memory to keep track of what has been generated so far and thereby guide future generation. We evaluate Scratchpad in the context of three well-studied natural language generation tasks — Machine Translation, Question Generation, and Text Summarization — and obtain state-of-the-art or comparable performance on standard datasets for each task. Qualitative assessments in the form of human judgements (question generation), attention visualization (MT), and sample output (summarization) provide further evidence of the ability of Scratchpad to generate fluent and expressive output.

Text infilling aims at filling in the missing part of a sentence or paragraph, which has been applied to a variety of real-world natural language generation scenarios. Given a well-trained sequential generative model, it is challenging for its unidirectional decoder to generate missing symbols conditioned on the past and future information around the missing part. In this paper, we propose an iterative inference algorithm based on gradient search, which could be the first inference algorithm that can be broadly applied to any neural sequence generative models for text infilling tasks. Extensive experimental comparisons show the effectiveness and efficiency of the proposed method on three different text infilling tasks with various mask ratios and different mask strategies, comparing with five state-of-the-art methods.

Direct comparison on point estimation of the precision (P), recall (R), and F1 measure of two natural language processing (NLP) models on a common test corpus is unreasonable and results in less replicable conclusions due to a lack of a statistical test. However, the existing t-tests in cross-validation (CV) for model comparison are inappropriate because the distributions of P, R, F1 are skewed and an interval estimation of P, R, and F1 based on a t-test may exceed [0,1]. In this study, we propose to use a block-regularized 3×2 CV (3×2 BCV) in model comparison because it could regularize the difference in certain frequency distributions over linguistic units between training and validation sets and yield stable estimators of P, R, and F1. On the basis of the 3×2 BCV, we calibrate the posterior distributions of P, R, and F1 and derive an accurate interval estimation of P, R, and F1. Furthermore, we formulate the comparison into a hypothesis testing problem and propose a novel Bayes test. The test could directly compute the probabilities of the hypotheses on the basis of the posterior distributions and provide more informative decisions than the existing significance t-tests. Three experiments with regard to NLP chunking tasks are conducted, and the results illustrate the validity of the Bayes test.

Learning representations such that the source and target distributions appear as similar as possible has benefited transfer learning tasks across several applications. Generally it requires labeled data from the source and only unlabeled data from the target to learn such representations. While these representations act like a bridge to transfer knowledge learned in the source to the target; they may lead to negative transfer when the source specific characteristics detract their ability to represent the target data. We present a novel neural network architecture to simultaneously learn a two-part representation which is based on the principle of segregating source specific representation from the common representation. The first part captures the source specific characteristics while the second part captures the truly common representation. Our architecture optimizes an objective function which acts adversarial for the source specific part if it contributes towards the cross-domain learning. We empirically show that two parts of the representation, in different arrangements, outperforms existing learning algorithms on the source learning as well as cross-domain tasks on multiple datasets.

Language usage can change across periods of time, but document classifiers models are usually trained and tested on corpora spanning multiple years without considering temporal variations. This paper describes two complementary ways to adapt classifiers to shifts across time. First, we show that diachronic word embeddings, which were originally developed to study language change, can also improve document classification, and we show a simple method for constructing this type of embedding. Second, we propose a time-driven neural classification model inspired by methods for domain adaptation. Experiments on six corpora show how these methods can make classifiers more robust over time.

Existing approaches for learning word embedding often assume there are sufficient occurrences for each word in the corpus, such that the representation of words can be accurately estimated from their contexts. However, in real-world scenarios, out-of-vocabulary (a.k.a. OOV) words that do not appear in training corpus emerge frequently. How to learn accurate representations of these words to augment a pre-trained embedding by only a few observations is a challenging research problem. In this paper, we formulate the learning of OOV embedding as a few-shot regression problem by fitting a representation function to predict an oracle embedding vector (defined as embedding trained with abundant observations) based on limited contexts. Specifically, we propose a novel hierarchical attention network-based embedding framework to serve as the neural regression function, in which the context information of a word is encoded and aggregated from K observations. Furthermore, we propose to use Model-Agnostic Meta-Learning (MAML) for adapting the learned model to the new corpus fast and robustly. Experiments show that the proposed approach significantly outperforms existing methods in constructing an accurate embedding for OOV words and improves downstream tasks when the embedding is utilized.

Heuristic-based active learning (AL) methods are limited when the data distribution of the underlying learning problems vary. Recent data-driven AL policy learning methods are also restricted to learn from closely related domains. We introduce a new sample-efficient method that learns the AL policy directly on the target domain of interest by using wake and dream cycles. Our approach interleaves between querying the annotation of the selected datapoints to update the underlying student learner and improving AL policy using simulation where the current student learner acts as an imperfect annotator. We evaluate our method on cross-domain and cross-lingual text classification and named entity recognition tasks. Experimental results show that our dream-based AL policy training strategy is more effective than applying the pretrained policy without further fine-tuning and better than the existing strong baseline methods that use heuristics or reinforcement learning.