EMNLP 2020 Papers

Representing, and reasoning over, long narratives requires models that can deal with complex event structures connected through multiple relationship types. This paper suggests to represent this type of information as a narrative graph and learn contextualized event representations over it using a relational graph neural network model. We train our model to capture event relations, derived from the Penn Discourse Tree Bank, on a huge corpus, and show that our multi-relational contextualized event representation can improve performance when learning script knowledge without direct supervision and provide a better representation for the implicit discourse sense classification task.

In this paper, we introduce NLP resources for 11 major Indian languages from two major language families. These resources include: (a) large-scale sentence-level monolingual corpora, (b) pre-trained word embeddings, (c) pre-trained language models, and (d) multiple NLU evaluation datasets (IndicGLUE benchmark). The monolingual corpora contains a total of 8.8 billion tokens across all 11 languages and Indian English, primarily sourced from news crawls. The word embeddings are based on FastText, hence suitable for handling morphological complexity of Indian languages. The pre-trained language models are based on the compact ALBERT model. Lastly, we compile the (IndicGLUE benchmark for Indian language NLU. To this end, we create datasets for the following tasks: Article Genre Classification, Headline Prediction, Wikipedia Section-Title Prediction, Cloze-style Multiple choice QA, Winograd NLI and COPA. We also include publicly available datasets for some Indic languages for tasks like Named Entity Recognition, Cross-lingual Sentence Retrieval, Paraphrase detection, etc. Our embeddings are competitive or better than existing pre-trained embeddings on multiple tasks. We hope that the availability of the dataset will accelerate Indic NLP research which has the potential to impact more than a billion people. It can also help the community in evaluating advances in NLP over a more diverse pool of languages. The data and models are available at https://indicnlp.ai4bharat.org.

Continuous efforts have been devoted to language understanding (LU) for conversational queries with the fast and wide-spread popularity of voice assistants. In this paper, we first study the LU problem in the spatial domain, which is a critical problem for providing location-based services by voice assistants but is without in-depth investigation in existing studies. Spatial domain queries have several unique properties making them be more challenging for language understanding than common conversational queries, including lexical-similar but diverse intents and highly ambiguous words. Thus, a special tailored LU framework for spatial domain queries is necessary. To the end, a dataset was extracted and annotated based on the real-life queries from a voice assistant service. We then proposed a new multi-task framework that jointly learns the intent detection and entity linking tasks on the with invented hierarchical intent detection method and triple-scoring mechanism for entity linking. A specially designed spatial GCN is also utilized to model spatial context information among entities. We have conducted extensive experimental evaluations with state-of-the-art entity linking and intent detection methods, which demonstrated that can outperform all baselines with a significant margin.

Natural language understanding (NLU) and Natural language generation (NLG) tasks hold a strong dual relationship, where NLU aims at predicting semantic labels based on natural language utterances and NLG does the opposite. The prior work mainly focused on exploiting the duality in model training in order to obtain the models with better performance. However, regarding the fast-growing scale of models in the current NLP area, sometimes we may have difficulty retraining whole NLU and NLG models. To better address the issue, this paper proposes to leverage the duality in the inference stage without the need of retraining. The experiments on three benchmark datasets demonstrate the effectiveness of the proposed method in both NLU and NLG, providing the great potential of practical usage.

Spoken languages are ever-changing, with new words entering them all the time. However, coming up with new words (neologisms) today relies exclusively on human creativity. In this paper we propose a system to automatically suggest neologisms. We focus on the Hebrew language as a test case due to the unusual regularity of its noun formation. User studies comparing our algorithm to experts and non-experts demonstrate that our algorithm is capable of generating high-quality outputs, as well as enhance human creativity. More broadly, we seek to inspire more computational work around the topic of linguistic creativity, which we believe offers numerous unexplored opportunities.

Active learning for sentence understanding aims at discovering informative unlabeled data for annotation and therefore reducing the demand for labeled data. We argue that the typical uncertainty sampling method for active learning is time-consuming and can hardly work in real-time, which may lead to ineffective sample selection. We propose adversarial uncertainty sampling in discrete space (AUSDS) to retrieve informative unlabeled samples more efficiently. AUSDS maps sentences into latent space generated by the popular pre-trained language models, and discover informative unlabeled text samples for annotation via adversarial attack. The proposed approach is extremely efficient compared with traditional uncertainty sampling with more than 10x speedup. Experimental results on five datasets show that AUSDS outperforms strong baselines on effectiveness.

We study the potential synergy between two different NLP tasks, both confronting predicate lexical variability: identifying predicate paraphrases, and event coreference resolution. First, we used annotations from an event coreference dataset as distant supervision to re-score heuristically-extracted predicate paraphrases. The new scoring gained more than 18 points in average precision upon their ranking by the original scoring method. Then, we used the same re-ranking features as additional inputs to a state-of-the-art event coreference resolution model, which yielded modest but consistent improvements to the model’s performance. The results suggest a promising direction to leverage data and models for each of the tasks to the benefit of the other.

Pretrained Language Models (LMs) have been shown to possess significant linguistic, common sense and factual knowledge. One form of knowledge that has not been studied yet in this context is information about the scalar magnitudes of objects. We show that pretrained language models capture a significant amount of this information but are short of the capability required for general common-sense reasoning. We identify contextual information in pre-training and numeracy as two key factors affecting their performance, and show that a simple method of canonicalizing numbers can have a significant effect on the results.

We present BRIDGE, a powerful sequential architecture for modeling dependencies between natural language questions and relational databases in cross-DB semantic parsing. BRIDGE represents the question and DB schema in a tagged sequence where a subset of the fields are augmented with cell values mentioned in the question. The hybrid sequence is encoded by BERT with minimal subsequent layers and the text-DB contextualization is realized via the fine-tuned deep attention in BERT. Combined with a pointer-generator decoder with schema-consistency driven search space pruning, BRIDGE attained state-of-the-art performance on the well-studied Spider benchmark (65.5% dev, 59.2% test), despite being much simpler than most recently proposed models for this task. Our analysis shows that BRIDGE effectively captures the desired cross-modal dependencies and has the potential to generalize to more text-DB related tasks. Our model implementation is available at https://github.com/salesforce/TabularSemanticParsing.

For decades, chitchat bots are designed as a listener to passively answer what people ask. This passive and relatively simple dialogue mechanism gains less attention from humans and consumes the interests of human beings rapidly. Therefore some recent researches attempt to endow the bots with proactivity through external knowledge to transform the role from a listener to a speaker with a hypothesis that the speaker expresses more just like a knowledge disseminator. However, along with the proactive manner introduced into a dialogue agent, an issue arises that, with too many knowledge facts to express, the agent starts to talks endlessly, and even completely ignores what the other expresses in dialogue sometimes, which greatly harms the interest of the other chatter to continue the conversation. To the end, we propose a novel model named Initiative-Imitate to interact with adaptive initiative throughout a dialogue. It forces the agent to express in parallel with the appropriate role during the whole conversation. The corresponding experiments show the proposed Initiative-Imitate obtains competitive results both on the automatic and manual metrics. And the fluency and engagement of the chatbot have also been improved significantly. Besides, the case study indicates the Initiative-Imitate can constantly transfer to appropriate role timely and response more properly during the whole continuous conversation.

The embedding layers transforming input words into real vectors are the key components of deep neural networks used in natural language processing. However, when the vocabulary is large, the corresponding weight matrices can be enormous, which precludes their deployment in a limited resource setting. We introduce a novel way of parameterizing embedding layers based on the Tensor Train decomposition, which allows compressing the model significantly at the cost of a negligible drop or even a slight gain in performance. We evaluate our method on a wide range of benchmarks in natural language processing and analyze the trade-off between performance and compression ratios for a wide range of architectures, from MLPs to LSTMs and Transformers.

Emotion Recognition in Conversations (ERC) aims to predict the emotional state of speakers in conversations, which is essentially a text classification task. Unlike the sentence-level text classification problem, the available supervised data for the ERC task is limited, which potentially prevents the models from playing their maximum effect. In this paper, we propose a novel approach to leverage unsupervised conversation data, which is more accessible. Specifically, we propose the Conversation Completion (ConvCom) task, which attempts to select the correct answer from candidate answers to fill a masked utterance in a conversation. Then, we Pre-train a basic COntext-Dependent Encoder (Pre-CODE) on the ConvCom task. Finally, we fine-tune the Pre-CODE on the datasets of ERC. Experimental results demonstrate that pre-training on unsupervised data achieves significant improvement of performance on the ERC datasets, particularly on the minority emotion classes.

Out-of-vocabulary (oov) words cause serious troubles in solving natural language tasks with a neural network. Existing approaches to this problem resort to using subwords, which are shorter and more ambiguous units than words, in order to represent oov words with a bag of subwords. In this study, inspired by the processes for creating words from known words, we propose a robust method of estimating oov word embeddings by referring to pre-trained word embeddings for known words with similar surfaces to target oov words. We collect known words by segmenting oov words and by approximate string matching, and we then aggregate their pre-trained embeddings. Experimental results show that the obtained oov word embeddings improve not only word similarity tasks but also downstream tasks in Twitter and biomedical domains where oov words often appear, even when the computed oov embeddings are integrated into a bert-based strong baseline.

The deployment of widely used Transformer architecture is challenging because of heavy computation load and memory overhead during inference, especially when the target device is limited in computational resources such as mobile or edge devices. Quantization is an effective technique to address such challenges. Our analysis shows that for a given number of quantization bits, each block of Transformer contributes to translation quality and inference computations in different manners. Moreover, even inside an embedding block, each word presents vastly different contributions. Correspondingly, we propose a mixed precision quantization strategy to represent Transformer weights by an extremely low number of bits (e.g., under 3 bits). For example, for each word in an embedding block, we assign different quantization bits based on statistical property. Our quantized Transformer model achieves 11.8× smaller model size than the baseline model, with less than -0.5 BLEU. We achieve 8.3× reduction in run-time memory footprints and 3.5× speed up (Galaxy N10+) such that our proposed compression strategy enables efficient implementation for on-device NMT.

There have been significant efforts to interpret the encoder of Transformer-based encoder-decoder architectures for neural machine translation (NMT); meanwhile, the decoder remains largely unexamined despite its critical role. During translation, the decoder must predict output tokens by considering both the source-language text from the encoder and the target-language prefix produced in previous steps. In this work, we study how Transformer-based decoders leverage information from the source and target languages – developing a universal probe task to assess how information is propagated through each module of each decoder layer. We perform extensive experiments on three major translation datasets (WMT En-De, En-Fr, and En-Zh). Our analysis provides insight on when and where decoders leverage different sources. Based on these insights, we demonstrate that the residual feed-forward module in each Transformer decoder layer can be dropped with minimal loss of performance – a significant reduction in computation and number of parameters, and consequently a significant boost to both training and inference speed.

Persuasive dialog systems have various usages, such as donation persuasion and physical exercise persuasion. Previous persuasive dialog systems research mostly focused on analyzing the persuader’s strategies and paid little attention to the persuadee (user). However, understanding and addressing users’ resistance strategies is an essential job of a persuasive dialog system. So, we adopt a preliminary framework on persuasion resistance in psychology and design a fine-grained resistance strategy annotation scheme. We annotate the PersuasionForGood dataset with the scheme. With the enriched annotations, we build a classifier to predict the resistance strategies. Furthermore, we analyze the relationships between persuasion strategies and persuasion resistance strategies. Our work lays the ground for developing a persuasive dialogue system that can understand and address user resistance strategy appropriately. The code and data will be released.

Named Entity Recognition (NER) is deeply explored and widely used in various tasks. Usually, some entity mentions are nested in other entities, which leads to the nested NER problem. Leading region based models face both the efficiency and effectiveness challenge due to the high subsequence enumeration complexity. To tackle these challenges, we propose a hierarchical region learning framework to automatically generate a tree hierarchy of candidate regions with nearly linear complexity and incorporate structure information into the region representation for better classification. Experiments on benchmark datasets ACE-2005, GENIA and JNLPBA demonstrate competitive or better results than state-of-the-art baselines.

The training process of scientific NER models is commonly performed in two steps: i) Pre-training a language model by self-supervised tasks on huge data and ii) fine-tune training with small labelled data. The success of the strategy depends on the relevance between the data domains and between the tasks. However, gaps are found in practice when the target domains are specific and small. We propose a novel framework to introduce a “pre-fine tuning” step between pre-training and fine-tuning. It constructs a corpus by selecting sentences from unlabeled documents that are the most relevant with the labelled training data. Instead of predicting tokens in random spans, the pre-fine tuning task is to predict tokens in entity candidates identified by text mining methods. Pre-fine tuning is automatic and light-weight because the corpus size can be much smaller than pre-training data to achieve a better performance. Experiments on seven benchmarks demonstrate the effectiveness.

We introduce TLDR generation, a new form of extreme summarization, for scientific papers. TLDR generation involves high source compression and requires expert background knowledge and understanding of complex domain-specific language. To facilitate study on this task, we introduce SCITLDR, a new multi-target dataset of 5.4K TLDRs over 3.2K papers. SCITLDR contains both author-written and expert-derived TLDRs, where the latter are collected using a novel annotation protocol that produces high-quality summaries while minimizing annotation burden. We propose CATTS, a simple yet effective learning strategy for generating TLDRs that exploits titles as an auxiliary training signal. CATTS improves upon strong baselines under both automated metrics and human evaluations. Data and code are publicly available at https://github.com/allenai/scitldr.