EMNLP 2020 Papers

Classifying and resolving coreferences of objects (e.g., product names) and attributes (e.g., product aspects) in opinionated reviews is crucial for improving the opinion mining performance. However, the task is challenging as one often needs to consider domain-specific knowledge (e.g., iPad is a tablet and has aspect resolution) to identify coreferences in opinionated reviews. Also, compiling a handcrafted and curated domain-specific knowledge base for each domain is very time consuming and arduous. This paper proposes an approach to automatically mine and leverage domain-specific knowledge for classifying objects and attribute coreferences. The approach extracts domain-specific knowledge from unlabeled review data and trains a knowledgeaware neural coreference classification model to leverage (useful) domain knowledge together with general commonsense knowledge for the task. Experimental evaluation on realworld datasets involving five domains (product types) shows the effectiveness of the approach

Advances in machine reading comprehension (MRC) rely heavily on the collection of large scale human-annotated examples in the form of (question, paragraph, answer) triples. In contrast, humans are typically able to generalize with only a few examples, relying on deeper underlying world knowledge, linguistic sophistication, and/or simply superior deductive powers. In this paper, we focus on “teaching” machines reading comprehension, using a small number of semi-structured explanations that explicitly inform machines why answer spans are correct. We extract structured variables and rules from explanations and compose neural module teachers that annotate instances for training downstream MRC models. We use learnable neural modules and soft logic to handle linguistic variation and overcome sparse coverage; the modules are jointly optimized with the MRC model to improve final performance. On the SQuAD dataset, our proposed method achieves 70.14% F1 score with supervision from 26 explanations, comparable to plain supervised learning using 1,100 labeled instances, yielding a 12x speed up.

Recent neural approaches to data-to-text generation have mostly focused on improving content fidelity while lacking explicit control over writing styles (e.g., sentence structures, word choices). More traditional systems use templates to determine the realization of text. Yet manual or automatic construction of high-quality templates is difficult, and a template acting as hard constraints could harm content fidelity when it does not match the record perfectly. We study a new way of stylistic control by using existing sentences as “soft” templates. That is, a model learns to imitate the writing style of any given exemplar sentence, with automatic adaptions to faithfully describe the record. The problem is challenging due to the lack of parallel data. We develop a neural approach that includes a hybrid attention-copy mechanism, learns with weak supervisions, and is enhanced with a new content coverage constraint. We conduct experiments in restaurants and sports domains. Results show our approach achieves stronger performance than a range of comparison methods. Our approach balances well between content fidelity and style control given exemplars that match the records to varying degrees.

Public works procurements move US$ 10 billion yearly in Brazil and are a preferred field for collusion and fraud. Federal Police and audit agencies investigate collusion (bid-rigging), over-pricing, and delivery fraud in this field and efforts have been employed to early detect fraud and collusion on public works procurements. The current automatic methods of fraud detection use structured data to classification and usually do not involve annotated data. The use of NLP for this kind of application is rare. Our work introduces a new dataset formed by public procurement calls available on Brazilian official journal (Diário Oficial da União), using by 15,132,968 textual entries of which 1,907 are annotated risky entries. Both bottleneck deep neural network and BiLSTM shown competitive compared with classical classifiers and achieved better precision (93.0% and 92.4%, respectively), which signs improvements in a criminal fraud investigation.

Dialogue state tracking (DST) is an important part of a spoken dialogue system. Existing DST models either ignore temporal feature dependencies across dialogue turns or fail to explicitly model temporal state dependencies in a dialogue. In this work, we propose Temporally Expressive Networks (TEN) to jointly model the two types of temporal dependencies in DST. The TEN model utilizes the power of recurrent networks and probabilistic graphical models. Evaluating on standard datasets, TEN is demonstrated to improve the accuracy of turn-level-state prediction and the state aggregation.

Automated Essay Scoring (AES) is a critical text regression task that automatically assigns scores to essays based on their writing quality. Recently, the performance of sentence prediction tasks has been largely improved by using Pre-trained Language Models via fusing representations from different layers, constructing an auxiliary sentence, using multi-task learning, etc. However, to solve the AES task, previous works utilize shallow neural networks to learn essay representations and constrain calculated scores with regression loss or ranking loss, respectively. Since shallow neural networks trained on limited samples show poor performance to capture deep semantic of texts. And without an accurate scoring function, ranking loss and regression loss measures two different aspects of the calculated scores. To improve AES’s performance, we find a new way to fine-tune pre-trained language models with multiple losses of the same task. In this paper, we propose to utilize a pre-trained language model to learn text representations first. With scores calculated from the representations, mean square error loss and the batch-wise ListNet loss with dynamic weights constrain the scores simultaneously. We utilize Quadratic Weighted Kappa to evaluate our model on the Automated Student Assessment Prize dataset. Our model outperforms not only state-of-the-art neural models near 3 percent but also the latest statistic model. Especially on the two narrative prompts, our model performs much better than all other state-of-the-art models.

Generating responses following a desired style has great potentials to extend applications of open-domain dialogue systems, yet is refrained by lacking of parallel data for training. In this work, we explore the challenging task with pre-trained language models that have brought breakthrough to various natural language tasks. To this end, we introduce a KL loss and a style classifier to the fine-tuning step in order to steer response generation towards the target style in both a word-level and a sentence-level. Comprehensive empirical studies with two public datasets indicate that our model can significantly outperform state-of-the-art methods in terms of both style consistency and contextual coherence.

We present CodeBERT, a bimodal pre-trained model for programming language (PL) and natural language (NL). CodeBERT learns general-purpose representations that support downstream NL-PL applications such as natural language code search, code documentation generation, etc. We develop CodeBERT with Transformer-based neural architecture, and train it with a hybrid objective function that incorporates the pre-training task of replaced token detection, which is to detect plausible alternatives sampled from generators. This enables us to utilize both “bimodal” data of NL-PL pairs and “unimodal data, where the former provides input tokens for model training while the latter helps to learn better generators. We evaluate CodeBERT on two NL-PL applications by fine-tuning model parameters. Results show that CodeBERT achieves state-of-the-art performance on both natural language code search and code documentation generation. Furthermore, to investigate what type of knowledge is learned in CodeBERT, we construct a dataset for NL-PL probing, and evaluate in a zero-shot setting where parameters of pre-trained models are fixed. Results show that CodeBERT performs better than previous pre-trained models on NLPL probing.

Natural language processing systems often struggle with out-of-vocabulary (OOV) terms, which do not appear in training data. Blends, such as “innoventor”, are one particularly challenging class of OOV, as they are formed by fusing together two or more bases that relate to the intended meaning in unpredictable manners and degrees. In this work, we run experiments on a novel dataset of English OOV blends to quantify the difficulty of interpreting the meanings of blends by large-scale contextual language models such as BERT. We first show that BERT’s processing of these blends does not fully access the component meanings, leaving their contextual representations semantically impoverished. We find this is mostly due to the loss of characters resulting from blend formation. Then, we assess how easily different models can recognize the structure and recover the origin of blends, and find that context-aware embedding systems outperform character-level and context-free embeddings, although their results are still far from satisfactory.

Psychologists routinely assess people’s emotions and traits, such as their personality, by collecting their responses to survey questionnaires. Such assessments can be costly in terms of both time and money, and often lack generalizability, as existing data cannot be used to predict responses for new survey questions or participants. In this study, we propose a method for predicting a participant’s questionnaire response using their social media texts and the text of the survey question they are asked. Specifically, we use Natural Language Processing (NLP) tools such as BERT embeddings to represent both participants (via the text they write) and survey questions as embeddings vectors, allowing us to predict responses for out-of-sample participants and questions. Our novel approach can be used by researchers to integrate new participants or new questions into psychological studies without the constraint of costly data collection, facilitating novel practical applications and furthering the development of psychological theory. Finally, as a side contribution, the success of our model also suggests a new approach to study survey questions using NLP tools such as text embeddings rather than response data used in traditional methods.

Local sequence transduction (LST) tasks are sequence transduction tasks where there exists massive overlapping between the source and target sequences, such as grammatical error correction and spell or OCR correction. Motivated by this characteristic of LST tasks, we propose Pseudo-Bidirectional Decoding (PBD), a simple but versatile approach for LST tasks. PBD copies the representation of source tokens to the decoder as pseudo future context that enables the decoder self-attention to attends to its bi-directional context. In addition, the bidirectional decoding scheme and the characteristic of LST tasks motivate us to share the encoder and the decoder of LST models. Our approach provides right-side context information for the decoder, reduces the number of parameters by half, and provides good regularization effects. Experimental results on several benchmark datasets show that our approach consistently improves the performance of standard seq2seq models on LST tasks.

Sentence fusion is the task of joining related sentences into coherent text. Current training and evaluation schemes for this task are based on single reference ground-truths and do not account for valid fusion variants. We show that this hinders models from robustly capturing the semantic relationship between input sentences. To alleviate this, we present an approach in which ground-truth solutions are automatically expanded into multiple references via curated equivalence classes of connective phrases. We apply this method to a large-scale dataset and use the augmented dataset for both model training and evaluation. To improve the learning of semantic representation using multiple references, we enrich the model with auxiliary discourse classification tasks under a multi-tasking framework. Our experiments highlight the improvements of our approach over state-of-the-art models.

Domain adaptation of Pretrained Language Models (PTLMs) is typically achieved by unsupervised pretraining on target-domain text. While successful, this approach is expensive in terms of hardware, runtime and CO 2 emissions. Here, we propose a cheaper alternative: We train Word2Vec on target-domain text and align the resulting word vectors with the wordpiece vectors of a general-domain PTLM. We evaluate on eight English biomedical Named Entity Recognition (NER) tasks and compare against the recently proposed BioBERT model. We cover over 60% of the BioBERT - BERT F1 delta, at 5% of BioBERT’s CO 2 footprint and 2% of its cloud compute cost. We also show how to quickly adapt an existing general-domain Question Answering (QA) model to an emerging domain: the Covid-19 pandemic.

The incompleteness of knowledge base (KB) is a vital factor limiting the performance of question answering (QA). This paper proposes a novel QA method by leveraging text information to enhance the incomplete KB. The model enriches the entity representation through semantic information contained in the text, and employs graph convolutional networks to update the entity status. Furthermore, to exploit the latent structural information of text, we treat the text as hyperedges connecting entities among it to complement the deficient relations in KB, and hypergraph convolutional networks are further applied to reason on the hypergraph-formed text. Extensive experiments on the WebQuestionsSP benchmark with different KB settings prove the effectiveness of our model.

Knowing whether a published research result can be replicated is important. Carrying out direct replication of published research incurs a high cost. There are efforts tried to use machine learning aided methods to predict scientific claims’ replicability. However, existing machine learning aided approaches use only hand-extracted statistics features such as p-value, sample size, etc. without utilizing research papers’ text information and train only on a very small size of annotated data without making the most use of a large number of unlabeled articles. Therefore, it is desirable to develop effective machine learning aided automatic methods which can automatically extract text information as features so that we can benefit from Natural Language Processing techniques. Besides, we aim for an approach that benefits from both labeled and the large number of unlabeled data. In this paper, we propose two weakly supervised learning approaches that use automatically extracted text information of research papers to improve the prediction accuracy of research replication using both labeled and unlabeled datasets. Our experiments over real-world datasets show that our approaches obtain much better prediction performance compared to the supervised models utilizing only statistic features and a small size of labeled dataset. Further, we are able to achieve an accuracy of 75.76% for predicting the replicability of research.

Much progress has been made in text summarization, fueled by neural architectures using large-scale training corpora. However, in the news domain, neural models easily overfit by leveraging position-related features due to the prevalence of the inverted pyramid writing style. In addition, there is an unmet need to generate a variety of summaries for different users. In this paper, we propose a neural framework that can flexibly control summary generation by introducing a set of sub-aspect functions (i.e. importance, diversity, position). These sub-aspect functions are regulated by a set of control codes to decide which sub-aspect to focus on during summary generation. We demonstrate that extracted summaries with minimal position bias is comparable with those generated by standard models that take advantage of position preference. We also show that news summaries generated with a focus on diversity can be more preferred by human raters. These results suggest that a more flexible neural summarization framework providing more control options could be desirable in tailoring to different user preferences, which is useful since it is often impractical to articulate such preferences for different applications a priori.

Data balancing is a known technique for improving the performance of classification tasks. In this work we define a novel balancing-viageneration framework termed BalaGen. BalaGen consists of a flexible balancing policy coupled with a text generation mechanism. Combined, these two techniques can be used to augment a dataset for more balanced distribution. We evaluate BalaGen on three publicly available semantic utterance classification (SUC) datasets. One of these is a new COVID-19 Q&A dataset published here for the first time. Our work demonstrates that optimal balancing policies can significantly improve classifier performance, while augmenting just part of the classes and under-sampling others. Furthermore, capitalizing on the advantages of balancing, we show its usefulness in all relevant BalaGen framework components. We validate the superiority of BalaGen on ten semantic utterance datasets taken from real-life goaloriented dialogue systems. Based on our results we encourage using data balancing prior to training for text classification tasks.

We introduce exBERT, a training method to extend BERT pre-trained models from a general domain to a new pre-trained model for a specific domain with a new additive vocabulary under constrained training resources (i.e., constrained computation and data). exBERT uses a small extension module to learn to adapt an augmenting embedding for the new domain in the context of the original BERT’s embedding of a general vocabulary. The exBERT training method is novel in learning the new vocabulary and the extension module while keeping the weights of the original BERT model fixed, resulting in a substantial reduction in required training resources. We pre-train exBERT with biomedical articles from ClinicalKey and PubMed Central, and study its performance on biomedical downstream benchmark tasks using the MTL-Bioinformatics-2016 datasets. We demonstrate that exBERT consistently outperforms prior approaches when using limited corpus and pre-training computation resources.

News recommendation aims to display news articles to users based on their personal interest. Existing news recommendation methods rely on centralized storage of user behavior data for model training, which may lead to privacy concerns and risks due to the privacy-sensitive nature of user behaviors. In this paper, we propose a privacy-preserving method for news recommendation model training based on federated learning, where the user behavior data is locally stored on user devices. Our method can leverage the useful information in the behaviors of massive number users to train accurate news recommendation models and meanwhile remove the need of centralized storage of them. More specifically, on each user device we keep a local copy of the news recommendation model, and compute gradients of the local model based on the user behaviors in this device. The local gradients from a group of randomly selected users are uploaded to server, which are further aggregated to update the global model in the server. Since the model gradients may contain some implicit private information, we apply local differential privacy (LDP) to them before uploading for better privacy protection. The updated global model is then distributed to each user device for local model update. We repeat this process for multiple rounds. Extensive experiments on a real-world dataset show the effectiveness of our method in news recommendation model training with privacy protection.