COLING 2022 论文列表

Models of natural language understanding often rely on question answering and logical inference benchmark challenges to evaluate the performance of a system. While informative, such task-oriented evaluations do not assess the broader semantic abilities that humans have as part of their linguistic competence when speaking and interpreting language. We define competence-based (CB) question generation, and focus on queries over lexical semantic knowledge involving implicit argument and subevent structure of verbs. We present a method to generate such questions and a dataset of English cooking recipes we use for implementing the generation method. Our primary experiment shows that even large pretrained language models perform poorly on CB questions until they are provided with additional contextualized semantic information. The data and the source code is available at: https: //github.com/brandeis-llc/CompQG.

With the rapid growth of scientific papers, understanding the changes and trends in a research area is rather time-consuming. The first challenge is to find related and comparable articles for the research. Comparative citations compare co-cited papers in a citation sentence and can serve as good guidance for researchers to track a research area. We thus go through comparative citations to find comparable objects and build a comparative scientific summarization corpus (CSSC). And then, we propose the comparative graph-based summarization (CGSUM) method to create comparative summaries using citations as guidance. The comparative graph is constructed using sentences as nodes and three different relationships of sentences as edges. The relationship that sentences occur in the same paper is used to calculate the salience of sentences, the relationship that sentences occur in two different papers is used to calculate the difference between sentences, and the relationship that sentences are related to citations is used to calculate the commonality of sentences. Experiments show that CGSUM outperforms comparative baselines on CSSC and performs well on DUC2006 and DUC2007.

As political attitudes have diverged ideologically in the United States, political speech has diverged lingusitically. The ever-widening polarization between the US political parties is accelerated by an erosion of mutual understanding between them. We aim to make these communities more comprehensible to each other with a framework that probes community-specific responses to the same survey questions using community language models CommunityLM. In our framework we identify committed partisan members for each community on Twitter and fine-tune LMs on the tweets authored by them. We then assess the worldviews of the two groups using prompt-based probing of their corresponding LMs, with prompts that elicit opinions about public figures and groups surveyed by the American National Election Studies (ANES) 2020 Exploratory Testing Survey. We compare the responses generated by the LMs to the ANES survey results, and find a level of alignment that greatly exceeds several baseline methods. Our work aims to show that we can use community LMs to query the worldview of any group of people given a sufficiently large sample of their social media discussions or media diet.

We present our novel, hyperparameter-free topic modelling algorithm, Community Topic. Our algorithm is based on mining communities from term co-occurrence networks. We empirically evaluate and compare Community Topic with Latent Dirichlet Allocation and the recently developed top2vec algorithm. We find that Community Topic runs faster than the competitors and produces topics that achieve higher coherence scores. Community Topic can discover coherent topics at various scales. The network representation used by Community Topic results in a natural relationship between topics and a topic hierarchy. This allows sub- and super-topics to be found on demand. These features make Community Topic the ideal tool for downstream applications such as applied research and conversational agents.

Quantization, knowledge distillation, and magnitude pruning are among the most popular methods for neural network compression in NLP. Independently, these methods reduce model size and can accelerate inference, but their relative benefit and combinatorial interactions have not been rigorously studied. For each of the eight possible subsets of these techniques, we compare accuracy vs. model size tradeoffs across six BERT architecture sizes and eight GLUE tasks. We find that quantization and distillation consistently provide greater benefit than pruning. Surprisingly, except for the pair of pruning and quantization, using multiple methods together rarely yields diminishing returns. Instead, we observe complementary and super-multiplicative reductions to model size. Our work quantitatively demonstrates that combining compression methods can synergistically reduce model size, and that practitioners should prioritize (1) quantization, (2) knowledge distillation, and (3) pruning to maximize accuracy vs. model size tradeoffs.

We study the problem of integrating cognitive language processing signals (e.g., eye-tracking or EEG data) into pre-trained language models like BERT. Existing methods typically fine-tune pre-trained models on cognitive data, ignoring the semantic gap between the texts and cognitive signals. To fill the gap, we propose CogBERT, a framework that can induce fine-grained cognitive features from cognitive data and incorporate cognitive features into BERT by adaptively adjusting the weight of cognitive features for different NLP tasks. Extensive experiments show that: (1) Cognition-guided pre-trained models can consistently perform better than basic pre-trained models on ten NLP tasks. (2) Different cognitive features contribute differently to different NLP tasks. Based on this observation, we give a fine-grained explanation of why cognitive data is helpful for NLP. (3) Different transformer layers of pre-trained models should encode different cognitive features, with word-level cognitive features at the bottom and semantic-level cognitive features at the top. (4) Attention visualization demonstrates that CogBERT aligns with human gaze patterns and improves its natural language comprehension ability.

Quotation extraction aims to extract quotations from written text. There are three components in a quotation: source refers to the holder of the quotation, cue is the trigger word(s), and content is the main body. Existing solutions for quotation extraction mainly utilize rule-based approaches and sequence labeling models. While rule-based approaches often lead to low recalls, sequence labeling models cannot well handle quotations with complicated structures. In this paper, we propose the Context and Former-Label Enhanced Net () for quotation extraction. is able to extract complicated quotations with components of variable lengths and complicated structures. On two public datasets (and ) and one proprietary dataset (), we show that our achieves state-of-the-art performance on complicated quotation extraction.

Generalized text representations are the foundation of many natural language understanding tasks. To fully utilize the different corpus, it is inevitable that models need to understand the relevance among them. However, many methods ignore the relevance and adopt a single-channel model (a coarse paradigm) directly for all tasks, which lacks enough rationality and interpretation. In addition, some existing works learn downstream tasks by stitches skill block (a fine paradigm), which might cause irrational results due to its redundancy and noise. In this work, we first analyze the task correlation through three different perspectives, , data property, manual design, and model-based relevance, based on which the similar tasks are grouped together. Then, we propose a hierarchical framework with a coarse-to-fine paradigm, with the bottom level shared to all the tasks, the mid-level divided to different groups, and the top-level assigned to each of the tasks. This allows our model to learn basic language properties from all tasks, boost performance on relevant tasks, and reduce the negative impact from irrelevant tasks. Our experiments on 13 benchmark datasets across five natural language understanding tasks demonstrate the superiority of our method.

Procedural text understanding is a challenging language reasoning task that requires models to track entity states across the development of a narrative. We identify three core aspects required for modeling this task, namely the local and global view of the inputs, as well as the global view of outputs. Prior methods have considered a subset of these aspects, which leads to either low precision or low recall. In this paper, we propose a new model Coalescing Global and Local Information (CGLI), which builds entity- and timestep-aware input representations (local input) considering the whole context (global input), and we jointly model the entity states with a structured prediction objective (global output). Thus, CGLI simultaneously optimizes for both precision and recall. Moreover, we extend CGLI with additional output layers and integrate it into a story reasoning framework. Extensive experiments on a popular procedural text understanding dataset show that our model achieves state-of-the-art results, while experiments on a story reasoning benchmark show the positive impact of our model on downstream reasoning.

We present CoNTACT: a Dutch language model adapted to the domain of COVID-19 tweets. The model was developed by continuing the pre-training phase of RobBERT (Delobelle et al., 2020) by using 2.8M Dutch COVID-19 related tweets posted in 2021. In order to test the performance of the model and compare it to RobBERT, the two models were tested on two tasks: (1) binary vaccine hesitancy detection and (2) detection of arguments for vaccine hesitancy. For both tasks, not only Twitter but also Facebook data was used to show cross-genre performance. In our experiments, CoNTACT showed statistically significant gains over RobBERT in all experiments for task 1. For task 2, we observed substantial improvements in virtually all classes in all experiments. An error analysis indicated that the domain adaptation yielded better representations of domain-specific terminology, causing CoNTACT to make more accurate classification decisions. For task 2, we observed substantial improvements in virtually all classes in all experiments. An error analysis indicated that the domain adaptation yielded better representations of domain-specific terminology, causing CoNTACT to make more accurate classification decisions.

Traditional training paradigms for extractive and abstractive summarization systems always only use token-level or sentence-level training objectives. However, the output summary is always evaluated from summary-level which leads to the inconsistency in training and evaluation. In this paper, we propose a Contrastive Learning based re-ranking framework for one-stage summarization called CoLo. By modeling a contrastive objective, we show that the summarization model is able to directly generate summaries according to the summary-level score without additional modules and parameters. Extensive experiments demonstrate that CoLo boosts the extractive and abstractive results of one-stage systems on CNN/DailyMail benchmark to 44.58 and 46.33 ROUGE-1 score while preserving the parameter efficiency and inference efficiency. Compared with state-of-the-art multi-stage systems, we save more than 100 GPU training hours and obtaining 3x 8x speed-up ratio during inference while maintaining comparable results.

Conversational question generation (CQG) serves as a vital task for machines to assist humans, such as interactive reading comprehension, through conversations. Compared to traditional single-turn question generation (SQG), CQG is more challenging in the sense that the generated question is required not only to be meaningful, but also to align with the provided conversation. Previous studies mainly focus on how to model the flow and alignment of the conversation, but do not thoroughly study which parts of the context and history are necessary for the model. We believe that shortening the context and history is crucial as it can help the model to optimise more on the conversational alignment property. To this end, we propose CoHS-CQG, a two-stage CQG framework, which adopts a novel CoHS module to shorten the context and history of the input. In particular, it selects the top-p sentences and history turns by calculating the relevance scores of them. Our model achieves state-of-the-art performances on CoQA in both the answer-aware and answer-unaware settings.

Cohesion devices, e.g., reiteration, coreference, are crucial for building cohesion links across sentences. In this paper, we propose a document-level neural machine translation framework, CoDoNMT, which models cohesion devices from two perspectives: Cohesion Device Masking (CoDM) and Cohesion Attention Focusing (CoAF). In CoDM, we mask cohesion devices in the current sentence and force NMT to predict them with inter-sentential context information. A prediction task is also introduced to be jointly trained with NMT. In CoAF, we attempt to guide the model to pay exclusive attention to relevant cohesion devices in the context when translating cohesion devices in the current sentence. Such a cohesion attention focusing strategy is softly applied to the self-attention layer. Experiments on three benchmark datasets demonstrate that our approach outperforms state-of-the-art document-level neural machine translation baselines. Further linguistic evaluation validates the effectiveness of the proposed model in producing cohesive translations.

The task of generating texts of different categories has attracted more and more attention in the area of natural language generation recently. Meanwhile, generative adversarial net (GAN) has demonstrated its effectiveness on text generation, and is further applied to category text generation in later works. Different from existing methods, which mainly consider the pairwise relations between the text embedding and the corresponding fixed one-hot class label (data-to-class relations), this paper proposes a novel Contrastive Category Generative Adversarial Net (CoCGAN) to incorporate contrastive learning into adversarial category text generation, considering more flexible data-to-class relations as well as relations between the multiple text embeddings in the same batch (data-to-data relations). The discriminator of CoCGAN discriminates the authenticity of given samples and optimizes a contrastive learning objective to capture both more flexible data-to-class relations and data-to-data relations among training samples. Accordingly, the generator tries to produce more realistic samples which can confuse the discriminator. Experimental results on both synthetic and real category text generation datasets demonstrate that CoCGAN can achieve significant improvements over the baseline category text generation models.

Supervised open relation extraction aims to discover novel relations by leveraging supervised data of pre-defined relations. However, most existing methods do not achieve effective knowledge transfer from pre-defined relations to novel relations, they have difficulties generating high-quality pseudo-labels for unsupervised data of novel relations and usually suffer from the error propagation issue. In this paper, we propose a Cluster-aware Pseudo-Labeling (CaPL) method to improve the pseudo-labels quality and transfer more knowledge for discovering novel relations. Specifically, the model is firstly pre-trained with the pre-defined relations to learn the relation representations. To improve the pseudo-labels quality, the distances between each instance and all cluster centers are used to generate the cluster-aware soft pseudo-labels for novel relations. To mitigate the catastrophic forgetting issue, we design the consistency regularization loss to make better use of the pseudo-labels and jointly train the model with both unsupervised and supervised data. Experimental results on two public datasets demonstrate that our proposed method achieves new state-of-the-arts performance.

Nowadays, transformer-based models gradually become the default choice for artificial intelligence pioneers. The models also show superiority even in the few-shot scenarios. In this paper, we revisit the classical methods and propose a new few-shot alternative. Specifically, we investigate the few-shot one-class problem, which actually takes a known sample as a reference to detect whether an unknown instance belongs to the same class. This problem can be studied from the perspective of sequence match. It is shown that with meta-learning, the classical sequence match method, i.e. Compare-Aggregate, significantly outperforms transformer ones. The classical approach requires much less training cost. Furthermore, we perform an empirical comparison between two kinds of sequence match approaches under simple fine-tuning and meta-learning. Meta-learning causes the transformer models’ features to have high-correlation dimensions. The reason is closely related to the number of layers and heads of transformer models. Experimental codes and data are available at https://github.com/hmt2014/FewOne.

The paper aims to identify cited text spans in the reference paper related to the given citance in the citing paper. We refer to it as cited text span retrieval (CTSR). Most current methods attempt this task by relying on pre-trained off-the-shelf deep learning models like SciBERT. Though these models are pre-trained on large datasets, they under-perform in out-of-domain settings. We introduce CitRet, a novel hybrid model for CTSR that leverages unique semantic and syntactic structural characteristics of scientific documents. This enables us to use significantly less data for finetuning. We use only 1040 documents for finetuning. Our model augments mildly-trained SBERT-based contextual embeddings with pre-trained non-contextual Word2Vec embeddings to calculate semantic textual similarity. We demonstrate the performance of our model on the CLSciSumm shared tasks. It improves the state-of-the-art results by over 15% on the F1 score evaluation.

Chinese couplet generation aims to generate a pair of clauses (usually generating a subsequent clause given an antecedent one) with certain rules (e.g., morphological and syntactical symmetry) adhered and has long been a challenging task with cultural background. To generate high-quality couplet (antecedent) clauses, it normally requires a model to learn the correspondences between antecedent and subsequent clauses under aforementioned rules and constraint of few characters with their concise usage. To tackle this task, previous studies normally directly adopt deep neural networks without explicitly taking into account fine-grained analysis of the clauses, in this paper, we propose to enhance Chinese couplet generation by leveraging syntactic information, i.e., part-of-speech (POS) tags and word dependencies. In doing so, we identify word boundaries in the antecedent clause and then use a special attention module to encode the syntactic information over the words for better generating the subsequent clause. Experimental results on a dataset for Chinese couplet generation illustrate the validity and effectiveness of our approach, which outperforms strong baselines with respect to automatic and manual evaluation metrics.

Compounding, a prevalent word-formation process, presents an interesting challenge for computational models. Indeed, the relations between compounds and their constituents are often complicated. It is particularly so in Chinese morphology, where each character is almost simultaneously bound and free when treated as a morpheme. To model such word-formation process, we propose the Notch (NOnlinear Transformation of CHaracter embeddings) model and the character Jacobians. The Notch model first learns the non-linear relations between the constituents and words, and the character Jacobians further describes the character’s role in each word. In a series of experiments, we show that the Notch model predicts the embeddings of the real words from their constituents but helps account for the behavioral data of the pseudowords. Moreover, we also demonstrated that character Jacobians reflect the characters’ meanings. Taken together, the Notch model and character Jacobians may provide a new perspective on studying the word-formation process and morphology with modern deep learning.