COLING 2022 论文列表

Training Neural Machine Translation (NMT) models suffers from sparse parallel data, in the infrequent translation scenarios towards low-resource source languages. The existing solutions primarily concentrate on the utilization of Parent-Child (PC) transfer learning. It transfers well-trained NMT models on high-resource languages (namely Parent NMT) to low-resource languages, so as to produce Child NMT models by fine-tuning. It has been carefully demonstrated that a variety of PC variants yield significant improvements for low-resource NMT. In this paper, we intend to enhance PC-based NMT by a bidirectionally-adaptive learning strategy. Specifically, we divide inner constituents (6 transformers) of Parent encoder into two “teams”, i.e., T1 and T2. During representation learning, T1 learns to encode low-resource languages conditioned on bilingual shareable latent space. Generative adversarial network and masked language modeling are used for space-shareable encoding. On the other hand, T2 is straightforwardly transferred to low-resource languages, and fine-tuned together with T1 for low-resource translation. Briefly, T1 and T2 take actions separately for different goals. The former aims to adapt to characteristics of low-resource languages during encoding, while the latter adapts to translation experiences learned from high-resource languages. We experiment on benchmark corpora SETIMES, conducting low-resource NMT for Albanian (Sq), Macedonian (Mk), Croatian (Hr) and Romanian (Ro). Experimental results show that our method yields substantial improvements, which allows the NMT performance to reach BLEU4-scores of 62.24%, 56.93%, 50.53% and 54.65% for Sq, Mk, Hr and Ro, respectively.

Various historical languages, which used to be lingua franca of science and arts, deserve the attention of current NLP research. In this work, we take the first data-driven steps towards this research line for Classical Arabic (CA) by addressing named entity recognition (NER) and topic modeling (TM) on the example of CA literature. We manually annotate the encyclopedic work of Tafsir Al-Tabari with span-based NEs, sentence-based topics, and span-based subtopics, thus creating the Tafsir Dataset with over 51,000 sentences, the first large-scale multi-task benchmark for CA. Next, we analyze our newly generated dataset, which we make open-source available, with current language models (lightweight BiLSTM, transformer-based MaChAmP) along a novel script compression method, thereby achieving state-of-the-art performance for our target task CA-NER. We also show that CA-TM from the perspective of historical topic models, which are central to Arabic studies, is very challenging. With this interdisciplinary work, we lay the foundations for future research on automatic analysis of CA literature.

Table-based fact verification aims to verify whether a statement sentence is trusted or fake. Most existing methods rely on graph feature or data augmentation but fail to investigate evidence correlation between the statement and table effectively. In this paper, we propose a self-Labeled Keypoint Alignment model, named LKA, to explore the correlation between the two. Specifically, a dual-view alignment module based on the statement and table views is designed to discriminate the salient words through multiple interactions, where one regular and one adversarial alignment network cooperatively character the alignment discrepancy. Considering the interaction characteristic inherent in the alignment module, we introduce a novel mixture-of experts block to elaborately integrate the interacted information for supporting the alignment and final classification. Furthermore, a contrastive learning loss is utilized to learn the precise representation of the structure-involved words, encouraging the words closer to words with the same table attribute and farther from the words with the unrelated attribute. Experimental results on three widely-studied datasets show that our model can outperform the state-of-the-art baselines and capture interpretable evidence words.

Causal Emotion Entailment (CEE) aims to discover the potential causes behind an emotion in a conversational utterance. Previous works formalize CEE as independent utterance pair classification problems, with emotion and speaker information neglected. From a new perspective, this paper considers CEE in a joint framework. We classify multiple utterances synchronously to capture the correlations between utterances in a global view and propose a Two-Stream Attention Model (TSAM) to effectively model the speaker’s emotional influences in the conversational history. Specifically, the TSAM comprises three modules: Emotion Attention Network (EAN), Speaker Attention Network (SAN), and interaction module. The EAN and SAN incorporate emotion and speaker information in parallel, and the subsequent interaction module effectively interchanges relevant information between the EAN and SAN via a mutual BiAffine transformation. Extensive experimental results demonstrate that our model achieves new State-Of-The-Art (SOTA) performance and outperforms baselines remarkably.

This paper is devoted to the extraction of entities and semantic relations between them from scientific texts, where we consider scientific terms as entities. In this paper, we present a dataset that includes annotations for two tasks and develop a system called TERMinator for the study of the influence of language models on term recognition and comparison of different approaches for relation extraction. Experiments show that language models pre-trained on the target language are not always show the best performance. Also adding some heuristic approaches may improve the overall quality of the particular task. The developed tool and the annotated corpus are publicly available at https://github.com/iis-research-team/terminator and may be useful for other researchers.

Knowledge-grounded dialogue generation consists of two subtasks: knowledge selection and response generation. The knowledge selector generally constructs a query based on the dialogue context and selects the most appropriate knowledge to help response generation. Recent work finds that realizing who (the user or the agent) holds the initiative and utilizing the role-initiative information to instruct the query construction can help select knowledge. It depends on whether the knowledge connection between two adjacent rounds is smooth to assign the role. However, whereby the user takes the initiative only when there is a strong semantic transition between two rounds, probably leading to initiative misjudgment. Therefore, it is necessary to seek a more sensitive reason beyond the initiative role for knowledge selection. To address the above problem, we propose a Topic-shift Aware Knowledge sElector(TAKE). Specifically, we first annotate the topic shift and topic inheritance labels in multi-round dialogues with distant supervision. Then, we alleviate the noise problem in pseudo labels through curriculum learning and knowledge distillation. Extensive experiments on WoW show that TAKE performs better than strong baselines.

In embodied cognition, physical experiences are believed to shape abstract cognition, such as natural language and reasoning. Image schemas were introduced as spatio-temporal cognitive building blocks that capture these recurring sensorimotor experiences. The few existing approaches for automatic detection of image schemas in natural language rely on specific assumptions about word classes as indicators of spatio-temporal events. Furthermore, the lack of sufficiently large, annotated datasets makes evaluation and supervised learning difficult. We propose to build on the recent success of large multilingual pretrained language models and a small dataset of examples from image schema literature to train a supervised classifier that classifies natural language expressions of varying lengths into image schemas. Despite most of the training data being in English with few examples for German, the model performs best in German. Additionally, we analyse the model’s zero-shot performance in Russian, French, and Mandarin. To further investigate the model’s behaviour, we utilize local linear approximations for prediction probabilities that indicate which words in a sentence the model relies on for its final classification decision. Code and dataset are publicly available.

Medical Relation Extraction (MRE) task aims to extract relations between entities in medical texts. Traditional relation extraction methods achieve impressive success by exploring the syntactic information, e.g., dependency tree. However, the quality of the 1-best dependency tree for medical texts produced by an out-of-domain parser is relatively limited so that the performance of medical relation extraction method may degenerate. To this end, we propose a method to jointly model semantic and syntactic information from medical texts based on causal explanation theory. We generate dependency forests consisting of the semantic-embedded 1-best dependency tree. Then, a task-specific causal explainer is adopted to prune the dependency forests, which are further fed into a designed graph convolutional network to learn the corresponding representation for downstream task. Empirically, the various comparisons on benchmark medical datasets demonstrate the effectiveness of our model.

Automatically summarizing patients’ main problems from daily progress notes using natural language processing methods helps to battle against information and cognitive overload in hospital settings and potentially assists providers with computerized diagnostic decision support. Problem list summarization requires a model to understand, abstract, and generate clinical documentation. In this work, we propose a new NLP task that aims to generate a list of problems in a patient’s daily care plan using input from the provider’s progress notes during hospitalization. We investigate the performance of T5 and BART, two state-of-the-art seq2seq transformer architectures, in solving this problem. We provide a corpus built on top of progress notes from publicly available electronic health record progress notes in the Medical Information Mart for Intensive Care (MIMIC)-III. T5 and BART are trained on general domain text, and we experiment with a data augmentation method and a domain adaptation pre-training method to increase exposure to medical vocabulary and knowledge. Evaluation methods include ROUGE, BERTScore, cosine similarity on sentence embedding, and F-score on medical concepts. Results show that T5 with domain adaptive pre-training achieves significant performance gains compared to a rule-based system and general domain pre-trained language models, indicating a promising direction for tackling the problem summarization task.

Abstractive dialogue summarization aims to convert a long dialogue content into its short form where the salient information is preserved while the redundant pieces are ignored. Different from the well-structured text, such as news and scientific articles, dialogues often consist of utterances coming from two or more interlocutors, where the conversations are often informal, verbose, and repetitive, sprinkled with false-starts, backchanneling, reconfirmations, hesitations, speaker interruptions and the salient information is often scattered across the whole chat. The above properties of conversations make it difficult to directly concentrate on scattered outstanding utterances and thus present new challenges of summarizing dialogues. In this work, rather than directly forcing a summarization system to merely pay more attention to the salient pieces, we propose to explicitly have the model perceive the redundant parts of an input dialogue history during the training phase. To be specific, we design two strategies to construct examples without salient pieces as negative cues. Then, the sequence-to-sequence likelihood loss is cooperated with the unlikelihood objective to drive the model to focus less on the unimportant information and also pay more attention to the salient pieces. Extensive experiments on the benchmark dataset demonstrate that our simple method significantly outperforms the baselines with regard to both semantic matching and factual consistent based metrics. The human evaluation also proves the performance gains.

The difficulty of generating coherent long texts lies in the fact that existing models overwhelmingly focus on the tasks of local word prediction, and cannot make high level plans on what to generate or capture the high-level discourse dependencies between chunks of texts. Inspired by how humans write, where a list of bullet points or a catalog is first outlined, and then each bullet point is expanded to form the whole article, we propose SOE, a pipelined system that involves of summarizing, outlining and elaborating for long text generation: the model first outlines the summaries for different segments of long texts, and then elaborates on each bullet point to generate the corresponding segment. To avoid the labor-intensive process of summary soliciting, we propose the reconstruction strategy, which extracts segment summaries in an unsupervised manner by selecting its most informative part to reconstruct the segment. The proposed generation system comes with the following merits: (1) the summary provides high-level guidance for text generation and avoids the local minimum of individual word predictions; (2) the high-level discourse dependencies are captured in the conditional dependencies between summaries and are preserved during the summary expansion process and (3) additionally, we are able to consider significantly more contexts by representing contexts as concise summaries. Extensive experiments demonstrate that SOE produces long texts with significantly better quality, along with faster convergence speed.

Both humans and neural language models are able to perform subject verb number agreement (SVA). In principle, semantics shouldn’t interfere with this task, which only requires syntactic knowledge. In this work we test whether meaning interferes with this type of agreement in English in syntactic structures of various complexities. To do so, we generate both semantically well-formed and nonsensical items. We compare the performance of BERT-base to that of humans, obtained with a psycholinguistic online crowdsourcing experiment. We find that BERT and humans are both sensitive to our semantic manipulation: They fail more often when presented with nonsensical items, especially when their syntactic structure features an attractor (a noun phrase between the subject and the verb that has not the same number as the subject). We also find that the effect of meaningfulness on SVA errors is stronger for BERT than for humans, showing higher lexical sensitivity of the former on this task.

Machine-learning-as-a-service (MLaaS) has attracted millions of users to their splendid large-scale models. Although published as black-box APIs, the valuable models behind these services are still vulnerable to imitation attacks. Recently, a series of works have demonstrated that attackers manage to steal or extract the victim models. Nonetheless, none of the previous stolen models can outperform the original black-box APIs. In this work, we conduct unsupervised domain adaptation and multi-victim ensemble to showing that attackers could potentially surpass victims, which is beyond previous understanding of model extraction. Extensive experiments on both benchmark datasets and real-world APIs validate that the imitators can succeed in outperforming the original black-box models on transferred domains. We consider our work as a milestone in the research of imitation attack, especially on NLP APIs, as the superior performance could influence the defense or even publishing strategy of API providers.

Structural bias has recently been exploited for aspect sentiment triplet extraction (ASTE) and led to improved performance. On the other hand, it is recognized that explicitly incorporating structural bias would have a negative impact on efficiency, whereas pretrained language models (PLMs) can already capture implicit structures. Thus, a natural question arises: Is structural bias still a necessity in the context of PLMs? To answer the question, we propose to address the efficiency issues by using an adapter to integrate structural bias in the PLM and using a cheap-to-compute relative position structure in place of the syntactic dependency structure. Benchmarking evaluation is conducted on the SemEval datasets. The results show that our proposed structural adapter is beneficial to PLMs and achieves state-of-the-art performance over a range of strong baselines, yet with a light parameter demand and low latency. Meanwhile, we give rise to the concern that the current evaluation default with data of small scale is under-confident. Consequently, we release a large-scale dataset for ASTE. The results on the new dataset hint that the structural adapter is confidently effective and efficient to a large scale. Overall, we draw the conclusion that structural bias shall still be a necessity even with PLMs.

Chinese Grammatical Error Diagnosis (CGED) suffers the problems of numerous types of grammatical errors and insufficiency of training data. In this paper, we propose a string editing based CGED model that requires less training data by using a unified workflow to handle various types of grammatical errors. Two measures are proposed in our model to enhance the performance of CGED. First, the detection and correction of grammatical errors are divided into different stages. In the stage of error detection, the model only outputs the types of grammatical errors so that the tag vocabulary size is significantly reduced compared with other string editing based models. Secondly, the correction of some grammatical errors is converted to the task of masked character inference, which has plenty of training data and mature solutions. Experiments on datasets of NLPTEA-CGED demonstrate that our model outperforms other CGED models in many aspects.

This paper analyses the degree to which dialect classifiers based on syntactic representations remain stable over space and time. While previous work has shown that the combination of grammar induction and geospatial text classification produces robust dialect models, we do not know what influence both changing grammars and changing populations have on dialect models. This paper constructs a test set for 12 dialects of English that spans three years at monthly intervals with a fixed spatial distribution across 1,120 cities. Syntactic representations are formulated within the usage-based Construction Grammar paradigm (CxG). The decay rate of classification performance for each dialect over time allows us to identify regions undergoing syntactic change. And the distribution of classification accuracy within dialect regions allows us to identify the degree to which the grammar of a dialect is internally heterogeneous. The main contribution of this paper is to show that a rigorous evaluation of dialect classification models can be used to find both variation over space and change over time.

Despite the revolutionary advances made by Transformer in Neural Machine Translation (NMT), inference efficiency remains an obstacle due to the heavy use of attention operations in auto-regressive decoding. We thereby propose a lightweight attention structure called Attention Refinement Network (ARN) for speeding up Transformer. Specifically, we design a weighted residual network, which reconstructs the attention by reusing the features across layers. To further improve the Transformer efficiency, we merge the self-attention and cross-attention components for parallel computing. Extensive experiments on ten WMT machine translation tasks show that the proposed model yields an average of 1.35x faster (with almost no decrease in BLEU) over the state-of-the-art inference implementation. Results on widely used WMT14 En-De machine translation tasks demonstrate that our model achieves a higher speed-up, giving highly competitive performance compared to AAN and SAN models with fewer parameter numbers.

Discourse parsing on multi-party dialogues is an important but difficult task in dialogue systems and conversational analysis. It is believed that speaker interactions are helpful for this task. However, most previous research ignores speaker interactions between different speakers. To this end, we present a speaker-aware model for this task. Concretely, we propose a speaker-context interaction joint encoding (SCIJE) approach, using the interaction features between different speakers. In addition, we propose a second-stage pre-training task, same speaker prediction (SSP), enhancing the conversational context representations by predicting whether two utterances are from the same speaker. Experiments on two standard benchmark datasets show that the proposed model achieves the best-reported performance in the literature. We will release the codes of this paper to facilitate future research.

We propose a speaker clustering model for textual dialogues, which groups the utterances of a multi-party dialogue without speaker annotations, so that the actual speakers are identical inside each cluster. We find that, without knowing the speakers, the interactions between utterances are still implied in the text, which suggest the relations between speakers. In this work, we model the semantic content of utterance with a pre-trained language model, and the relations between speakers with an utterance-level pairwise matrix. The semantic content representation can be further instructed by cross-corpus dialogue act modeling. The speaker labels are finally generated by spectral clustering. Experiments show that our model outperforms the sequence classification baseline, and benefits from the auxiliary dialogue act classification task. We also discuss the detail of determining the number of speakers (clusters), eliminating the interference caused by semantic similarity, and the impact of utterance distance.