COLING 2024 论文列表

Cross-domain text classification is a crucial task as it enables models to adapt to a target domain that lacks labeled data. It leverages or reuses rich labeled data from the different but related source domain(s) and unlabeled data from the target domain. To this end, previous work focuses on either extracting domain-invariant features or task-agnostic features, ignoring domain-aware features that may be present in the target domain and could be useful for the downstream task. In this paper, we propose a two-stage framework for cross-domain text classification. In the first stage, we finetune the model with mask language modeling (MLM) and labeled data from the source domain. In the second stage, we further fine-tune the model with self-supervised distillation (SSD) and unlabeled data from the target domain. We evaluate its performance on a public cross-domain text classification benchmark and the experiment results show that our method achieves new state-of-the-art results for both single-source domain adaptations (94.17% +1.03%) and multi-source domain adaptations (95.09% +1.34%).

We present the first parallel dataset for English–Tulu translation. Tulu, classified within the South Dravidian linguistic family branch, is predominantly spoken by approximately 2.5 million individuals in southwestern India. Our dataset is constructed by integrating human translations into the multilingual machine translation resource FLORES-200. Furthermore, we use this dataset for evaluation purposes in developing our English–Tulu machine translation model. For the model’s training, we leverage resources available for related South Dravidian languages. We adopt a transfer learning approach that exploits similarities between high-resource and low-resource languages. This method enables the training of a machine translation system even in the absence of parallel data between the source and target language, thereby overcoming a significant obstacle in machine translation development for low-resource languages. Our English–Tulu system, trained without using parallel English–Tulu data, outperforms Google Translate by 19 BLEU points (in September 2023). The dataset and code are available here: https://github.com/manunarayanan/Tulu-NMT.

While deep learning models are powerful, they have limitations in tasks that require commonsense reasoning, as these tasks often involve interpreting information that may not be directly available in the input. Providing evidence has been proven to significantly enhance performance in commonsense reasoning tasks. However, there are various perspectives on evidence, including natural language explanations generated by pre-trained language models, facts derived from world knowledge like text corpora and knowledge bases, and rationales extracted from the input context. Hence, it is crucial to determine how to estimate the confidence degree of different evidence and how to combine them reliably. To address these challenges, this study proposes a trusted multi-view evidential fusion framework for reliable commonsense reasoning tasks that dynamically assesses the confidence of evidence and combines different views of evidence in a trustworthy manner. The proposed method is applied to three commonsense question-answering benchmarks, demonstrating that this approach can effectively reason with multi-view evidence and can compete with state-of-the-art performance.

Asia Minor Greek (AMG) dialects are endangered dialects rich in history and cultAsia Minor Greek (AMG) dialects are endangered dialects rich in history and cultAsia Minor Greek (AMG) dialects are endangered dialects rich in history and cultAsia Minor Greek (AMG) dialects are endangered dialects rich in history and cultAsia Minor Greek (AMG) dialects are endangered dialects rich in history and culture that face a dire struggle for preservation due to declining speaker base and scarce linguistic resources. To address this need, we introduce a Universal Dependencies treebank of Pharasiot Greek, one of the severly endangerd AMG dialects. The present treebank is fully manually annotated and currently consists of 350 sentences from six fairy tales in Pharasiot dialect. Besides describing the treebank and the annotation process, we provide and discuss interesting phenomena we observed in the treebank. Most phenomena we discuss are related to contact-induced linguistic changes that these dialects are well known for. Beyond linguistic inquiry, like other treebanks for truly low-resource languages, the AMG treebank we present offers potentials for diverse applications, such as language preservation and revitalization, as well as NLP tools that have to be developed with scarce resources.

Comparing annotations is a constant and necessary step in corpus analysis. Although the nature of these annotations is normally research-specific, the tools used for this purpose do not have to be. Here, we present a tool for extracting and comparing annotations from ELAN, despite their idiosyncrasies. The intention behind this tool is to provide a handy way to analyze ELAN annotated files, by comparing tiers to a reference unit. Using the presented tool, it is possible to see how tiers overlap (even if they are of symbolic type), to which ratio, and the displacement regarding a reference unit. We present an example of multimodal corpus analysis, regarding the coordination between speech and gesture units based on a pragmatic reference. We argue that looking into overlap ratios can be more informative of the association between speech and gestures, and that considering a time buffer between speech and gestural events can be misleading.

Natural language processing for programming aims to use NLP techniques to assist programming. It is increasingly prevalent for its effectiveness in improving productivity. Distinct from natural language, a programming language is highly structured and functional. Constructing a structure-based representation and a functionality-oriented algorithm is at the heart of program understanding and generation. In this paper, we conduct a systematic review covering tasks, datasets, evaluation methods, techniques, and models from the perspective of the structure-based and functionality-oriented property, aiming to understand the role of the two properties in each component. Based on the analysis, we illustrate unexplored areas and suggest potential directions for future work.

This study examines whether the attention scores between tokens in the BERT model significantly vary based on lexical categories during the fine-tuning process for downstream tasks. Drawing inspiration from the notion that in human language processing, syntactic and semantic information is parsed differently, we categorize tokens in sentences according to their lexical categories and focus on changes in attention scores among these categories. Our hypothesis posits that in downstream tasks that prioritize semantic information, attention scores centered on content words are enhanced, while in cases emphasizing syntactic information, attention scores centered on function words are intensified. Through experimentation conducted on six tasks from the GLUE benchmark dataset, we substantiate our hypothesis regarding the fine-tuning process. Furthermore, our additional investigations reveal the presence of BERT layers that consistently assign more bias to specific lexical categories, irrespective of the task, highlighting the existence of task-agnostic lexical category preferences.

Few-shot named entity recognition (NER) is a challenging task that aims to recognize new named entities with only a limited amount of labeled examples. In this paper, we introduce SSF, which is a streamlined span-based factorization method that addresses the problem of few-shot NER. Our approach formulates few-shot NER as a span-level alignment problem between query and support instances. To achieve this goal, SSF decomposes the span-level alignment problem into several refined span-level procedures. The proposed approach encompasses several key modules such as the Span Boosting Module, Span Prototypical Module, Span Alignment Module, and Span Optimization Module. Our experimental results demonstrate a significant improvement over the previous state-of-the-art performance. Specifically, compared to previous methods, our proposed approach achieves an average F1 score improvement of 12 points on the FewNERD dataset and 10 points on the SNIPS dataset. Moreover, our approach has surpassed the latest state-of-the-art performance on both datasets.

Low-Rank Adaptation (LoRA) is a widely used Parameter-Efficient Fine-Tuning (PEFT) method that updates an initial weight matrix W0 with a delta matrix 𝛥 W consisted by two low-rank matrices A and B. A previous study suggested that there is correlation between W0 and 𝛥 W. In this study, we aim to delve deeper into relationships between W0 and low-rank matrices A and B to further comprehend the behavior of LoRA. In particular, we analyze a conversion matrix that transform W0 into low-rank matrices, which encapsulates information about the relationships. Our analysis reveals that the conversion matrices are similar across each layer. Inspired by these findings, we hypothesize that a single linear layer, which takes each layer’s W0 as input, can yield task-adapted low-rank matrices. To confirm this hypothesis, we devise a method named Conditionally Parameterized LoRA (CondLoRA) that updates initial weight matrices with low-rank matrices derived from a single linear layer. Our empirical results show that CondLoRA maintains a performance on par with LoRA, despite the fact that the trainable parameters of CondLoRA are fewer than those of LoRA. Therefore, we conclude that “a single linear layer yields task-adapted low-rank matrices.” The code used in our experiments is available at https://github.com/CyberAgentAILab/CondLoRA.

Document-level Event Argument Extraction (DEAE) aims to identify arguments and their specific roles from an unstructured document. The advanced approaches on DEAE utilize prompt-based methods to guide pre-trained language models (PLMs) in extracting arguments from input documents. They mainly concentrate on establishing relations between triggers and entity mentions within documents, leaving two unresolved problems: a) independent modeling of entity mentions; b) document-prompt isolation. To this end, we propose a semantic mention Graph Augmented Model (GAM) to address these two problems in this paper. Firstly, GAM constructs a semantic mention graph that captures relations within and between documents and prompts, encompassing co-existence, co-reference and co-type relations. Furthermore, we introduce an ensemble graph transformer module to address mentions and their three semantic relations effectively. Later, the graph-augmented encoder-decoder module incorporates the relation-specific graph into the input embedding of PLMs and optimizes the encoder section with topology information, enhancing the relations comprehensively. Extensive experiments on the RAMS and WikiEvents datasets demonstrate the effectiveness of our approach, surpassing baseline methods and achieving a new state-of-the-art performance.

Simile tasks are challenging in natural language processing (NLP) because models require adequate world knowledge to produce predictions. In recent years, pre-trained language models (PLMs) have succeeded in NLP since they learn generic knowledge from a large corpus. The knowledge embedded in PLMs can be used for different kinds of Simile tasks. However, previous work usually explored one type of simile knowledge for a specific simile task, how to fully utilize different types of knowledge embedded in the PLMs requires further exploration. This paper proposes a self-verified method for exploring simile knowledge from PLMs, which allows the PLMs to leverage one type of simile knowledge to self-validate another. To this end, we first enhance PLMs with a novel multi-level simile recognition (MLSR) task that trains PLMs to evaluate the quality of similes. Then the PLMs leverage this evaluation score to assist the simile interpretation and generation tasks. In this way, we connect different types of simile knowledge in PLMs and make better use of them. Experiments on different pre-trained models and multiple publicly available datasets show that our method works for different kinds of PLMs and can explore more accurate simile knowledge for PLMs. Our code/data will be released on GitHub.

Due to the lack of parallel data, the mainstream fine-tuning-based domain adaptation methods have the overfitting problem in the translation of low-resource domains, and it is difficult for the model to learn the in-domain generalization knowledge. To address the above issue, in this work, we propose a novel Reinforcement Learning Domain Adaptation method for Neural Machine Translation (RLDA-NMT) in the low-resource domain. RLDA-NMT utilizes in-domain source monolingual data to make up for the lack of parallel data, and reinforces domain features learning to make the translation model learn the domain-specific knowledge more fully. Specifically, we first train a ranking-based model with a small-scale in-domain parallel corpus, and then adopt it as the reward model to select higher-quality generated translations for reinforcement when fine-tuning pre-trained NMT model using in-domain source monolingual data. We conduct experiments on Education, Laws, Thesis, and Patent domains of Chinese⇔English translation tasks. Experimental results demonstrate that RLDA-NMT can alleviate overfitting and reinforce the NMT model to learn domain-specific knowledge. Additionally, the results also show that RLDA-NMT and back-translation (BT) are nicely complementary to each other, where combining RLDA-NMT with BT can further improve translation quality.

Information extraction (IE) aims to extract complex structured information from the text. Numerous datasets have been constructed for various IE tasks, leading to time-consuming and labor-intensive data annotations. Nevertheless, most prevailing methods focus on training task-specific models, while the common knowledge among different IE tasks is not explicitly modeled. Moreover, the same phrase may have inconsistent labels in different tasks, which poses a big challenge for knowledge transfer using a unified model. In this study, we propose a regularization-based transfer learning method for IE (TIE) via an instructed graph decoder. Specifically, we first construct an instruction pool for datasets from all well-known IE tasks, and then present an instructed graph decoder, which decodes various complex structures into a graph uniformly based on corresponding instructions. In this way, the common knowledge shared with existing datasets can be learned and transferred to a new dataset with new labels. Furthermore, to alleviate the label inconsistency problem among various IE tasks, we introduce a task-specific regularization strategy, which does not update the gradients of two tasks with ‘opposite direction’. We conduct extensive experiments on 12 datasets spanning four IE tasks, and the results demonstrate the great advantages of our proposed method.

Enabling machines with the capability to recognize and comprehend metaphors is a crucial step toward achieving artificial intelligence. In linguistic theories, metaphor can be identified through Metaphor Identification Procedure (MIP) or Selectional Preference Violation (SPV), both of which are typically considered as matching tasks in the field of natural language processing. However, the implementation of MIP poses a challenge due to the semantic uncertainty and ambiguity of literal meanings of words. Simultaneously, SPV often struggles to recognize conventional metaphors. Inspired by Quantum Language Model (QLM) for modeling semantic uncertainty and fine-grained feature matching, we propose a quantum-inspired matching network for metaphor detection. Specifically, we use the density matrix to explicitly characterize the literal meanings of the target word for MIP, in order to model the uncertainty and ambiguity of the literal meanings of words. This can make SPV effective even in the face of conventional metaphors. MIP and SPV are then achieved by fine-grained feature matching. The results of the experiment finally demonstrated our approach has strong competitiveness.

The process of adapting and creating Easy-to-Read (E2R) texts is very expensive and time-consuming. Due to the success of Large Language Models (LLMs) such as ChatGPT and their ability to generate written language, it is likely to think that such models can help in the adaptation or creation of text in E2R. In this paper, we explore the concept of E2R, its underlying principles and applications, and provides a preliminary study on the usefulness of ChatGPT-4 for E2R text adaptation. We focus on the Spanish language and its E2R variant, Lectura Fácil (LF). We consider a range of prompts that can be used and the differences in output that this produces. We then carry out a three-folded evaluation on 10 texts adapted by ChatGPT-4: (1) an automated evaluation to check values related to the readability of texts, (2) a checklist-based manual evaluation (for which we also propose three new capabilities) and (3) a users’ evaluation with people with cognitive disabilities. We show that it is difficult to choose the best prompt to make ChatGPT-4 adapt texts to LF. Furthermore, the generated output does not follow the E2R text rules, so it is often not suitable for the target audience.

While Natural Language Processing (NLP) models have gained substantial attention, only in recent years has research opened new paths for tackling Human-Computer Design (HCD) from the perspective of natural language. We focus on developing a human-centered corpus, more specifically, a persona-based corpus in a particular healthcare domain (diabetes mellitus self-care). In order to follow an HCD approach, we created personas to model interpersonal interaction (expert and non-expert users) in that specific domain. We show that an HCD approach benefits language generation from different perspectives, from machines to humans - contributing with new directions for low-resource contexts (languages other than English and sensitive domains) where the need to promote effective communication is essential.

Machine Translation (MT) has greatly advanced over the years due to the developments in deep neural networks. However, the emergence of Large Language Models (LLMs) like GPT-4 and ChatGPT is introducing a new phase in the MT domain. In this context, we believe that the future of MT is intricately tied to the capabilities of LLMs. These models not only offer vast linguistic understandings but also bring innovative methodologies, such as prompt-based techniques, that have the potential to further elevate MT. In this paper, we provide an overview of the significant enhancements in MT that are influenced by LLMs and advocate for their pivotal role in upcoming MT research and implementations. We highlight several new MT directions, emphasizing the benefits of LLMs in scenarios such as Long-Document Translation, Stylized Translation, and Interactive Translation. Additionally, we address the important concern of privacy in LLM-driven MT and suggest essential privacy-preserving strategies. By showcasing practical instances, we aim to demonstrate the advantages that LLMs offer, particularly in tasks like translating extended documents. We conclude by emphasizing the critical role of LLMs in guiding the future evolution of MT and offer a roadmap for future exploration in the sector.

Different from most existing tasks relying on abundant labeled data, Few-shot Named Entity Recognition (NER) aims to develop NER systems that are capable of learning from a small set of labeled samples and then generalizing well to new, unseen data.In this paper, with the intention of obtaining a model that can better adapt to new domains, we design a novel three-stage framework for Few-shot NER, including teacher span recognizer, student span recognizer and entity classifier.We first train a teacher span recognizer which is based on a global boundary matrix to obtain soft boundary labels.Then we leverage the soft boundary labels learned by the teacher model to assist in training the student span recognizer,which can smooth the training process of span recognizer.Finally, we adopt the traditional prototypical network as entity classifier and incorporate the idea of prompt learning to construct a more generalizable semantic space.Extensive experiments on various benchmarks demonstrate that our approach surpasses prior methods.

Medical imaging is critical to the diagnosis, surveillance, and treatment of many health conditions, including oncological, neurological, cardiovascular, and musculoskeletal disorders, among others. Radiologists interpret these complex, unstructured images and articulate their assessments through narrative reports that remain largely unstructured. This unstructured narrative must be converted into a structured semantic representation to facilitate secondary applications such as retrospective analyses or clinical decision support. Here, we introduce the Corpus of Annotated Medical Imaging Reports (CAMIR), which includes 609 annotated radiology reports from three imaging modality types: Computed Tomography, Magnetic Resonance Imaging, and Positron Emission Tomography-Computed Tomography. Reports were annotated using an event-based schema that captures clinical indications, lesions, and medical problems. Each event consists of a trigger and multiple arguments, and a majority of the argument types, including anatomy, normalize the spans to pre-defined concepts to facilitate secondary use. CAMIR uniquely combines a granular event structure and concept normalization. To extract CAMIR events, we explored two BERT (Bi-directional Encoder Representation from Transformers)-based architectures, including an existing architecture (mSpERT) that jointly extracts all event information and a multi-step approach (PL-Marker++) that we augmented for the CAMIR schema.