COLING 2024 论文列表

Large Language Models (LLMs) have ushered in a new era in Natural Language Processing, but their massive size demands effective compression techniques for practicality. Although numerous model compression techniques have been investigated, they typically rely on a calibration set that overlooks the multilingual context and results in significant accuracy degradation for low-resource languages. This paper introduces Multilingual Brain Surgeon (MBS), a novel calibration data sampling method for multilingual LLMs compression. MBS overcomes the English-centric limitations of existing methods by sampling calibration data from various languages proportionally to the language distribution of the model training datasets. Our experiments, conducted on the BLOOM multilingual LLM, demonstrate that MBS improves the performance of existing English-centric compression methods, especially for low-resource languages. We also uncover the dynamics of language interaction during compression, revealing that the larger the proportion of a language in the training set and the more similar the language is to the calibration language, the better performance the language retains after compression. In conclusion, MBS presents an innovative approach to compressing multilingual LLMs, addressing the performance disparities and improving the language inclusivity of existing compression techniques. Keywords: Large Language Model, Multilingual Model Compression

Understanding the relation between the meanings of words is an important part of comprehending natural language. Prior work has either focused on analysing lexical semantic relations in word embeddings or probing pretrained language models (PLMs), with some exceptions. Given the rarity of highly multilingual benchmarks, it is unclear to what extent PLMs capture relational knowledge and are able to transfer it across languages. To start addressing this question, we propose MultiLexBATS, a multilingual parallel dataset of lexical semantic relations adapted from BATS in 15 languages including low-resource languages, such as Bambara, Lithuanian, and Albanian. As experiment on cross-lingual transfer of relational knowledge, we test the PLMs’ ability to (1) capture analogies across languages, and (2) predict translation targets. We find considerable differences across relation types and languages with a clear preference for hypernymy and antonymy as well as romance languages.

Text sanitization is the task of detecting and removing personal information from the text. While it has been well-studied in monolingual settings, today, there is also a need for multilingual text sanitization. In this paper, we introduce MultiLeg: a parallel, multilingual named entity (NE) dataset consisting of documents from the Court of Justice of the European Union annotated with semantic categories suitable for text sanitization. The dataset is available in 8 languages, and it contains 3082 parallel text segments for each language. We also show that the pseudonymized dataset remains useful for downstream tasks.

Emotional support conversation (ESC) task aims to relieve the emotional distress of users who have high-intensity of negative emotions. However, due to the ignorance of emotion intensity modelling which is essential for ESC, previous methods fail to capture the transition of emotion intensity effectively. To this end, we propose a Multi-stream information Fusion Framework (MFF-ESC) to thoroughly fuse three streams (text semantics stream, emotion intensity stream, and feedback stream) for the modelling of emotion intensity, based on a designed multi-stream fusion unit. As the difficulty of modelling subtle transitions of emotion intensity and the strong emotion intensity-feedback correlations, we use the KL divergence between feedback distribution and emotion intensity distribution to further guide the learning of emotion intensities. Experimental results on automatic and human evaluations indicate the effectiveness of our method.

Knowledge graph completion (KGC) is a widely used method to tackle incompleteness in knowledge graphs (KGs) by making predictions for missing links. Description-based KGC leverages pre-trained language models to learn entity and relation representations with their names or descriptions, which shows promising results. However, the performance of description-based KGC is still limited by the quality of text and the incomplete structure, as it lacks sufficient entity descriptions and relies solely on relation names, leading to sub-optimal results. To address this issue, we propose MPIKGC, a general framework to compensate for the deficiency of contextualized knowledge and improve KGC by querying large language models (LLMs) from various perspectives, which involves leveraging the reasoning, explanation, and summarization capabilities of LLMs to expand entity descriptions, understand relations, and extract structures, respectively. We conducted extensive evaluation of the effectiveness and improvement of our framework based on four description-based KGC models, for both link prediction and triplet classification tasks. All codes and generated data will be publicly available after review.

Multi-modal semantic understanding requires integrating information from different modalities to extract users’ real intention behind words. Most previous work applies a dual-encoder structure to separately encode image and text, but fails to learn cross-modal feature alignment, making it hard to achieve cross-modal deep information interaction. This paper proposes a novel CLIP-guided contrastive-learning-based architecture to perform multi-modal feature alignment, which projects the features derived from different modalities into a unified deep space. On multi-modal sarcasm detection (MMSD) and multi-modal sentiment analysis (MMSA) tasks, the experimental results show that our proposed model significantly outperforms several baselines, and our feature alignment strategy brings obvious performance gain over models with different aggregating methods and models even enriched with knowledge. More importantly, our model is simple to implement without using task-specific external knowledge, and thus can easily migrate to other multi-modal tasks. Our source codes are available at https://github.com/ChangKe123/CLFA.

Social networks have become venues where people can share and spread hate speech, especially when the platforms allow users to remain anonymous. Hate speech can have significant social and cultural effects, especially when it targets specific groups of people in terms of religion, race, ethnicity, culture or a specific social situation such as immigrants and refugees. In this study, we propose a hate speech detection model, BERTurk-DualCL, using a mixed objective with contrastive learning loss that is combined with the traditional cross-entropy loss used for classification. In addition, we study the effects of paralinguistic features, namely emojis and hashtags, on the performance of our model. We trained and evaluated our model on tweets in four different topics with heated discussions from two separate datasets, ranging from discussions about migrants to the Israel-Palestine conflict. Our multi-domain model outperforms comparable results in literature and the average results of four domain-specific models, achieving a macro-F1 score of 81.04% and 58.89% on two- and five-class tasks respectively.

Word segmentation for Chinese text data is essential for compiling corpora and any other tasks where the notion of “word” is assumed, since Chinese orthography does not have conventional word boundaries as languages such as English do. A perennial issue, however, is that there is no consensus about the definition of “word” in Chinese, which makes word segmentation challenging. Recent work in Chinese word segmentation has begun to embrace the idea of multiple word segmentation possibilities. In a similar spirit, this paper focuses on Cantonese, another major Chinese variety. We propose a linguistically motivated, multi-tiered word segmentation system for Cantonese, and release a Cantonese corpus of 150,000 characters word-segmented by this proposal. Our work will be of interest to researchers whose work involves Cantonese corpus data.

Recent advances in machine learning have demonstrated that multi-modal pre-training can improve automatic speech recognition (ASR) performance compared to randomly initialized models, even when models are fine-tuned on uni-modal tasks. Existing multi-modal pre-training methods for the ASR task have primarily focused on single-stage pre-training where a single unsupervised task is used for pre-training followed by fine-tuning on the downstream task. In this work, we introduce a novel method combining multi-modal and multi-task unsupervised pre-training with a translation-based supervised mid-training approach. We empirically demonstrate that such a multi-stage approach leads to relative word error rate (WER) improvements of up to 38.45% over baselines on both Librispeech and SUPERB. Additionally, we share several important findings for choosing pre-training methods and datasets.

Product review summarization aims to generate a concise summary based on product reviews to facilitate purchasing decisions. This intricate task gives rise to three challenges in existing work: factual accuracy, aspect comprehensiveness, and content relevance. In this paper, we first propose an FB-Thinker framework to improve the summarization ability of LLMs with multi-objective forward reasoning and multi-reward backward refinement. To enable LLM with these dual capabilities, we present two Chinese product review summarization datasets, Product-CSum and Product-CSum-Cross, for both instruction-tuning and cross-domain evaluation. Specifically, these datasets are collected via GPT-assisted manual annotations from an online forum and public datasets. We further design an evaluation mechanism Product-Eval, integrating both automatic and human evaluation across multiple dimensions for product summarization. Experimental results show the competitiveness and generalizability of our proposed framework in the product review summarization tasks.

Text semantic matching is crucial in natural language processing, applied in information retrieval, question answering, and recommendation systems. Traditional text-matching methods struggle with semantic nuances in short text. Recent advancements in multi-granularity representation learning have led to increased interest in improving text semantic matching models. We propose a novel multi-granularity fusion model that harnesses WoBERT, a pre-trained language model, to enhance the accuracy of text semantic information capture. Initially, we process text using WoBERT to acquire semantic representations, effectively capturing individual text semantic nuances. Next, we employ a soft attention alignment mechanism, enabling multi-granularity fusions among characters, words, and sentences, thus further improving matching performance. Our approach was evaluated through experiments on common Chinese short text matching datasets, BQ and LCQMC. Results reveal a significant improvement in performance compared to traditional methods, particularly in terms of accuracy.

Retrieval-based dialogue agents aim at selecting a proper response according to multi-turn conversational history. Existing methods have achieved great progress in terms of retrieval accuracy on benchmarks with pre-trained language models. However, these methods simply concatenate all turns in the dialogue history as the input, ignoring the dialogue dependency and structural information between the utterances. Besides, they usually reason the relationship of the context-response pair at a single level of abstraction (e.g., utterance level), which can not comprehensively capture the fine-grained relation between the context and response. In this paper, we present the multi-grained conversational graph network (MCGN) that considers multiple levels of abstraction from dialogue histories and semantic dependencies within multi-turn dialogues for addressing. Evaluation results on two benchmarks indicate that the proposed multi-grained conversational graph network is helpful for dialogue context understanding and can bring consistent and significant improvement over the state-of-the-art methods.

Almost all frameworks for the manual or automatic evaluation of machine translation characterize the quality of an MT output with a single number. An exception is the Multidimensional Quality Metrics (MQM) framework which offers a fine-grained ontology of quality dimensions for scoring (such as style, fluency, accuracy, and terminology). Previous studies have demonstrated the feasibility of MQM annotation but there are, to our knowledge, no computational models that predict MQM scores for novel texts, due to a lack of resources. In this paper, we address these shortcomings by (a) providing a 1200-sentence MQM evaluation benchmark for the language pair English-Korean and (b) reframing MT evaluation as the multi-task problem of simultaneously predicting several MQM scores using SOTA language models, both in a reference-based MT evaluation setup and a reference-free quality estimation (QE) setup. We find that reference-free setup outperforms its counterpart in the style dimension while reference-based models retain an edge regarding accuracy. Overall, RemBERT emerges as the most promising model. Through our evaluation, we offer an insight into the translation quality in a more fine-grained, interpretable manner.

The task of Sign Language Production (SLP) in machine learning involves converting text-based spoken language into corresponding sign language expressions. Sign language conveys meaning through the continuous movement of multiple articulators, including manual and non-manual channels. However, most current Transformer-based SLP models convert these multi-channel sign poses into a unified feature representation, ignoring the inherent structural correlations between channels. This paper introduces a novel approach called MCST-Transformer for skeletal sign language production. It employs multi-channel spatial attention to capture correlations across various channels within each frame, and temporal attention to learn sequential dependencies for each channel over time. Additionally, the paper explores and experiments with multiple fusion techniques to combine the spatial and temporal representations into naturalistic sign sequences. To validate the effectiveness of the proposed MCST-Transformer model and its constituent components, extensive experiments were conducted on two benchmark sign language datasets from diverse cultures. The results demonstrate that this new approach outperforms state-of-the-art models on both datasets.

Motivational interviewing (MI) is a counseling approach that aims to increase intrinsic motivation and commitment to change. Despite its effectiveness in various disorders such as addiction, weight loss, and smoking cessation, publicly available annotated MI datasets are scarce, limiting the development and evaluation of MI language generation models. We present MI-TAGS, a new annotated dataset of MI therapy sessions written in English collected from video recordings available on public sources. The dataset includes 242 MI demonstration transcripts annotated with the MI Treatment Integrity (MITI) 4.2 therapist behavioral codes and global scores, and Client Language EAsy Rating (CLEAR) 1.0 tags for client speech. In this paper we describe the process of data collection, transcription, and annotation, and provide an analysis of the new dataset. Additionally, we explore the potential use of the dataset for training language models to perform several MITI classification tasks; our results suggest that models may be able to automatically provide utterance-level annotation as well as global scores, with performance comparable to human annotators.

The task of **text2motion** is to generate human motion sequences from given textual descriptions, where the model explores diverse mappings from natural language instructions to human body movements. While most existing works are confined to coarse-grained motion descriptions, e.g., _”A man squats.”_, fine-grained descriptions specifying movements of relevant body parts are barely explored. Models trained with coarse-grained texts may not be able to learn mappings from fine-grained motion-related words to motion primitives, resulting in the failure to generate motions from unseen descriptions. In this paper, we build a large-scale language-motion dataset specializing in fine-grained textual descriptions, FineHumanML3D, by feeding GPT-3.5-turbo with step-by-step instructions with pseudo-code compulsory checks. Accordingly, we design a new text2motion model, FineMotionDiffuse, making full use of fine-grained textual information. Our quantitative evaluation shows that FineMotionDiffuse trained on FineHumanML3D improves FID by a large margin of 0.38, compared with competitive baselines. According to the qualitative evaluation and case study, our model outperforms MotionDiffuse in generating spatially or chronologically composite motions, by learning the implicit mappings from fine-grained descriptions to the corresponding basic motions. We release our data at [https://github.com/KunhangL/finemotiondiffuse](https://github.com/KunhangL/finemotiondiffuse).

Across a number of sign languages, temporal and spatial characteristics of dominant hand articulation are used to express semantic and grammatical features. In this study of Austrian Sign Language (Österreichische Gebärdensprache, or ÖGS), motion capture data of four Deaf signers is used to quantitatively characterize the kinematic parameters of sign production in verbs and adjectives. We investigate (1) the difference in production between verbs involving a natural endpoint (telic verbs; e.g. arrive) and verbs lacking an endpoint (atelic verbs; e.g. analyze), and (2) adjective signs in intensified vs. non-intensified (plain) forms. Motion capture data analysis using linear-mixed effects models (LME) indicates that both the endpoint marking in verbs, as well as marking of intensification in adjectives, are expressed by movement modulation in ÖGS. While the semantic distinction between verb types (telic/atelic) is marked by higher peak velocity and shorter duration for telic signs compared to atelic ones, the grammatical distinction (intensification) in adjectives is expressed by longer duration for intensified compared to non-intensified adjectives. The observed individual differences of signers might be interpreted as personal signing style.

Morphemes serve as a strong linguistic feature to capture lexical semantics, with higher coverage than words and more natural than sememes. However, due to the lack of morpheme-informed resources and the expense of manual annotation, morpheme-enhanced methods remain largely unexplored in Computational Linguistics. To address this issue, we propose the task of Morpheme Sense Disambiguation (MSD), with two subtasks in-text and in-word, similar to Word Sense Disambiguation (WSD) and Sememe Prediction (SP), to generalize morpheme features on more tasks. We first build the MorDis resource for Chinese, including MorInv as a morpheme inventory, MorTxt and MorWrd as two types of morpheme-annotated datasets. Next, we provide two baselines in each evaluation; the best model yields a promising precision of 77.66% on in-text MSD and 88.19% on in-word MSD, indicating its comparability with WSD and superiority over SP. Finally, we demonstrate that predicted morphemes achieve comparable performance with the ground-truth ones on a downstream application of Definition Generation (DG). This validates the feasibility and applicability of our proposed tasks. The resources and workflow of MSD will provide new insights and solutions for downstream tasks, including DG as well as WSD, training pre-trained models, etc.

Paraphrase detection is a task to identify if two sentences are semantically similar or not. It plays an important role in maintaining the integrity of written work such as plagiarism detection and text reuse detection. Formerly, researchers focused on developing large corpora for English. However, no research has been conducted on sentence-level paraphrase detection in low-resource Pashto language. To bridge this gap, we introduce the first fully manually annotated Pashto sentential paraphrase detection corpus collected from authentic cases in journalism covering 10 different domains, including Sports, Health, Environment, and more. Our proposed corpus contains 6,727 sentences, encompassing 3,687 paraphrased and 3,040 non-paraphrased. Experimental findings reveal that our proposed corpus is sufficient to train XLM-RoBERTa to accurately detect paraphrased sentence pairs in Pashto with an F1 score of 84%. To compare our corpus with those in other languages, we also applied our fine-tuned model to the Indonesian and English paraphrase datasets in a zero-shot manner, achieving F1 scores of 82% and 78%, respectively. This result indicates that the quality of our corpus is not less than commonly used datasets. It‘s a pioneering contribution to the field. We will publicize a subset of 1,800 instances from our corpus, free from any licensing issues.