COLING 2024 论文列表

The introduction of large language models (LLMs) has advanced natural language processing (NLP), but their effectiveness is largely dependent on pre-training resources. This is especially evident in low-resource languages, such as Sinhala, which face two primary challenges: the lack of substantial training data and limited benchmarking datasets. In response, this study introduces NSina, a comprehensive news corpus of over 500,000 articles from popular Sinhala news websites, along with three NLP tasks: news media identification, news category prediction, and news headline generation. The release of NSina aims to provide a solution to challenges in adapting LLMs to Sinhala, offering valuable resources and benchmarks for improving NLP in the Sinhala language. NSina is the largest news corpus for Sinhala, available up to date.

Pre-trained Language Models (PLMs) like BERT have achieved superior performance on different downstream tasks, even when such a model is trained on a general domain. Moreover, recent studies have shown that continued pre-training on task-specific data, known as task adaptive pre-training (TAPT), can further improve downstream task performance. However, conventional TAPT adjusts all the parameters of the PLMs, which distorts the learned generic knowledge embedded in the original PLMs weights, and it is expensive to store a whole model copy for each downstream task. In this paper, we propose NLoPT, a two-step n-gram enhanced low-rank task adaptive pre-training method, to effectively and efficiently customize a PLM to the downstream task. Specifically, we first apply low-rank adaption (LoRA), a prevalent parameter-efficient technique, for efficient TAPT. We further explicitly incorporate the task-specific multi-granularity n-gram information via the cross-attention mechanism. Experimental results on six datasets from four domains illustrate the effectiveness of NLoPT, demonstrating the superiority of LoRA based TAPT and the necessity of incorporating task-specific n-gram information.

With the advancements of transformer-based architectures, we observe the rise of natural language preprocessing (NLPre) tools capable of solving preliminary NLP tasks (e.g. tokenisation, part-of-speech tagging, dependency parsing, or morphological analysis) without any external linguistic guidance. It is arduous to compare novel solutions to well-entrenched preprocessing toolkits, relying on rule-based morphological analysers or dictionaries. Aware of the shortcomings of existing NLPre evaluation approaches, we investigate a novel method of reliable and fair evaluation and performance reporting. Inspired by the GLUE benchmark, the proposed language-centric benchmarking system enables comprehensive ongoing evaluation of multiple NLPre tools, while credibly tracking their performance. The prototype application is configured for Polish and integrated with the thoroughly assembled NLPre-PL benchmark. Based on this benchmark, we conduct an extensive evaluation of a variety of Polish NLPre systems. To facilitate the construction of benchmarking environments for other languages, e.g. NLPre-GA for Irish or NLPre-ZH for Chinese, we ensure full customization of the publicly released source code of the benchmarking system. The links to all the resources (deployed platforms, source code, trained models, datasets etc.) can be found on the project website: https://sites.google.com/view/nlpre-benchmark.

The Nguni languages have over 20 million home language speakers in South Africa. There has been considerable growth in the datasets for Nguni languages, but so far no analysis of the performance of NLP models for these languages has been reported across languages and tasks. In this paper we study pretrained language models for the 4 Nguni languages - isiXhosa, isiZulu, isiNdebele, and Siswati. We compile publicly available datasets for natural language understanding and generation, spanning 6 tasks and 11 datasets. This benchmark, which we call NGLUEni, is the first centralised evaluation suite for the Nguni languages, allowing us to systematically evaluate the Nguni-language capabilities of pretrained language models (PLMs). Besides evaluating existing PLMs, we develop new PLMs for the Nguni languages through multilingual adaptive finetuning. Our models, Nguni-XLMR and Nguni-ByT5, outperform their base models and large-scale adapted models, showing that performance gains are obtainable through limited language group-based adaptation. We also perform experiments on cross-lingual transfer and machine translation. Our models achieve notable cross-lingual transfer improvements in the lower resourced Nguni languages (isiNdebele and Siswati). To facilitate future use of NGLUEni as a standardised evaluation suite for the Nguni languages, we create a web portal to access the collection of datasets and publicly release our models.

Hierarchical text classification (HTC) is a significant but challenging task in natural language processing (NLP) due to its complex taxonomic label hierarchy. Recently, there have been a number of approaches that applied prompt learning to HTC problems, demonstrating impressive efficacy. The majority of prompt-based studies emphasize global hierarchical features by employing graph networks to represent the hierarchical structure as a whole, with limited research on maintaining path consistency within the internal hierarchy of the structure. In this paper, we formulate prompt-based HTC as a named entity recognition (NER) task and introduce conditional random fields (CRF) and Global Pointer to establish hierarchical dependencies. Specifically, we approach single- and multi-path HTC as flat and nested entity recognition tasks and model them using span- and token-based methods. By narrowing the gap between HTC and NER, we maintain the consistency of internal paths within the hierarchical structure through a simple and effective way. Extensive experiments on three public datasets show that our method achieves state-of-the-art (SoTA) performance.

We present a Norwegian pronunciation lexicon with Bokmål orthographic word forms and up to eight alternate phonological transcriptions per word form. The lexicon covers dialectal variations for five geographical areas, as well as pronunciation variations for spontaneous and manuscript-read speech. It is based on the NST Bokmål lexicon for East Norwegian, whose original phonological transcriptions have been corrected, before they were converted with dialect specific regular expression rules. To evaluate the quality and consistency of the new, rule-generated transcriptions, we trained grapheme-to phoneme (G2P) models and report our results with word- (WER) and phoneme-error-rate (PER) metrics. We found that the G2P models trained on lexica for Southwest and West Norwegian close-to written transcriptions have the lowest WER scores, and that all error-corrected, close-to-written lexica yield better WER scores than the original NST lexicon. The lexicon is available under an open license, and can be used for various language technology applications and in linguistic research.

It remains a question that how simultaneous interpretation (SI) data affects simultaneous machine translation (SiMT). Research has been limited due to the lack of a large-scale training corpus. In this work, we aim to fill in the gap by introducing NAIST-SIC-Aligned, which is an automatically-aligned parallel English-Japanese SI dataset. Starting with a non-aligned corpus NAIST-SIC, we propose a two-stage alignment approach to make the corpus parallel and thus suitable for model training. The first stage is coarse alignment where we perform a many-to-many mapping between source and target sentences, and the second stage is fine-grained alignment where we perform intra- and inter-sentence filtering to improve the quality of aligned pairs. To ensure the quality of the corpus, each step has been validated either quantitatively or qualitatively. This is the first open-sourced large-scale parallel SI dataset in the literature. We also manually curated a small test set for evaluation purposes. Our results show that models trained with SI data lead to significant improvement in translation quality and latency over baselines. We hope our work advances research on SI corpora construction and SiMT. Our data will be released upon the paper’s acceptance.

This article describes the [corpus-name] corpus developed as part of the [project-name] project. To the best of our knowledge, this is one of the largest collections of children’s conversational speech that is freely available for non-commercial use under the creative commons license (CC BY-NC-SA 4.0). It comprises approximately 400 hours of speech, spanning some 230K utterances spread across about 10,500 virtual tutor sessions. Roughly 1,300 third, fourth and fifth grade students contributed to this corpus. The current release contains roughly 100K transcribed utterances. It is our hope that the corpus can be used to improve automatic speech recognition models and algorithms. We report the word error rate achieved on the test set using a model trained on the training and development portion of the corpus. The git repository of the corpus contains the complete training and evaluation setup in order to facilitate a fair and consistent evaluation. It is our hope that this corpus will contribute to the creation and evaluation of conversational AI agents having a better understanding of children’s speech, potentially opening doors to novel, effective, learning and therapeutic interventions.

This paper presents Murre24, a collection of dialectal messages posted on the largest Finnish internet forum, Suomi24. The messages posted in Finnish on the forum between 2001 and 2020 are classified to present either the standard language, one of the seven traditional dialects, a colloquial style or the Helsinki slang. We present a manually annotated dataset used to train dialect identification models as well as the automatic annotation of almost 94 million messages in total. We experiment with five different dialect identification methods and evaluate them on dialectally balanced and random test samples. The best performing method for differentiating standard Finnish from non-standard Finnish is a character n-gram based support vector machine (SVM), while fine-tuning a BERT-based model achieves best scores in the final dialect identification task. According to the automatic classification, most of the messages written on the forum are in standard Finnish, and most of the non-standard messages are in a colloquial variety used typically by young speakers in Finland. We moreover show that the proportion of non-standard messages declines over time, but the proportion of the traditional dialects stays relatively steady.

This paper presents the development of a novel multimodal multilingual dataset in Russian and English, with a particular emphasis on the exploration of laughter detection techniques. Data was collected from YouTube stand-up comedy videos with manually annotated subtitles, and our research covers data preparation and laughter labeling. We explore two laughter detection approaches presented in the literature: peak detection using preprocessed voiceless audio with an energy-based algorithm and machine learning approach with pretrained models to identify laughter presence and duration. While the machine learning approach currently outperforms peak detection in accuracy and generalization, the latter shows promise and warrants further study. Additionally, we explore unimodal and multimodal humor detection on the new dataset, showing the effectiveness of neural models in capturing humor in both languages, even with textual data. Multimodal experiments indicate that even basic models benefit from visual data, improving detection results. However, further research is needed to enhance laughter detection labeling quality and fully understand the impact of different modalities in a multimodal and multilingual context.

Multimodal large language models (MLLMs) are gaining popularity as partial solutions to the “symbol grounding problem” faced by language models trained on text alone. However, little is known about whether and how these multiple modalities are integrated. We draw inspiration from analogous work in human psycholinguistics on embodied simulation, i.e., the hypothesis that language comprehension is grounded in sensorimotor representations. We show that MLLMs are sensitive to implicit visual features like object shape (e.g., “The egg was in the skillet” implies a frying egg rather than one in a shell). This suggests that MLLMs activate implicit information about object shape when it is implied by a verbal description of an event. We find mixed results for color and orientation, and rule out the possibility that this is due to models’ insensitivity to those features in our dataset overall. We suggest that both human psycholinguistics and computational models of language could benefit from cross-pollination, e.g., with the potential to establish whether grounded representations play a functional role in language processing.

Cross-lingual phrase retrieval aims to retrieve parallel phrases among languages. Current approaches only deals with textual modality. There lacks multimodal data resources and explorations for multimodal cross-lingual phrase retrieval (MXPR). In this paper, we create the first MXPR data resource and propose a novel approach for MXPR to explore the effectiveness of multi-modality. The MXPR data resource is built by marrying the benchmark dataset for textual cross-lingual phrase retrieval with Wikimedia Commons, which is a media store containing tremendous texts and related images. In the built resource, the phrase pairs of the textual benchmark dataset are equipped with their related images. Based on this novel data resource, we introduce a strategy to bridge the gap between different modalities by multimodal relation generation with a large multimodal pre-trained model and consistency training. Experiments on benchmarked dataset covering eight language pairs show that our MXPR approach, which deals with multimodal phrases, performs significantly better than pure textual cross-lingual phrase retrieval.

Event coreference resolution (ECR) is the task of determining whether distinct mentions of events within a multi-document corpus are actually linked to the same underlying occurrence. Images of the events can help facilitate resolution when language is ambiguous. Here, we propose a multimodal cross-document event coreference resolution method that integrates visual and textual cues with a simple linear map between vision and language models. As existing ECR benchmark datasets rarely provide images for all event mentions, we augment the popular ECB+ dataset with event-centric images scraped from the internet and generated using image diffusion models. We establish three methods that incorporate images and text for coreference: 1) a standard fused model with finetuning, 2) a novel linear mapping method without finetuning and 3) an ensembling approach based on splitting mention pairs by semantic and discourse-level difficulty. We evaluate on 2 datasets: the augmented ECB+, and AIDA Phase 1. Our ensemble systems using cross-modal linear mapping establish an upper limit (91.9 CoNLL F1) on ECB+ ECR performance given the preprocessing assumptions used, and establish a novel baseline on AIDA Phase 1. Our results demonstrate the utility of multimodal information in ECR for certain challenging coreference problems, and highlight a need for more multimodal resources in the coreference resolution space.

This paper introduces a novel multimodal corpus consisting of 12 video recordings of Zoom meetings held in English by an international group of researchers from September 2021 to March 2023. The meetings have an average duration of about 40 minutes each, for a total of 8 hours. The number of participants varies from 5 to 9 per meeting. The participants’ speech was transcribed automatically using WhisperX, while visual coordinates of several keypoints of the participants’ head, their shoulders and wrists, were extracted using OpenPose. The audio-visual recordings will be distributed together with the orthographic transcription as well as the visual coordinates. In the paper we describe the way the corpus was collected, transcribed and enriched with the visual coordinates, we give descriptive statistics concerning both the speech transcription and the visual keypoint values and we present and discuss visualisations of these values. Finally, we carry out a short preliminary analysis of the role of feedback in the meetings, and show how visualising the coordinates extracted via OpenPose can be used to see how gestural behaviour supports the use of feedback words during the interaction.

Machine translation quality is highly reliant on large amounts of training data, and, when a limited amount of parallel data is available, synthetic back-translated or multilingual data can be used in addition. In this work, we introduce SynEst, a synthetic corpus of translations from 11 languages into Estonian which totals over 1 billion sentence pairs. Using this corpus, we investigate whether adding synthetic or English-centric additional data yields better translation quality for translation directions that do not include English. Our results show that while both strategies are effective, synthetic data gives better results. Our final models improve the performance of the baseline No Language Left Behind model while retaining its source-side multilinguality.

This paper investigates the application of voice activity projection (VAP), a predictive turn-taking model for spoken dialogue, on multilingual data, encompassing English, Mandarin, and Japanese. The VAP model continuously predicts the upcoming voice activities of participants in dyadic dialogue, leveraging a cross-attention Transformer to capture the dynamic interplay between participants. The results show that a monolingual VAP model trained on one language does not make good predictions when applied to other languages. However, a multilingual model, trained on all three languages, demonstrates predictive performance on par with monolingual models across all languages. Further analyses show that the multilingual model has learned to discern the language of the input signal. We also analyze the sensitivity to pitch, a prosodic cue that is thought to be important for turn-taking. Finally, we compare two different audio encoders, contrastive predictive coding (CPC) pre-trained on English, with a recent model based on multilingual wav2vec 2.0 (MMS).

Word Sense Induction (WSI) is the task of discovering senses of an ambiguous word by grouping usages of this word into clusters corresponding to these senses. Many approaches were proposed to solve WSI in English and a few other languages, but these approaches are not easily adaptable to new languages. We present multilingual substitution-based WSI methods that support any of 100 languages covered by the underlying multilingual language model with minimal to no adaptation required. Despite the multilingual capabilities, our methods perform on par with the existing monolingual approaches on popular English WSI datasets. At the same time, they will be most useful for lower-resourced languages which miss lexical resources available for English, thus, have higher demand for unsupervised methods like WSI.

Prior work on multilingual sentence embedding has demonstrated that the efficient use of natural language inference (NLI) data to build high-performance models can outperform conventional methods. However, the potential benefits from the recent “exponential” growth of language models with billions of parameters have not yet been fully explored. In this paper, we introduce Multilingual Sentence T5 (m-ST5), as a larger model of NLI-based multilingual sentence embedding, by extending Sentence T5, an existing monolingual model. By employing the low-rank adaptation (LoRA) technique, we have achieved a successful scaling of the model’s size to 5.7 billion parameters. We conducted experiments to evaluate the performance of sentence embedding and verified that the method outperforms the NLI-based prior approach. Furthermore, we also have confirmed a positive correlation between the size of the model and its performance. It was particularly noteworthy that languages with fewer resources or those with less linguistic similarity to English benefited more from the parameter increase. Our model is available at https://huggingface.co/pkshatech/m-ST5.

The emergence of pre-trained models marks a significant juncture for the multilingual generation, offering unprecedented capabilities to comprehend and produce text across multiple languages. These models display commendable efficiency in high-resource languages. However, their performance notably falters in low-resource languages due to the extensive linguistic diversity encountered. Moreover, the existing works lack thorough analysis impairs the discovery of effective multilingual strategies, further complicating the advancement of current multilingual generation systems. This paper aims to appraise the efficacy of multilingual generation tasks, with a focus on summarization, through three resource availability scenarios: high-resource, low-resource, and zero-shot. We classify multilingual generation methodologies into three foundational categories based on their underlying modeling principles: Fine-tuning, Parameter-isolation, and Constraint-based approaches. Following this classification, we conduct a comprehensive comparative study of these methodologies across different resource contexts using two datasets that span six languages. This analysis provides insights into the unique advantages and limitations of each method. In addition, we introduce an innovative yet simple automatic metric LANGM designed to mitigate the prevalent problem of spurious correlations associated with language mixing. LANGM accurately measures the degree of code-mixing at the language level. Finally, we highlight several challenges and suggest potential avenues for future inquiry, aiming to spur further advancements within the field of multilingual text generation.