COLING 2024 论文列表

With the evolution of LLMs, they are endowed with impressive logical reasoning, or vertical thinking capabilities. But can they think out of the box? Do they possess proficient lateral thinking abilities? Following the setup of Lateral Thinking Puzzles, we propose a novel evaluation benchmark, LatEval, which assesses the model’s lateral thinking within an interactive framework. In our benchmark, we challenge LLMs with 2 aspects: (1) posing high-quality questions that break out of conventional norms but are beneficial for puzzle-solving. (2) integrating existing information to gradually deduce the truth through reasoning. We observe that it is hard for most LLMs to accomplish lateral thinking during interactions. Even the most powerful LLM, GPT-4, faces challenges in achieving satisfactory performance, and for most open-source models, simply completing this task is quite difficult. This evaluation benchmark provides LLMs with a highly challenging and differentiating task that is crucial to an effective AI assistant. Our dataset and source codes are available at https://github.com/THUKElab/LatEval.

Large language models (LLM) not only have revolutionized the field of natural language processing (NLP) but also have the potential to reshape many other fields, e.g., recommender systems (RS). However, most of the related work treats an LLM as a component of the conventional recommendation pipeline (e.g., as a feature extractor), which may not be able to fully leverage the generative power of LLM. Instead of separating the recommendation process into multiple stages, such as score computation and re-ranking, this process can be simplified to one stage with LLM: directly generating recommendations from the complete pool of items. This survey reviews the progress, methods, and future directions of LLM-based generative recommendation by examining three questions: 1) What generative recommendation is, 2) Why RS should advance to generative recommendation, and 3) How to implement LLM-based generative recommendation for various RS tasks. We hope that this survey can provide the context and guidance needed to explore this interesting and emerging topic.

Collecting labeled datasets in finance is challenging due to scarcity of domain experts and higher cost of employing them. While Large Language Models (LLMs) have demonstrated remarkable performance in data annotation tasks on general domain datasets, their effectiveness on domain specific datasets remains under-explored. To address this gap, we investigate the potential of LLMs as efficient data annotators for extracting relations in financial documents. We compare the annotations produced by three LLMs (GPT-4, PaLM 2, and MPT Instruct) against expert annotators and crowdworkers. We demonstrate that the current state-of-the-art LLMs can be sufficient alternatives to non-expert crowdworkers. We analyze models using various prompts and parameter settings and find that customizing the prompts for each relation group by providing specific examples belonging to those groups is paramount. Furthermore, we introduce a reliability index (LLM-RelIndex) used to identify outputs that may require expert attention. Finally, we perform an extensive time, cost and error analysis and provide recommendations for the collection and usage of automated annotations in domain-specific settings.

Topic modelling, as a well-established unsupervised technique, has found extensive use in automatically detecting significant topics within a corpus of documents. However, classic topic modelling approaches (e.g., LDA) have certain drawbacks, such as the lack of semantic understanding and the presence of overlapping topics. In this work, we investigate the untapped potential of large language models (LLMs) as an alternative for uncovering the underlying topics within extensive text corpora. To this end, we introduce a framework that prompts LLMs to generate topics from a given set of documents and establish evaluation protocols to assess the clustering efficacy of LLMs. Our findings indicate that LLMs with appropriate prompts can stand out as a viable alternative, capable of generating relevant topic titles and adhering to human guidelines to refine and merge topics. Through in-depth experiments and evaluation, we summarise the advantages and constraints of employing LLMs in topic extraction.

Modern large language models and chatbots based on them show impressive results in text generation and dialog tasks. At the same time, these models are subject to criticism in many aspects, e.g., they can generate hate speech and untrue and biased content. In this work, we show another problematic feature of such chatbots: they are echo chambers in the sense that they tend to agree with the opinions of their users. Social media, such as Facebook, was criticized for a similar problem and called an echo chamber. We experimentally test five LLM-based chatbots, which we feed with opinionated inputs. We annotate the chatbot answers whether they agree or disagree with the input. All chatbots tend to agree. However, the echo chamber effect is not equally strong. We discuss the differences between the chatbots and make the dataset publicly available.

Kurdish, an Indo-European language spoken by over 30 million speakers, is considered a dialect continuum and known for its diversity in language varieties. Previous studies addressing language and speech technology for Kurdish handle it in a monolithic way as a macro-language, resulting in disparities for dialects and varieties for which there are few resources and tools available. In this paper, we take a step towards developing resources for language and speech technology for varieties of Central Kurdish, creating a corpus by transcribing movies and TV series as an alternative to fieldwork. Additionally, we report the performance of machine translation, automatic speech recognition, and language identification as downstream tasks evaluated on Central Kurdish subdialects. Data and models are publicly available under an open license at https://github.com/sinaahmadi/CORDI.

Language identification is an important first step in many NLP applications. Most publicly available language identification datasets, however, are compiled under the assumption that the gold label of each instance is determined by where texts are retrieved from. Research has shown that this is a problematic assumption, particularly in the case of very similar languages (e.g., Croatian and Serbian) and national language varieties (e.g., Brazilian and European Portuguese), where texts may contain no distinctive marker of the particular language or variety. To overcome this important limitation, this paper presents DSL True Labels (DSL-TL), the first human-annotated multilingual dataset for language variety identification. DSL-TL contains a total of 12,900 instances in Portuguese, split between European Portuguese and Brazilian Portuguese; Spanish, split between Argentine Spanish and Castilian Spanish; and English, split between American English and British English. We trained multiple models to discriminate between these language varieties, and we present the results in detail. The data and models presented in this paper provide a reliable benchmark toward the development of robust and fairer language variety identification systems. We make DSL-TL freely available to the research community.

In this position paper we argue that researchers interested in language and/or language technologies should attend to challenges of linguistic and algorithmic injustice together with language communities. We put forward that this can be done by drawing together diverse scholarly and experiential insights, building strong interdisciplinary teams, and paying close attention to the wider social, cultural and historical contexts of both language communities and the technologies we aim to develop.

Word Sense Disambiguation (WSD) is an important task in NLP, which serves the purpose of automatically disambiguating a polysemous word with its most likely sense in context. Recent studies have advanced the state of the art in this task, but most of the work has been carried out on contemporary English or other modern languages, leaving challenges posed by low-resource languages and diachronic change open. Although the problem with low-resource languages has recently been mitigated by using existing multilingual resources to propagate otherwise expensive annotations from English to other languages, such techniques have hitherto not been applied to historical languages such as Latin. In this work, we make the following two major contributions. First, we test such a strategy on a historical language and propose a new approach in this framework which makes use of existing bilingual corpora instead of native English datasets. Second, we fine-tune a Latin WSD model on the data produced and achieve state-of-the-art results on a standard benchmark for the task. Finally, we release the dataset generated with our approach, which is the largest dataset for Latin WSD to date. This work opens the door to further research, as our approach can be used for different historical and, generally, under-resourced languages.

Recent foundational language models have shown state-of-the-art performance in many NLP tasks in zero- and few-shot settings. An advantage of these models over more standard approaches based on fine-tuning is the ability to understand instructions written in natural language (prompts), which helps them generalise better to different tasks and domains without the need for specific training data. This makes them suitable for addressing text classification problems for domains with limited amounts of annotated instances. However, existing research is limited in scale and lacks understanding of how text generation models combined with prompting techniques compare to more established methods for text classification such as fine-tuning masked language models. In this paper, we address this research gap by performing a large-scale evaluation study for 16 text classification datasets covering binary, multiclass, and multilabel problems. In particular, we compare zero- and few-shot approaches of large language models to fine-tuning smaller language models. We also analyse the results by prompt, classification type, domain, and number of labels. In general, the results show how fine-tuning smaller and more efficient language models can still outperform few-shot approaches of larger language models, which have room for improvement when it comes to text classification.

Since they rely on the distributional hypothesis, static and contextual language models are closely linked to lexical semantic relations. In this paper, we exploit this link for enhancing a BERT model. More precisely, we propose to extract lexical semantic relations with two unsupervised methods, one based on a static language model, the other on a contextual model, and to inject the extracted relations into a BERT model for improving its semantic capabilities. Through various evaluations performed for English and focusing on semantic similarity at the word and sentence levels, we show the interest of this approach, allowing us to semantically enrich a BERT model without using any external semantic resource.

Lip reading, the process of interpreting silent speech from visual lip movements, has gained rising attention for its wide range of realistic applications. Deep learning approaches greatly improve current lip reading systems. However, lip reading in cross-speaker scenarios where the speaker identity changes, poses a challenging problem due to inter-speaker variability. A well-trained lip reading system may perform poorly when handling a brand new speaker. To learn a speaker-robust lip reading model, a key insight is to reduce visual variations across speakers, avoiding the model overfitting to specific speakers. In this work, in view of both input visual clues and latent representations based on a hybrid CTC/attention architecture, we propose to exploit the lip landmark-guided fine-grained visual clues instead of frequently-used mouth-cropped images as input features, diminishing speaker-specific appearance characteristics. Furthermore, a max-min mutual information regularization approach is proposed to capture speaker-insensitive latent representations. Experimental evaluations on public lip reading datasets demonstrate the effectiveness of the proposed approach under the intra-speaker and inter-speaker conditions.

The combination of topic modeling and automatic topic labeling sheds light on understanding large corpora of text. It can be used to add semantic information for existing metadata. In addition, one can use the documents and the corresponding topic labels for topic classification. While there are existing algorithms for topic modeling readily accessible for processing texts, there is a need to postprocess the result to make the topics more interpretable and self-explanatory. The topic words from the topic model are ranked and the first/top word could easily be considered as a label. However, it is imperative to use automatic topic labeling, because the highest scored word is not the word that sums up the topic in the best way. Using the lexical-semantic word net GermaNet, the first step is to disambiguate words that are represented in GermaNet with more than one sense. We show how to find the correct sense in the context of a topic with the method of word sense disambiguation. To enhance accuracy, we present a similarity measure based on vectors of topic words that considers semantic relations of the senses demonstrating superior performance of the investigated cases compared to existing methods.

We address the challenge of detecting questionable content in online media, specifically the subcategory of comic mischief. This type of content combines elements such as violence, adult content, or sarcasm with humor, making it difficult to detect. Employing a multimodal approach is vital to capture the subtle details inherent in comic mischief content. To tackle this problem, we propose a novel end-to-end multimodal system for the task of comic mischief detection. As part of this contribution, we release a novel dataset for the targeted task consisting of three modalities: video, text (video captions and subtitles), and audio. We also design a HIerarchical Cross-attention model with CAPtions (HICCAP) to capture the intricate relationships among these modalities. The results show that the proposed approach makes a significant improvement over robust baselines and state-of-the-art models for comic mischief detection and its type classification. This emphasizes the potential of our system to empower users, to make informed decisions about the online content they choose to see.

Over-correction is a critical problem in Chinese grammatical error correction (CGEC) task. Recent work using model ensemble methods based on voting can effectively mitigate over-correction and improve the precision of the GEC system. However, these methods still require the output of several GEC systems and inevitably lead to reduced error recall. In this light, we propose the LM-Combiner, a rewriting model that can directly modify the over-correction of GEC system outputs without a model ensemble. Specifically, we train the model on an over-correction dataset constructed through the proposed K-fold cross inference method, which allows it to directly generate filtered sentences by combining the original and the over-corrected text. In the inference stage, we directly take the original sentences and the output results of other systems as input and then obtain the filtered sentences through LM-Combiner. Experiments on the FCGEC dataset show that our proposed method effectively alleviates the over-correction of the original system (+18.2 Precision) while ensuring the error recall remains unchanged. Besides, we find that LM-Combiner still has a good rewriting performance even with small parameters and few training data, and thus can cost-effectively mitigate the over-correction of black-box GEC systems (e.g., ChatGPT).

Morphological word segmentation splits a given word into its morphemes (roots and affixes), the smallest meaning-bearing units of language. We introduce a novel approach, called LLMSegm, to surface-level morphological segmentation leveraging large language models (LLMs). The proposed approach is applicable in low-data settings as well as for low-resourced languages. We show how to transform the surface-level morphological segmentation task to a binary classification problem and train LLMs to solve it efficiently. For input, we leverage the information from the default LLM subword tokenisation, and a custom morphological segmentation using novel encoding. The evaluation of LLMSegm across seven morphologically diverse languages demonstrates substantial gains in minimally-supervised settings as well as for low-resourced languages, compared to several existing competitive approaches. In terms of F1-scores and accuracy, we achieve improved results compared to the competing methods in six out of seven datasets. Keywords: morphological segmentation, surface-level segmentation, large language models, low-resource settings

Large language models have become increasingly popular and demonstrated remarkable performance in various natural language processing (NLP) tasks. However, these models are typically computationally expensive and difficult to be deployed in resource-constrained environments. In this paper, we propose LLMR, a novel knowledge distillation (KD) method based on a reward function induced from large language models. We conducted experiments on multiple datasets in the dialogue generation and summarization tasks. Empirical results demonstrate that our LLMR approach consistently outperforms traditional KD methods in different tasks and datasets.

Fact verification constitutes a pivotal application in the effort to combat the dissemination of disinformation, a concern that has recently garnered considerable attention. However, previous studies in the field of fact verification, particularly those focused on question-answering dialogue, have exhibited limitations, such as failing to fully exploit the potential of question structures and ignoring relevant label information during the verification process. In this paper, we introduce Label-Infused Iterative Information Interacting (LI4), a novel approach designed for the task of question-answering dialogue based fact verification. LI4 consists of two meticulously designed components, namely the Iterative Information Refining and Filtering Module (IIRF) and the Fact Label Embedding Module (FLEM). The IIRF uses the Interactive Gating Mechanism to iteratively filter out the noise of question and evidence, concurrently refining the claim information. The FLEM is conceived to strengthen the understanding ability of the model towards labels by injecting label knowledge. We evaluate the performance of the proposed LI4 on HEALTHVER, FAVIQ, and COLLOQUIAL. The experimental results confirm that our LI4 model attains remarkable progress, manifesting as a new state-of-the-art performance.

Chinese Large Language Models (LLMs) have recently demonstrated impressive capabilities across various NLP benchmarks and real-world applications. However, the existing benchmarks for comprehensively evaluating these LLMs are still insufficient, particularly in terms of measuring knowledge that LLMs capture. Current datasets collect questions from Chinese examinations across different subjects and educational levels to address this issue. Yet, these benchmarks primarily focus on objective questions such as multiple-choice questions, leading to a lack of diversity in question types. To tackle this problem, we propose LHMKE, a Large-scale, Holistic, and Multi-subject Knowledge Evaluation benchmark in this paper. LHMKE is designed to provide a comprehensive evaluation of the knowledge acquisition capabilities of Chinese LLMs. It encompasses 10,465 questions across 75 tasks covering 30 subjects, ranging from primary school to professional certification exams. Notably, LHMKE includes both objective and subjective questions, offering a more holistic evaluation of the knowledge level of LLMs. We have assessed 11 Chinese LLMs under the zero-shot setting, which aligns with real examinations, and compared their performance across different subjects. We also conduct an in-depth analysis to check whether GPT-4 can automatically score subjective predictions. Our findings suggest that LHMKE is a challenging and advanced testbed for Chinese LLMs.