COLING 2024 Papers

Pre-trained language models (PLMs) leverage chains-of-thought (CoT) to simulate human reasoning and inference processes, achieving proficient performance in multi-hop QA. However, a gap persists between PLMs’ reasoning abilities and those of humans when tackling complex problems. Psychological studies suggest a vital connection between explicit information in passages and human prior knowledge during reading. Nevertheless, current research has given insufficient attention to linking input passages and PLMs’ pre-training-based knowledge from the perspective of human cognition studies. In this study, we introduce a Prompting Explicit and Implicit knowledge (PEI) framework, which uses prompts to connect explicit and implicit knowledge, aligning with human reading process for multi-hop QA. We consider the input passages as explicit knowledge, employing them to elicit implicit knowledge through unified prompt reasoning. Furthermore, our model incorporates type-specific reasoning via prompts, a form of implicit knowledge. Experimental results show that PEI performs comparably to the state-of-the-art on HotpotQA. Ablation studies confirm the efficacy of our model in bridging and integrating explicit and implicit knowledge.

Given the importance of identifying and monitoring news stories within the continuous flow of news articles, this paper presents PromptStream, a novel method for unsupervised news story discovery. In order to identify coherent and comprehensive stories across the stream, it is crucial to create article representations that incorporate as much topic-related information from the articles as possible. PromptStream constructs these article embeddings using cloze-style prompting. These representations continually adjust to the evolving context of the news stream through self-supervised learning, employing a contrastive loss and a memory of the most confident article-story assignments from the most recent days. Extensive experiments with real news datasets highlight the notable performance of our model, establishing a new state of the art. Additionally, we delve into selected news stories to reveal how the model’s structuring of the article stream aligns with story progression.

Answering logical queries on knowledge graphs (KG) poses a significant challenge for machine reasoning. The primary obstacle in this task stems from the inherent incompleteness of KGs. Existing research has predominantly focused on addressing the issue of missing edges in KGs, thereby neglecting another aspect of incompleteness: the emergence of new entities. Furthermore, most of the existing methods tend to reason over each logical operator separately, rather than comprehensively analyzing the query as a whole during the reasoning process. In this paper, we propose a query-aware prompt-fused framework named Pro-QE, which could incorporate existing query embedding methods and address the embedding of emerging entities through contextual information aggregation. Additionally, a query prompt, which is generated by encoding the symbolic query, is introduced to gather information relevant to the query from a holistic perspective. To evaluate the efficacy of our model in the inductive setting, we introduce two new challenging benchmarks. Experimental results demonstrate that our model successfully handles the issue of unseen entities in logical queries. Furthermore, the ablation study confirms the efficacy of the aggregator and prompt components.

In relation triplet extraction (RTE), recognizing unseen relations for which there are no training instances is a challenging task. Efforts have been made to recognize unseen relations based on question-answering models or relation descriptions. However, these approaches miss the semantic information about connections between seen and unseen relations. In this paper, We propose a prompt-based model with semantic knowledge augmentation (ZS-SKA) to recognize unseen relations under the zero-shot setting. We present a new word-level analogy-based sentence translation rule and generate augmented instances with unseen relations from instances with seen relations using that new rule. We design prompts with weighted virtual label construction based on an external knowledge graph to integrate semantic knowledge information learned from seen relations. Instead of using the actual label sets in the prompt template, we construct weighted virtual label words. We learn the representations of both seen and unseen relations with augmented instances and prompts. We then calculate the distance between the generated representations using prototypical networks to predict unseen relations. Extensive experiments conducted on three public datasets FewRel, Wiki-ZSL, and NYT, show that ZS-SKA outperforms other methods under zero-shot setting. Results also demonstrate the effectiveness and robustness of ZS-SKA.

Synthetic lethality (SL) offers a promising approach for targeted anti-cancer therapy. Deeply understanding SL gene pair mechanisms is vital for anti-cancer drug discovery. However, current wet-lab and machine learning-based SL prediction methods lack user-friendly and quantitatively evaluable explanations. To address these problems, we propose a prompt-based pipeline for generating natural language explanations. We first construct a natural language dataset named NexLeth. This dataset is derived from New Bing through prompt-based queries and expert annotations and contains 707 instances. NexLeth enhances the understanding of SL mechanisms and it is a benchmark for evaluating SL explanation methods. For the task of natural language generation for SL explanations, we combine subgraph explanations from an SL knowledge graph (KG) with instructions to construct novel personalized prompts, so as to inject the domain knowledge into the generation process. We then leverage the prompts to fine-tune pre-trained biomedical language models on our dataset. Experimental results show that the fine-tuned model equipped with designed prompts performs better than existing biomedical language models in terms of text quality and explainability, suggesting the potential of our dataset and the fine-tuned model for generating understandable and reliable explanations of SL mechanisms.

Recently, prompt-tuning has achieved very significant results for few-shot tasks. The core idea of prompt-tuning is to insert prompt templates into the input, thus converting the classification task into a masked language modeling problem. However, for few-shot relation extraction tasks, how to mine more information from limited resources becomes particularly important. In this paper, we first construct a global relation graph based on label consistency to optimize the feature representation of samples between different relations. Then the global relation graph is further divided to form a local relation subgraph for each relation type to optimize the feature representation of samples within the same relation. This fully uses the limited supervised information and improves the tuning efficiency. In addition, the existence of rich semantic knowledge in relation labels cannot be ignored. For this reason, this paper incorporates the knowledge in relation labels into prompt-tuning. Specifically, the potential knowledge implicit in relation labels is injected into constructing learnable prompt templates. In this paper, we conduct extensive experiments on four datasets under low-resource settings, showing that this method achieves significant results.

The development of large language models (LLMs) raises the importance of assessing the fairness and completeness of various evaluation benchmarks. Regrettably, these benchmarks predominantly utilize uniform manual prompts, which may not fully capture the expansive capabilities of LLMs—potentially leading to an underestimation of their performance. To unlock the potential of LLMs, researchers pay attention to automated prompt search methods, which employ LLMs as optimizers to discover optimal prompts. However, previous methods generate the solutions implicitly, which overlook the underlying thought process and lack explicit feedback. In this paper, we propose a novel prompt introspective search framework, namely PromISe, to better release the capabilities of LLMs. It converts the process of optimizing prompts into an explicit chain of thought, through a step-by-step procedure that integrates self-introspect and self-refine. Extensive experiments, conducted over 73 tasks on two major benchmarks, demonstrate that our proposed PromISe significantly boosts the performance of 12 well-known LLMs compared to the baseline approach. Moreover, our study offers enhanced insights into the interaction between humans and LLMs, potentially serving as a foundation for future designs and implementations. Keywords: large language models, prompt search, self-introspect, self-refine

Mitigation of gender bias in NLP has a long history tied to debiasing static word embeddings. More recently, attention has shifted to debiasing pre-trained language models. We study to what extent the simplest projective debiasing methods, developed for word embeddings, can help when applied to BERT’s internal representations. Projective methods are fast to implement, use a small number of saved parameters, and make no updates to the existing model parameters. We evaluate the efficacy of the methods in reducing both intrinsic bias, as measured by BERT’s next sentence prediction task, and in mitigating observed bias in a downstream setting when fine-tuned. To this end, we also provide a critical analysis of a popular gender-bias assessment test for quantifying intrinsic bias, resulting in an enhanced test set and new bias measures. We find that projective methods can be effective at both intrinsic bias and downstream bias mitigation, but that the two outcomes are not necessarily correlated. This finding serves as a warning that intrinsic bias test sets, based either on language modeling tasks or next sentence prediction, should not be the only benchmark in developing a debiased language model.

Second language acquisition (SLA) is a complex and dynamic process. Many SLA studies that have attempted to record and analyze this process have typically focused on a single modality (e.g., textual output of learners), covered only a short period of time, and/or lacked control (e.g., failed to capture every aspect of the learning process). In Project MOSLA (Moments of Second Language Acquisition), we have created a longitudinal, multimodal, multilingual, and controlled dataset by inviting participants to learn one of three target languages (Arabic, Spanish, and Chinese) from scratch over a span of two years, exclusively through online instruction, and recording every lesson using Zoom. The dataset is semi-automatically annotated with speaker/language IDs and transcripts by both human annotators and fine-tuned state-of-the-art speech models. Our experiments reveal linguistic insights into learners’ proficiency development over time, as well as the potential for automatically detecting the areas of focus on the screen purely from the unannotated multimodal data. Our dataset is freely available for research purposes and can serve as a valuable resource for a wide range of applications, including but not limited to SLA, proficiency assessment, language and speech processing, pedagogy, and multimodal learning analytics.

In this paper we present the initial construction of a treebank of Ancient Greek containing portions of the Septuagint, a translation of the Hebrew Scriptures (1576 sentences, 39K tokens, roughly 7% of the total corpus). We construct the treebank by word-aligning and projecting from the parallel text in Ancient Hebrew before automatically correcting systematic syntactic mismatches and manually correcting other errors.

The advancement of Multimodal Large Language Models (MLLMs) has greatly accelerated the development of applications in understanding integrated texts and images. Recent works leverage image-caption datasets to train MLLMs, achieving state-of-the-art performance on image-to-text tasks. However, there are few studies exploring which layers of MLLMs make the most effort to the global image information, which plays vital roles in multimodal comprehension and generation. In this study, we find that the intermediate layers of models can encode more global semantic information, whose representation vectors perform better on visual-language entailment tasks, rather than the topmost layers. We further probe models regarding local semantic representations through object recognition tasks. We find that the topmost layers may excessively focus on local information, leading to a diminished ability to encode global information. Our code and data are released via https://github.com/kobayashikanna01/probing_MLLM_rep.

Scalar adjectives pertain to various domain scales and vary in intensity within each scale (e.g. certain is more intense than likely on the likelihood scale). Scalar implicatures arise from the consideration of alternative statements which could have been made. They can be triggered by scalar adjectives and require listeners to reason pragmatically about them. Some scalar adjectives are more likely to trigger scalar implicatures than others. This phenomenon is referred to as scalar diversity. In this study, we probe different families of Large Language Models such as GPT-4 for their knowledge of the lexical semantics of scalar adjectives and one specific aspect of their pragmatics, namely scalar diversity. We find that they encode rich lexical-semantic information about scalar adjectives. However, the rich lexical-semantic knowledge does not entail a good understanding of scalar diversity. We also compare current models of different sizes and complexities and find that larger models are not always better. Finally, we explain our probing results by leveraging linguistic intuitions and model training objectives.

Step-by-step reasoning methods, such as the Chain-of-Thought (CoT), have been demonstrated to be highly effective in harnessing the reasoning capabilities of Large Language Models (LLMs). Recent research efforts have sought to distill LLMs into Small Language Models (SLMs), with a significant focus on transferring the reasoning capabilities of LLMs to SLMs via CoT. However, the outcomes of CoT distillation are inadequate for knowledge-intensive reasoning tasks. This is because generating accurate rationales requires crucial factual knowledge, which SLMs struggle to retain due to their parameter constraints. We propose a retrieval-based CoT distillation framework, named Probe then Retrieve and Reason (PRR), which distills the question probing and reasoning capabilities from LLMs into SLMs. We train two complementary distilled SLMs, a probing model and a reasoning model, in tandem. When presented with a new question, the probing model first identifies the necessary knowledge to answer it, generating queries for retrieval. Subsequently, the reasoning model uses the retrieved knowledge to construct a step-by-step rationale for the answer. In knowledge-intensive reasoning tasks, such as StrategyQA and OpenbookQA, our distillation framework yields superior performance for SLMs compared to conventional methods, including simple CoT distillation and knowledge-augmented distillation using raw questions.

Retrieval-based code question answering seeks to match user queries in natural language to relevant code snippets. Previous approaches typically rely on pretraining models using crafted bi-modal and uni-modal datasets to align text and code representations. In this paper, we introduce ProCQA, a large-scale programming question answering dataset extracted from the StackOverflow community, offering naturally structured mixed-modal QA pairs. To validate its effectiveness, we propose a modality-agnostic contrastive pre-training approach to improve the alignment of text and code representations of current code language models. Compared to previous models that primarily employ bimodal and unimodal pairs extracted from CodeSearchNet for pre-training, our model exhibits significant performance improvements across a wide range of code retrieval benchmarks.

Knowledge Graph Completion (KGC) is a task aimed at uncovering the inherent relationships among known knowledge triplets in a Knowledge Graph (KG) and subsequently predicting missing links. Presently, there is a rising interest in inductive knowledge graph completion, where missing links may pertain to previously unobserved entities. Previous inductive KGC methods mainly rely on descriptive information of entities to improve the representation of unseen entities, neglecting to provide effective prior knowledge for relation modeling. To tackle this challenge, we capture prior schema-level interactions related to relations by leveraging entity type information, thereby furnishing effective prior constraints when reasoning with newly introduced entities. Moreover, We employ normal in-batch negatives and introduce schema-guided negatives to bolster the efficiency of normal contrastive representation learning. Experimental results demonstrate that our approach consistently achieves state-of-the-art performance on various established metrics across multiple benchmark datasets for link prediction. Notably, our method achieves a 20.5% relative increase in Hits@1 on the HumanWiki-Ind dataset.

Bayesian approaches to reconstructing the evolutionary history of languages rely on the tree model, which assumes that these languages descended from a common ancestor and underwent modifications over time. However, this assumption can be violated to different extents due to contact and other factors. Understanding the degree to which this assumption is violated is crucial for validating the accuracy of phylolinguistic inference. In this paper, we propose a simple sanity check: projecting a reconstructed tree onto a space generated by principal component analysis. By using both synthetic and real data, we demonstrate that our method effectively visualizes anomalies, particularly in the form of jogging.

While impressive performance has been achieved in image captioning, the limited diversity of the generated captions and the large parameter scale remain major barriers to the real-word application of these systems. In this work, we propose a lightweight image captioning network in combination with continuous diffusion, called Prefix-diffusion. To achieve diversity, we design an efficient method that injects prefix image embeddings into the denoising process of the diffusion model. In order to reduce trainable parameters, we employ a pre-trained model to extract image features and further design an extra mapping network. Prefix-diffusion is able to generate diverse captions with relatively less parameters, while maintaining the fluency and relevance of the captions benefiting from the generative capabilities of the diffusion model. Our work paves the way for scaling up diffusion models for image captioning, and achieves promising performance compared with recent approaches.

In this study, we analyze spontaneous speech transcripts from Hungarian patients with schizophrenia, schizoaffective, and bipolar disorders. Our goal is to identify distinctive linguistic features in these patient groups and controls. To our knowledge, no prior study has systematically examined the linguistic features of these disorders or explored their use in distinguishing between these patient groups. We collected recordings from 77 participants during three directed spontaneous speech tasks in a clinical setting, resulting in 458 texts. Our research group manually transcribed the recordings. We processed the written corpus texts using Natural Language Processing methods and tools. The final corpus consists of 179,515 tokens, excluding punctuation. Using this data, we analyze different linguistic features’ predictive power by computing and comparing their frequency distributions. We then attempt to automatically differentiate between patient groups and controls using our extensive set of linguistic features, employing the random forest algorithm in these experiments. Our results indicate that applying machine learning techniques based on distinctive features can effectively distinguish SZ, SAD, BD, and controls, surpassing baseline results.

Recent large-scale vision-language pre-training depends on image-text global alignment by contrastive learning and is further boosted by fine-grained alignment in a weakly contrastive manner for cross-modal retrieval. Nonetheless, besides semantic matching learned by contrastive learning, cross-modal retrieval also largely relies on object matching between modalities. This necessitates fine-grained categorical discriminative learning, which however suffers from scarce data in full-supervised scenarios and information asymmetry in weakly-supervised scenarios when applied to cross-modal retrieval. To address these issues, we propose expansive lexicon-patch alignment (ELA) to align image patches with a vocabulary rather than only the words explicitly in the text for annotation-free alignment and information augmentation, thus enabling more effective fine-grained categorical discriminative learning for cross-modal retrieval. Experimental results show that ELA could effectively learn representative fine-grained information and outperform state-of-the-art methods on cross-modal retrieval.