EMNLP 2023 Papers

Recent efforts have endeavored to employ the sequence-to-sequence (Seq2Seq) model in Information Extraction~(IE) due to its potential to tackle multiple IE tasks in a unified manner. Under this formalization, multiple structured objects are concatenated as the target sequence in a predefined order. However, structured objects, by their nature, constitute an unordered set. Consequently, this formalization introduces a potential order bias, which can impair model learning. Targeting this issue, this paper proposes a set learning approach that considers multiple permutations of structured objects to optimize set probability approximately. Notably, our approach does not require any modifications to model structures, making it easily integrated into existing generative IE frameworks. Experiments show that our method consistently improves existing frameworks on vast tasks and datasets.

Visual Word Sense Disambiguation (VWSD) is a novel challenging task with the goal of retrieving an image among a set of candidates, which better represents the meaning of an ambiguous word within a given context. In this paper, we make a substantial step towards unveiling this interesting task by applying a varying set of approaches. Since VWSD is primarily a text-image retrieval task, we explore the latest transformer-based methods for multimodal retrieval. Additionally, we utilize Large Language Models (LLMs) as knowledge bases to enhance the given phrases and resolve ambiguity related to the target word. We also study VWSD as a unimodal problem by converting to text-to-text and image-to-image retrieval, as well as question-answering (QA), to fully explore the capabilities of relevant models. To tap into the implicit knowledge of LLMs, we experiment with Chain-of-Thought (CoT) prompting to guide explainable answer generation. On top of all, we train a learn to rank (LTR) model in order to combine our different modules, achieving competitive ranking results. Extensive experiments on VWSD demonstrate valuable insights to effectively drive future directions.

One of the main problems in speech translation is the mismatches between different modalities. The second problem, scarcity of parallel data covering multiple modalities, means that the end-to-end multi-modal models tend to perform worse than cascade models, although there are exceptions under favorable conditions. To address these problems, we propose an end-to-end zero-shot speech translation model, connecting two pre-trained uni-modality modules via word rotator's distance. The model retains the ability of zero-shot, which is like cascade models, and also can be trained in an end-to-end style to avoid error propagation. Our comprehensive experiments on the MuST-C benchmarks show that our end-to-end zero-shot approach performs better than or as well as those of the CTC-based cascade models and that our end-to-end model with supervised training also matches the latest baselines.

Prevalent supervised learning methods in natural language processing (NLP) are notoriously data-hungry, which demand large amounts of high-quality annotated data. In practice, acquiring such data is a costly endeavor. Recently, the superior few-shot performance of large language models (LLMs) has propelled the development of dataset generation, where the training data are solely synthesized from LLMs. However, such an approach usually suffers from low-quality issues, and requires orders of magnitude more labeled data to achieve satisfactory performance. To fully exploit the potential of LLMs and make use of massive unlabeled data, we propose LLMaAA, which takes LLMs as annotators and puts them into an active learning loop to determine what to annotate efficiently. To learn robustly with pseudo labels, we optimize both the annotation and training processes: (1) we draw $k$-NN examples from a small demonstration pool as in-context examples, and (2) we adopt the example reweighting technique to assign training samples with learnable weights. Compared with previous approaches, LLMaAA features both efficiency and reliability. We conduct experiments and analysis on two classic NLP tasks, named entity recognition and relation extraction. With LLMaAA, task-specific models trained from LLM-generated labels can outperform the teacher within only hundreds of annotated examples, which is much more cost-effective than other baselines.

A natural way to design a negotiation dialogue system is via self-play RL: train an agent that learns to maximize its performance by interacting with a simulated user that has been designed to imitate human-human dialogue data. Although this procedure has been adopted in prior work, we find that it results in a fundamentally flawed system that fails to learn the value of compromise in a negotiation, which can often lead to no agreements (i.e., the partner walking away without a deal), ultimately hurting the model’s overall performance. We investigate this observation in the context of DealOrNoDeal task, a multi-issue negotiation over books, hats, and balls. Grounded in negotiation theory from Economics, we modify the training procedure in two novel ways to design agents with diverse personalities and analyze their performance with human partners. We find that although both techniques show promise, a selfish agent, which maximizes its own performance while also avoiding walkaways, performs superior to other variants by implicitly learning to generate value for both itself and the negotiation partner. We discuss the implications of our findings for what it means to be a successful negotiation dialogue system and how these systems should be designed in the future.

Improving the quality of academic writing is a meaningful but challenging task. Conventional methods of language refinement focus on narrow, specific linguistic features within isolated sentences, such as grammatical errors and improper word use. We propose a more general task, Academic Writing Formalization (AWF), to improve the overall quality of formal academic writing at the paragraph level. We formulate this language refinement task as a formal text style transfer task which transfers informal-academic text to formal-academic and contribute a large-scale non-parallel dataset, Doolittle, for this purpose. Concurrently, we apply a method named metric-oriented reinforcement learning (MORL) to two large language models (LLM) where we incorporate different levels of automatic feedback into the training process. Our experiments reveal that existing text transfer models and grammatical error correction models address certain aspects of AWF but still have a significant performance gap compared to human performance. Meanwhile, language models fine-tuned with our MORL method exhibit considerably improved performance, rivaling the latest chatbot ChatGPT, but still have a non-negligible gap compared to the ground truth formal-academic texts in Doolittle.

This paper introduces a novel approach for identifying the possible large language models (LLMs) involved in text generation. Instead of adding an additional classification layer to a base LM, we reframe the classification task as a next-token prediction task and directly fine-tune the base LM to perform it. We utilize the Text-to-Text Transfer Transformer (T5) model as the backbone for our experiments. We compared our approach to the more direct approach of utilizing hidden states for classification. Evaluation shows the exceptional performance of our method in the text classification task, highlighting its simplicity and efficiency. Furthermore, interpretability studies on the features extracted by our model reveal its ability to differentiate distinctive writing styles among various LLMs even in the absence of an explicit classifier. We also collected a dataset named OpenLLMText, containing approximately 340k text samples from human and LLMs, including GPT3.5, PaLM, LLaMA, and GPT2.

Negation is the fundamental component in a natural language that reverses the semantic meaning of a sentence. It plays an extremely important role across a wide range of applications, yet they are underrepresented in pre-trained language models (LMs), resulting often in wrong inferences. In this work, we try to improve the underlying understanding of the negation in the pre-trained LMs. To augment negation understanding, we propose a language model objective with a weighted cross-entropy loss and elastic weight consolidation regularization. We reduce the mean top 1 error rate for BERT-base to 1.1\%, BERT-large to 0.78\%, RoBERTA-base to 3.74\%, RoBERTA-large to 0.01\% on the negated LAMA dataset. It minimizes the BERT error rate by a margin of 8\% and also outperform the existing negation models. We also provide empirical evidences that negated augmented models outperform the classical models on original as well as negation benchmarks on natural language inference tasks.

This paper presents a benchmark dataset of Bangla word analogies for evaluating the quality of existing Bangla word embeddings. Despite being the 7th largest spoken language in the world, Bangla is still a low-resource language and popular NLP models often struggle to perform well on Bangla data sets. Therefore, developing a robust evaluation set is crucial for benchmarking and guiding future research on improving Bangla word embeddings, which is currently missing. To address this issue, we introduce a new evaluation set of 16,678 unique word analogies in Bangla as well as a translated and curated version of the original Mikolov dataset (10,594 samples) in Bangla. Our experiments with different state-of-the-art embedding models reveal that current Bangla word embeddings struggle to achieve high accuracy on both data sets, demonstrating a significant gap in multilingual NLP research.

Prompt tuning attempts to update few task-specific parameters in pre-trained models. It has achieved comparable performance to fine-tuning of the full parameter set on both language understanding and generation tasks. In this work, we study the problem of prompt tuning for neural text retrievers. We introduce parameter-efficient prompt tuning for text retrieval across in-domain, cross-domain, and cross-topic settings. Through an extensive analysis, we show that the strategy can mitigate the two issues---parameter-inefficiency and weak generalizability---faced by fine-tuning based retrieval methods. Notably, it can significantly improve the out-of-domain zero-shot generalization of the retrieval models. By updating only 0.1% of the model parameters, the prompt tuning strategy can help retrieval models achieve better generalization performance than traditional methods in which all parameters are updated. Finally, to facilitate research on retrievers' cross-topic generalizability, we curate and release an academic retrieval dataset with 18K query-results pairs in 87 topics, making it the largest topic-specific one to date.

Cross-lingual transfer (XLT) is an emergent ability of multilingual language models that preserves their performance on a task to a significant extent when evaluated in languages that were not included in the fine-tuning process. While English, due to its widespread usage, is typically regarded as the primary language for model adaption in various tasks, recent studies have revealed that the efficacy of XLT can be amplified by selecting the most appropriate source languages based on specific conditions. In this work, we propose the utilization of sub-network similarity between two languages as a proxy for predicting the compatibility of the languages in the context of XLT. Our approach is model-oriented, better reflecting the inner workings of foundation models. In addition, it requires only a moderate amount of raw text from candidate languages, distinguishing it from the majority of previous methods that rely on external resources. In experiments, we demonstrate that our method is more effective than baselines across diverse tasks. Specifically, it shows proficiency in ranking candidates for zero-shot XLT, achieving an improvement of 4.6% on average in terms of NDCG@3. We also provide extensive analyses that confirm the utility of sub-networks for XLT prediction.

Meetings typically involve multiple participants and lengthy conversations, resulting in redundant and trivial content. To overcome these challenges, we propose a two-step framework, Reconstruct before Summarize (RbS), for effective and efficient meeting summarization. RbS first leverages a self-supervised paradigm to annotate essential contents by reconstructing the meeting transcripts. Secondly, we propose a relative positional bucketing (RPB) algorithm to equip (conventional) summarization models to generate the summary. Despite the additional reconstruction process, our proposed RPB significantly compresses the input, leading to faster processing and reduced memory consumption compared to traditional summarization methods. We validate the effectiveness and efficiency of our method through extensive evaluations and analyses. On two meeting summarization datasets, AMI and ICSI, our approach outperforms previous state-of-the-art approaches without relying on large-scale pre-training or expert-grade annotating tools.

Large multilingual language models typically rely on a single vocabulary shared across 100+ languages. As these models have increased in parameter count and depth, vocabulary size has remained largely unchanged. This \textit{vocabulary bottleneck} limits the representational capabilities of multilingual models like XLM-R. In this paper, we introduce a new approach for scaling to very large multilingual vocabularies by de-emphasizing token sharing between languages with little lexical overlap and assigning vocabulary capacity to achieve sufficient coverage for each individual language. Tokenizations using our vocabulary are typically more semantically meaningful and shorter compared to XLM-R. Leveraging this improved vocabulary, we train XLM-V, a multilingual language model with a one million token vocabulary. XLM-V outperforms XLM-R on every task we tested on ranging from natural language inference (XNLI), question answering (MLQA, XQuAD, TyDiQA), to named entity recognition (WikiAnn). XLM-V is particularly effective on low-resource language tasks and outperforms XLM-R by 11.2\% and 5.8\% absolute on MasakhaNER and Americas NLI, respectively.

Distantly supervised relation extraction (DSRE) aims to extract relational facts from texts but suffers from noisy instances. To mitigate the influence of noisy labels, current methods typically use the Multi-Instance-Learning framework to extract relations for each bag. However, these approaches are not capable of extracting relation labels for individual sentences. Several studies have focused on sentence-level DSRE to solve the above problem. These studies primarily aim to develop methods for identifying noisy samples and filtering them out to mitigate the impact of noise. However, discarding noisy samples directly leads to the loss of useful information. To this end, we propose SSLRE, a novel Semi-Supervised-Learning Relation Extraction framework for sentence-level DSRE. We discard only the labels of the noisy samples and utilize these instances without labels as unlabeled samples. Our SSLRE framework utilizes a weighted K-NN graph to select confident samples as labeled data and the rest as unlabeled. We then design a robust semi-supervised learning framework that can efficiently handle remaining label noise present in the labeled dataset, while also making effective use of unlabeled samples. Based on our experiments on two real-world datasets, the SSLRE framework we proposed has achieved significant enhancements in sentence-level relation extraction performance compared to the existing state-of-the-art methods. Moreover, it has also attained a state-of-the-art level of performance in bag-level relation extraction with ONE aggregation strategy.

Large language models (LLMs) can be used to serve as agents to simulate human behaviors, given the powerful ability to understand human instructions and provide high-quality generated texts. Such ability stimulates us to wonder whether LLMs can simulate a person in a higher form than simple human behaviors. Therefore, we aim to train an agent with the profile, experience, and emotional states of a specific person instead of using limited prompts to instruct ChatGPT API. In this work, we introduce Character-LLM that teach LLMs to act as specific people such as Beethoven, Queen Cleopatra, Julius Caesar, etc. Our method focuses on editing profiles as experiences of a certain character and training models to be personal simulacra with these experiences. To assess the effectiveness of our approach, we build a test playground that interviews trained agents and evaluates whether the agents \textit{memorize} their characters and experiences. Experimental results show interesting observations that help build future simulacra of humankind.\footnote{Code and datasets are public at \url{https://github.com/choosewhatulike/trainable-agents}}

When people interpret text, they rely on inferences that go beyond the observed language itself. Inspired by this observation, we introduce a method for the analysis of text that takes implicitly communicated content explicitly into account. We use a large language model to produce sets of propositions that are inferentially related to the text that has been observed, then validate the plausibility of the generated content via human judgments. Incorporating these explicit representations of implicit content proves useful in multiple problem settings that involve the human interpretation of utterances: assessing the similarity of arguments, making sense of a body of opinion data, and modeling legislative behavior. Our results suggest that modeling the meanings behind observed language, rather than the literal text alone, is a valuable direction for NLP and particularly its applications to social science.

Table of contents (ToC) extraction centres on structuring documents in a hierarchical manner. In this paper, we propose a new dataset, ESGDoc, comprising 1,093 ESG annual reports from 563 companies spanning from 2001 to 2022. These reports pose significant challenges due to their diverse structures and extensive length. To address these challenges, we propose a new framework for Toc extraction, consisting of three steps: (1) Constructing an initial tree of text blocks based on reading order and font sizes; (2) Modelling each tree node (or text block) independently by considering its contextual information captured in node-centric subtree; (3) Modifying the original tree by taking appropriate action on each tree node (Keep, Delete, or Move). This construction-modelling-modification (CMM) process offers several benefits. It eliminates the need for pairwise modelling of section headings as in previous approaches, making document segmentation practically feasible. By incorporating structured information, each section heading can leverage both local and long-distance context relevant to itself. Experimental results show that our approach outperforms the previous state-of-the-art baseline with a fraction of running time. Our framework proves its scalability by effectively handling documents of any length.

Controlling chatbot utterance generation with multiple attributes such as personalities, emotions and dialogue acts is a practically useful but under-studied problem. We propose a novel framework called DASC that possesses strong controllability with a weighted decoding paradigm, while improving generation quality with the grounding in an attribute semantics space. Generation with multiple attributes is then intuitively implemented with an interpolation of multiple attribute embeddings, which results in substantial reduction in the model sizes. Experiments show that DASC can achieve high control accuracy in generation task with the simultaneous control of 3 aspects while also producing interesting and reasonably sensible responses, even in an out-of-distribution robustness test.

Concepts play a pivotal role in various human cognitive functions, including learning, reasoning and communication. However, there is very little work on endowing machines with the ability to form and reason with concepts. In particular, state-of-the-art large language models (LLMs) work at the level of tokens, not concepts. In this work, we analyze how well contemporary LLMs capture human concepts and their structure. We then discuss ways to develop concept-aware LLMs, taking place at different stages of the pipeline. We sketch a method for pretraining LLMs using concepts, and also explore the simpler approach that uses the output of existing LLMs. Despite its simplicity, our proof-of-concept is shown to better match human intuition, as well as improve the robustness of predictions. These preliminary results underscore the promise of concept-aware LLMs.