NAACL 2024 Papers

Argument mining has focused so far mainly on the identification, extraction, and formalization of arguments. An important yet unaddressedtask consists in the prediction of the argumentative behavior of stakeholders in a debate. Predicting the argumentative behavior in advance can support foreseeing issues in public policy making or help recognize potential disagreements early on and help to resolve them. In this paper, we consider the novel task of predicting the argumentative behavior of individual stakeholders. We present ARGENST, a framework that relies on a recommender-based architecture to predict the stance and the argumentative main point on a specific controversial topic for a given stakeholder, which is described in terms of a profile including properties related to demographic attributes, religious and political orientation, socio-economic background, etc. We evaluate our approach on the well-known debate.org dataset in terms of accuracy for predicting stance as well as in terms of similarity of the generated arguments to the ground truth arguments using BERTScore. As part of a case study, we show how juries of members representing different stakeholder groups and perspectives can be assembled to simulate the public opinion on a given topic.

Fairness-related assumptions about what constitute appropriate NLG system behaviors range from invariance, where systems are expected to behave identically for social groups, to adaptation, where behaviors should instead vary across them. To illuminate tensions around invariance and adaptation, we conduct five case studies, in which we perturb different types of identity-related language features (names, roles, locations, dialect, and style) in NLG system inputs. Through these cases studies, we examine people’s expectations of system behaviors, and surface potential caveats of these contrasting yet commonly held assumptions. We find that motivations for adaptation include social norms, cultural differences, feature-specific information, and accommodation; in contrast, motivations for invariance include perspectives that favor prescriptivism, view adaptation as unnecessary or too difficult for NLG systems to do appropriately, and are wary of false assumptions. Our findings highlight open challenges around what constitute “fair” or “good” NLG system behaviors.

Modeling evolving knowledge over temporal knowledge graphs (TKGs) has become a heated topic. Various methods have been proposed to forecast links on TKGs. Most of them are embedding-based, where hidden representations are learned to represent knowledge graph (KG) entities and relations based on the observed graph contexts. Although these methods show strong performance on traditional TKG forecasting (TKGF) benchmarks, they face a strong challenge in modeling the unseen zero-shot relations that have no prior graph context. In this paper, we try to mitigate this problem as follows. We first input the text descriptions of KG relations into large language models (LLMs) for generating relation representations, and then introduce them into embedding-based TKGF methods. LLM-empowered representations can capture the semantic information in the relation descriptions. This makes the relations, whether seen or unseen, with similar semantic meanings stay close in the embedding space, enabling TKGF models to recognize zero-shot relations even without any observed graph context. Experimental results show that our approach helps TKGF models to achieve much better performance in forecasting the facts with previously unseen relations, while still maintaining their ability in link forecasting regarding seen relations.

Interventions on model-internal states are fundamental operations in many areas of AI, including model editing, steering, robustness, and interpretability. To facilitate such research, we introduce pyvene, an open-source Python library that supports customizable interventions on a range of different PyTorch modules. pyvene supports complex intervention schemes with an intuitive configuration format, and its interventions can be static or include trainable parameters. We show how pyvene provides a unified and extensible framework for performing interventions on neural models and sharing the intervened upon models with others. We illustrate the power of the library via interpretability analyses using causal abstraction and knowledge localization. We publish our library through Python Package Index (PyPI) and provide code, documentation, and tutorials at ‘https://github.com/stanfordnlp/pyvene‘.

Many pretrained multilingual models exhibit cross-lingual transfer ability, which is often attributed to a learned language-neutral representation during pretraining. However, it remains unclear what factors contribute to the learning of a language-neutral representation, and whether the learned language-neutral representation suffices to facilitate cross-lingual transfer. We propose a synthetic task, Multilingual Othello (mOthello), as a testbed to delve into these two questions. We find that: (1) models trained with naive multilingual pretraining fail to learn a language-neutral representation across all input languages; (2) the introduction of “anchor tokens” (i.e., lexical items that are identical across languages) helps cross-lingual representation alignment; and (3) the learning of a language-neutral representation alone is not sufficient to facilitate cross-lingual transfer. Based on our findings, we propose a novel approach – multilingual pretraining with unified output space – that both induces the learning of language-neutral representation and facilitates cross-lingual transfer.

We introduce mEdIT, a multi-lingual extension to CoEdIT – the recent state-of-the-art text editing models for writing assistance. mEdIT models are trained by fine-tuning multi-lingual large, pre-trained language models (LLMs) via instruction tuning. They are designed to take instructions from the user specifying the attributes of the desired text in the form of natural language instructions, such as “Grammatik korrigieren” (German) or “이 텍스 트를 단순화” (Korean). We build mEdIT by curating data from multiple publicly available human-annotated text editing datasets for three text editing tasks (Grammatical Error Correction (GEC), Text Simplification, and Paraphrasing) across diverse languages belonging to six different language families. We detail the design and training of mEdIT models and demonstrate their strong performance on many multi-lingual text editing benchmarks against other multilingual LLMs. We also find that mEdIT generalizes effectively to new languages over multilingual baselines. We publicly release our data, code, and trained models.

Task-Oriented Parsing (TOP) enables conversational assistants to interpret user commands expressed in natural language, transforming them into structured outputs that combine elements of both natural language and intent/slot tags. Recently, Large Language Models (LLMs) have achieved impressive performance in synthesizing computer programs based on a natural-language prompt, mitigating the gap between natural language and structured programs. Our paper focuses on harnessing the capabilities of LLMs for semantic parsing tasks, addressing the following three key research questions: 1) How can LLMs be effectively utilized for semantic parsing tasks? 2) What defines an effective prompt? and 3) How can LLM overcome the length constraint and streamline prompt design by including all examples as prompts? We introduce k Nearest Neighbor In-Context Learning (kNN-ICL), which simplifies prompt engineering by allowing it to be built on top of any design strategy while providing access to all demo examples. Extensive experiments show that: 1) Simple ICL without kNN search can achieve a comparable performance with strong supervised models on the TOP tasks, and 2) kNN-ICL significantly improves the comprehension of complex requests by seamlessly integrating ICL with a nearest-neighbor approach. Notably, this enhancement is achieved without the need for additional data or specialized prompts.

We introduce an evaluation system designed to compute PARSEVAL measures, offering a viable alternative to evalb commonly used for constituency parsing evaluation. The widely used evalb script has traditionally been employed for evaluating the accuracy of constituency parsing results, albeit with the requirement for consistent tokenization and sentence boundaries. In contrast, our approach, named jp-evalb, is founded on an alignment method. This method aligns sentences and words when discrepancies arise. It aims to overcome several known issues associated with evalb by utilizing the ‘jointly preprocessed (JP)’ alignment-based method. We introduce a more flexible and adaptive framework, ultimately contributing to a more accurate assessment of constituency parsing performance.

Aspect-based sentiment analysis (ABSA) have been extensively studied, but little light has been shed on the quadruple extraction consisting of four fundamental elements: aspects, categories, opinions and sentiments, especially with implicit aspects and opinions. In this paper, we propose a new method iACOS for extracting Implicit Aspects with Categories and Opinions with Sentiments. First, iACOS appends two implicit tokens at the end of a text to capture the context-aware representation of all tokens including implicit aspects and opinions. Second, iACOS develops a sequence labeling model over the context-aware token representation to co-extract explicit and implicit aspects and opinions. Third, iACOS devises a multi-label classifier with a specialized multi-head attention for discovering aspect-opinion pairs and predicting their categories and sentiments simultaneously. Fourth, iACOS leverages informative and adaptive negative examples to jointly train the multi-label classifier and the other two classifiers on categories and sentiments by multi-task learning. Finally, the experimental results show that iACOS significantly outperforms other quadruple extraction baselines according to the F1 score on two public benchmark datasets.

The convergence of text, visual, and audio data is crucial towards human-like artificial intelligence, however the current Vision-Language-Speech landscape is dominated by encoder-only models that lack generative abilities. We propose closing this gap with i-Code V2, one of the first models capable of generating natural language from any combination of Vision, Language, and Speech data. i-Code V2 leverages state-of-the-art single-modality encoders, combining their outputs with a new modality-fusing encoder to project combinations of modalities into a shared representational space. Language tokens are generated from these representations via an autoregressive decoder. i-Code V2 is pretrained end-to-end on a large collection of dual- and single-modality datasets with a novel text completion objective that can be generalized across arbitrary combinations of modalities. i-Code V2 matches or outperforms state-of-the-art single- and dual-modality baselines on 7 multimodal tasks, demonstrating the power of generative multimodal pretraining across a diversity of tasks and signals.

The swift advancement in large language models (LLMs) has heightened the importance of model evaluations. LLMs have acquired a substantial amount of knowledge, and evaluating the knowledge of these LLMs is crucial. To address this, we introduce the ZhuJiu-Knowledge benchmark which carefully considers the following factors: (1) For knowledge scope, we concentrate on three domains: commonsense knowledge, world knowledge, language knowledge, which comes from ATOMIC, Conceptnet, Wikidata, and Wordnet. (2) For data construction, to prevent data contamination, we utilize knowledge derived from corpora and knowledge graphs to formulate novel questions which are ensured not to appear in the training corpus. A multitude of prompts is purposefully devised to mitigate the impact of prompt design on evaluation and to further analyze the LLMs’ sensitivity to various prompts. (3) For evaluation criteria, we propose a novel voting methodology for assessing generative text, aligning the model’s evaluation with human preferences to reduce biases inherent in individual model assessments. We evaluate 14 current mainstream LLMs and conduct a comprehensive discussion and analysis of their results. The ZhuJiu-Knowledge benchmark and open-participation leaderboard are publicly released at http://zhujiu-knowledge.top and we also provide a demo video at https://youtu.be/QJp4qlEHVH8.

Generative linguistic steganography attempts to hide secret messages into covertext. Previous studies have generally focused on the statistical differences between the covertext and stegotext, however, ill-formed stegotext can readily be identified by humans. In this paper, we propose a novel zero-shot approach based on in-context learning for linguistic steganography to achieve better perceptual and statistical imperceptibility. We also design several new metrics and reproducible language evaluations to measure the imperceptibility of the stegotext. Our experimental results indicate that our method produces 1.926× more innocent and intelligible stegotext than any other method.

Cross-lingual transfer can be achieved through two main approaches: zero-shot transfer or machine translation (MT). While the former has been the dominant approach, both have been shown to be competitive. In this work, we compare the current performance and long-term viability of these methods. We leverage lexical gaps to create a multilingual question answering dataset, which provides a difficult domain for evaluation. Both approaches struggle in this setting, though zero-shot transfer performs better, as current MT outputs are not specific enough for the task. Using oracle translation offers the best performance, showing that this approach can perform well long-term, however current MT quality is a bottleneck. We also conduct an exploratory study to see if humans produce translations sufficient for the task with only general instructions. We find this to be true for the majority of translators, but not all. This indicates that while translation has the potential to outperform zero-shot approaches, creating MT models that generate accurate task-specific translations may not be straightforward.

Automated synthesis of zeolite, one of the most important catalysts in chemical industries, holds great significance for attaining economic and environmental benefits. Structural synthesis data extracted through NLP technologies from zeolite experimental procedures can significantly expedite automated synthesis owing to its machine readability. However, the utilization of NLP technologies in information extraction of zeolite synthesis remains restricted due to the lack of annotated datasets. In this paper, we formulate an event extraction task to mine structural synthesis actions from experimental narratives for modular automated synthesis. Furthermore, we introduce ZSEE, a novel dataset containing fine-grained event annotations of zeolite synthesis actions. Our dataset features 16 event types and 13 argument roles which cover all the experimental operational steps of zeolite synthesis. We explore current state-of-the-art event extraction methods on ZSEE, perform error analysis based on the experimental results, and summarize the challenges and corresponding research directions to further facilitate the automated synthesis of zeolites. The code is publicly available at https://github.com/Hi-0317/ZSEE.

Despite recent significant progress, Multi-Object Tracking (MOT) faces limitations such as reliance on prior knowledge and predefined categories and struggles with unseen objects. To address these issues, Generic Multiple Object Tracking (GMOT) has emerged as an alternative approach, requiring less prior information. However, current GMOT methods often rely on initial bounding boxes and struggle to handle variations in factors such as viewpoint, lighting, occlusion, and scale, among others. Our contributions commence with the introduction of the Referring GMOT dataset a collection of videos, each accompanied by detailed textual descriptions of their attributes. Subsequently, we propose Z-GMOT, a cutting-edge tracking solution capable of tracking objects from never-seen categories without the need of initial bounding boxes or predefined categories. Within our Z-GMOT framework, we introduce two novel components: (i) iGLIP, an improved Grounded language-image pretraining, for accurately detecting unseen objects with specific characteristics. (ii) MA-SORT, a novel object association approach that adeptly integrates motion and appearance-based matching strategies to tackle the complex task of tracking objects with high similarity. Our contributions are benchmarked through extensive experiments conducted on the Referring GMOT dataset for GMOT task. Additionally, to assess the generalizability of the proposed Z-GMOT, we conduct ablation studies on the DanceTrack and MOT20 datasets for the MOT task. Our dataset, code, and models are released at: https://fsoft-aic.github.io/Z-GMOT

The versatility of Large Language Models (LLMs) on natural language understanding tasks has made them popular for research in social sciences. To properly understand the properties and innate personas of LLMs, researchers have performed studies that involve using prompts in the form of questions that ask LLMs about particular opinions. In this study, we take a cautionary step back and examine whether the current format of prompting LLMs elicits responses in a consistent and robust manner. We first construct a dataset that contains 693 questions encompassing 39 different instruments of persona measurement on 115 persona axes. Additionally, we design a set of prompts containing minor variations and examine LLMs’ capabilities to generate answers, as well as prompt variations to examine their consistency with respect to content-level variations such as switching the order of response options or negating the statement. Our experiments on 17 different LLMs reveal that even simple perturbations significantly downgrade a model’s question-answering ability, and that most LLMs have low negation consistency. Our results suggest that the currently widespread practice of prompting is insufficient to accurately and reliably capture model perceptions, and we therefore discuss potential alternatives to improve these issues.

In this paper, we introduce a benchmark for evaluating the overall quality of emergent languages using data-driven methods. Specifically, we interpret the notion of the “quality” of an emergent language as its similarity to human language within a deep learning framework. We measure this by using the emergent language as pretraining data for a downstream NLP tasks in human language—the better the downstream performance, the better the emergent language. We implement this benchmark as an easy-to-use Python package that only requires a text file of utterances from the emergent language to be evaluated. Finally, we empirically test the benchmark’s validity using human, synthetic, and emergent language baselines.

Without proper safeguards, large language models will readily follow malicious instructions and generate toxic content. This risk motivates safety efforts such as red-teaming and large-scale feedback learning, which aim to make models both helpful and harmless. However, there is a tension between these two objectives, since harmlessness requires models to refuse to comply with unsafe prompts, and thus not be helpful. Recent anecdotal evidence suggests that some models may have struck a poor balance, so that even clearly safe prompts are refused if they use similar language to unsafe prompts or mention sensitive topics. In this paper, we introduce a new test suite called XSTest to identify such eXaggerated Safety behaviours in a systematic way. XSTest comprises 250 safe prompts across ten prompt types that well-calibrated models should not refuse to comply with, and 200 unsafe prompts as contrasts that models, for most applications, should refuse. We describe XSTest’s creation and composition, and then use the test suite to highlight systematic failure modes in state-of-the-art language models as well as more general challenges in building safer language models.

XNLI is a popular Natural Language Inference (NLI) benchmark widely used to evaluate cross-lingual Natural Language Understanding (NLU) capabilities across languages. In this paper, we expand XNLI to include Basque, a low-resource language that can greatly benefit from transfer-learning approaches. The new dataset, dubbed XNLIeu, has been developed by first machine-translating the English XNLI corpus into Basque, followed by a manual post-edition step. We have conducted a series of experiments using mono- and multilingual LLMs to assess a) the effect of professional post-edition on the MT system; b) the best cross-lingual strategy for NLI in Basque; and c) whether the choice of the best cross-lingual strategy is influenced by the fact that the dataset is built by translation. The results show that post-edition is necessary and that the translate-train cross-lingual strategy obtains better results overall, although the gain is lower when tested in a dataset that has been built natively from scratch. Our code and datasets are publicly available under open licenses.