Thailand

An automatic citation generation system aims to concisely and accurately describe the relationship between two scientific articles. To do so, such a system must ground its outputs to the content of the cited paper to avoid non-factual hallucinations. Due to the length of scientific documents, existing abstractive approaches have conditioned only on cited paper abstracts. We demonstrate empirically that the abstract is not always the most appropriate input for citation generation and that models trained in this way learn to hallucinate. We propose to condition instead on the cited text span (CTS) as an alternative to the abstract. Because manual CTS annotation is extremely time- and labor-intensive, we experiment with distant labeling of candidate CTS sentences, achieving sufficiently strong performance to substitute for expensive human annotations in model training, and we propose a human-in-the-loop, keyword-based CTS retrieval approach that makes generating citation texts grounded in the full text of cited

ACL 2024

Cited Text Spans for Scientific Citation Text Generation

cited text span; citation text generation

workshop paper

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

Citation generation aims to generate a citation sentence that refers to a chosen paper in the context of a manuscript. However, a rigid citation generation process is at odds with an author's desire to control specific attributes, such as 1) the citation intent, e.g., either introducing background information or comparing results, and 2) keywords that should appear in the citation text. To provide these degrees of controllability during citation generation, we propose to integrate the manuscript context, the context of the referenced paper, and the desired control attributes into a structured template and use it to fine-tune a language model (LM) via next-token prediction. We then utilize Proximal Policy Optimization to directly optimize the LM in favor of a high score of our proposed controllability metric. The proposed workflow harmoniously combines citation attribute suggestion and conditional citation generation into one LM, allowing for better user control.

Controllable Citation Sentence Generation with Language Models

The inception of a research agenda typically commences with the creation of a comprehensive research proposal. The efficacy of the proposal often hinges on its ability to connect with the existing scientific literature that supports its ideas. To effectively assess the relevance of existing articles to a research proposal, it is imperative to categorize these articles into high-level thematic groups, referred to as topics, that align with the proposal. This paper introduces a novel task of aligning scientific articles, relevant to a proposal, with researcher-provided proposal topics. Additionally, we construct a dataset to serve as a benchmark for this task. We establish human and Large Language Model (LLM) baselines and propose a novel three-stage approach to address this challenge. We synthesize and use pseudo-labels that map proposal topics to text spans from cited articles to train Language Models (LMs) for two purposes: (i) as a retriever, to extract relevant text spans from cited articles for each topic, and (ii) as a classifier, to categorize the articles into the proposal topics. Our retriever-classifier pipeline, which employs very small open-source LMs fine-tuned with our constructed dataset, achieves results comparable to a vanilla paid LLM-based classifier, demonstrating its efficacy. However, a notable gap of 23.57 F1 score between our approach and the human baseline highlights the complexity of this task and emphasizes the need for further research.

Beyond Retrieval: Topic-based Alignment of Scientific Papers to Research Proposal

When examining reviews of research papers, we can distinguish between two hypothetical referees: the maximally lenient referee who accepts any paper with a vacuous review and the maximally strict one who rejects any paper with an overly pedantic review. Clearly, both are of no practical value. Our interest is in a referee who makes a balanced judgement and provides a review abiding by the guidelines. In this paper, we present a case study of automatic correction of an existing machine-generated or human review. The AutoRef system implements an iterative approach that progressively "refines'' a review by attempting to make it more compliant with pre-defined requirements of a "good'' review. It implements the following steps: (1) Translate the review requirements into a specification in natural language, of "yes/no'' questions; (2) Given a (paper, review) pair, extract answers to the questions; (3) Use the results in (2) to generate a new review; and (4) Return to Step (2) with the paper and the new review. Here, (2) and (3) are implemented by large language model (LLM) based agents. We present a case study using papers and reviews made available for the International Conference on Learning Representations (ICLR). Our initial empirical results suggest that AutoRef progressively improves the compliance of the generated reviews to the specification. Currently designed specification makes AutoRef progressively generate reviews which are stricter, making the decisions more inclined towards "rejections''. This demonstrates the applicability of AutoRef for: (1) The progressive correction of overly lenient reviews, being useful for referees and meta-reviewers; and (2) The generation of progressively stricter reviews for a paper, starting from a vacuous review ("Great paper. Accept.''), facilitating authors when trying to assess weaknesses in their papers.

AutoRef: Generating Refinements of Reviews Given Guidelines

This paper describes a system designed to distinguish between AI-generated and human-written scientific excerpts in the DAGPap24 competition hosted within the Fourth Workshop on Scientific Document Processing. In this competition the task is to find artificially generated token-level text fragments in documents of a scientific domain. Our work focuses on the use of a multi-task learning architecture with two heads. The application of this approach is justified by the specificity of the task, where class spans are continuous over several hundred characters. We considered different encoder variations to obtain a state vector for each token in the sequence, as well as a variation in splitting fragments into tokens to further feed into the input of a transform-based encoder. This approach allows us to achieve a 9% quality improvement relative to the baseline solution score on the development set (from 0.86 to 0.95) using the average macro $F_{1}$-score, as well as a score of 0.96 on a closed test part of the dataset from the competition.

Multi-head Span-based Detector for AI-generated Fragments in Scientific Papers

Taking note of the current challenges of the peer review system, this paper inventories the research tasks for analysing and possibly automating parts of the reviewing process, like matching submissions with a reviewer's domain of expertise. For each of these tasks we list their associated datasets, analysing their quality in terms of available documentation of creation and use. Building up on this, we give a set of recommendations to take into account when collecting and releasing data.

An Analysis of Tasks and Datasets in Peer Reviewing

To help readers understand the novelty and the research context, an excellent related work section is structured (i.e., the section consists of paragraphs determined by categorizing papers into several topics) and includes descriptions of novelty. However, previous studies viewed related work generation as multi-document summarization, and the structure and novelty statement are ignored in such studies. In this paper, we redefine the related work generation task as summarization with structure (i.e., multiple paragraphs with citation) and novelty statement. For this task, we propose a quality-oriented dataset and evaluation metrics. Experiments evaluated the state-of-the-art language models on our tasks, and we confirmed the issues with the current models and the validity of the evaluation indicators.

Toward Related Work Generation with Structure and Novelty Statement

Despite the dramatic progress in Large Language Model (LLM) development, LLMs often provide seemingly plausible but not factual information, often referred to as hallucinations. Retrieval-augmented LLMs provide a non-parametric approach to solve these issues by retrieving relevant information from external data sources and augment the training process. These models help to trace evidence from an externally provided knowledge base allowing the model predictions to be better interpreted and verified. In this work, we critically evaluate these models in their ability to perform in scientific document reasoning tasks. To this end, we tuned multiple such model variants with science-focused instructions and evaluated them on a scientific document reasoning benchmark for the usefulness of the retrieved document passages. Our findings suggest that models justify predictions in science tasks with fabricated evidence and leveraging scientific corpus as pretraining data does not alleviate the risk of evidence fabrication.

Evaluating the Effectiveness of Retrieval-Augmented Large Language Models in Scientific Document Reasoning

Most task-oriented dialogue (TOD) benchmarks assume users that know exactly how to use the system by constraining the user behaviors within the system's capabilities via strict user goals, namely "user familiarity" bias. This data bias deepens when it combines with data-driven TOD systems, as it is impossible to fathom the effect of it with existing static evaluations. Hence, we conduct an interactive user study to unveil how vulnerable TOD systems are against realistic scenarios. In particular, we compare users with 1) detailed goal instructions that conform to the system boundaries (closed-goal) and 2) vague goal instructions that are often unsupported but realistic (open-goal). Our study reveals that conversations in open-goal settings lead to catastrophic failures of the system, in which 92% of the dialogues had significant issues. Moreover, we conduct a thorough analysis to identify distinctive features between the two settings through error annotation. From this, we discover a novel "pretending" behavior, in which the system pretends to handle the user requests even though they are beyond the system's capabilities. We discuss its characteristics and toxicity while showing recent large language models can also suffer from this behavior.

Revealing User Familiarity Bias in Task-Oriented Dialogue via Interactive Evaluation

Engineering Conversational Search Systems: A Review of Applications, Architectures, and Functional Components

Although humans inherently have diverse values, current large language model (LLM) alignment methods often assume that aligning LLMs with the general public’s preferences is optimal. A major challenge in adopting a more individualized approach to LLM alignment is scalability, as it involves repeatedly acquiring preference data and training new reward models and LLMs for each individual’s preferences. To address these challenges, we propose a new paradigm where users specify what they value most within the system messages steering the LLM’s generation behavior to better align with the user's intentions. However, a naive adaptation of such an approach is non-trivial since LLMs are typically trained on a uniform system messages (e.g., "You are a helpful assistant") which limits their ability to generalize to diverse, unseen system messages. To improve this generalization, we create the Multifaceted Collection, a preference dataset with 192k combinations of values beyond generic helpfulness and harmlessness, spanning 65k user instructions. Using this dataset, we train a 7B LLM called Janus and test it on 921 prompts from 5 benchmarks (AlpacaEval 2.0, FLASK, Koala, MT-Bench, and Self-Instruct) by adding various unseen system messages that reflect user preferences. Janus demonstrates performance that is as good as or better than Mistral 7B Instruct v0.2, GPT-3.5 Turbo, and GPT-4, 75.2%, 72.4%, and 66.4% of the time, respectively. Unexpectedly, on three benchmarks focused on response helpfulness (AlpacaEval 2.0, MT-Bench, Arena Hard Auto v0.1), Janus also outperforms LLaMA 3 8B Instruct by a +4.0%, +0.1%, +3.0% margin, underscoring that training with a vast array of system messages could also enhance alignment to the general public's preference as well. Our code, dataset, benchmark, and models are available at https://anonymous.4open.science/r/janus.

Cited Text Spans for Scientific Citation Text Generation

Downloads

Next from ACL 2024

Controllable Citation Sentence Generation with Language Models

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES