ACL 2024 论文列表

Selecting which claims to check is a time-consuming task for human fact-checkers, especially from documents consisting of multiple sentences and containing multiple claims. However, existing claim extraction approaches focus more on identifying and extracting claims from individual sentences, e.g., identifying whether a sentence contains a claim or the exact boundaries of the claim within a sentence. In this paper, we propose a method for document-level claim extraction for fact-checking, which aims to extract check-worthy claims from documents and decontextualise them so that they can be understood out of context. Specifically, we first recast claim extraction as extractive summarization in order to identify central sentences from documents, then rewrite them to include necessary context from the originating document through sentence decontextualisation. Evaluation with both automatic metrics and a fact-checking professional shows that our method is able to extract check-worthy claims from documents at a higher rate than previous work, while also improving evidence retrieval.

Despite the fact that document-level machine translation has inherent advantages over sentence-level machine translation due to additional information available to a model from document context, most translation systems continue to operate at a sentence level. This is primarily due to the severe lack of publicly available large-scale parallel corpora at the document level. We release a large-scale open parallel corpus with document context extracted from ParaCrawl in five language pairs, along with code to compile document-level datasets for any language pair supported by ParaCrawl. We train context-aware models on these datasets and find improvements in terms of overall translation quality and targeted document-level phenomena. We also analyse how much long-range information is useful to model some of these discourse phenomena and find models are able to utilise context from several preceding sentences.

Digital documents, such as PDFs, are vital in business workflows, enabling communication, documentation, and collaboration. Handling PDFs can involve navigating complex workflows and numerous tools (e.g., comprehension, annotation, editing), which can be tedious and time-consuming for users. We introduce DocPilot, an AI-assisted document workflow Copilot system capable of understanding user intent and executing tasks accordingly to help users streamline their workflows. DocPilot undertakes intelligent orchestration of various tools through LLM prompting in four steps: (1) Task plan generation, (2) Task plan verification and self-correction, (3) Multi-turn User Feedback, and (4) Task Plan Execution via Code Generation and Error log-based Code Self-Revision. The primary goal of this system is to free the user from the intricacies of document editing, enabling them to focus on the creative aspects and enrich their document management experience.

Recent LLMs have demonstrated remarkable performance in solving exam-like math word problems. However, the degree to which these numerical reasoning skills are effective in real-world scenarios, particularly in expert domains, is still largely unexplored. This paper introduces DocMath-Eval, a comprehensive benchmark specifically designed to evaluate the numerical reasoning capabilities of LLMs in the context of understanding and analyzing specialized documents containing both text and tables. We conduct an extensive evaluation of 48 LLMs with Chain-of-Thought and Program-of-Thought prompting methods, aiming to comprehensively assess the capabilities and limitations of existing LLMs in DocMath-Eval. We found that even the current best-performing system (i.e., GPT-4o) still significantly lags behind human experts in solving complex numerical reasoning problems grounded in long contexts. We believe that DocMath-Eval can serve as a valuable benchmark for evaluating LLMs' capabilities in solving challenging numerical reasoning problems within expert domains.

Medical text generation aims to assist with administrative work and highlight salient information to support decision-making.To reflect the specific requirements of medical text, in this paper, we propose a set of metrics to evaluate the completeness, conciseness, and attribution of the generated text at a fine-grained level. The metrics can be computed by various types of evaluators including instruction-following (both proprietary and open-source) and supervised entailment models. We demonstrate the effectiveness of the resulting framework, DocLens, with three evaluators on three tasks: clinical note generation, radiology report summarization, and patient question summarization. A comprehensive human study shows that DocLens exhibits substantially higher agreement with the judgments of medical experts than existing metrics. The results also highlight the need to improve open-source evaluators and suggest potential directions. We released the code at https://github.com/yiqingxyq/DocLens.

Enterprise documents such as forms, receipts, reports, and other such records, often carry rich semantics at the intersection of textual and spatial modalities. The visual cues offered by their complex layouts play a crucial role in comprehending these documents effectively. In this paper, we present DocLLM, a lightweight extension to traditional large language models (LLMs) for reasoning over visual documents, taking into account both textual semantics and spatial layout. Our model differs from existing multimodal LLMs by avoiding expensive image encoders and focuses exclusively on bounding box information to incorporate the spatial layout structure. Specifically, the cross-alignment between text and spatial modalities is captured by decomposing the attention mechanism in classical transformers to a set of disentangled matrices. Furthermore, we devise a pre-training objective that learns to infill text segments. This approach allows us to address irregular layouts and heterogeneous content frequently encountered in visual documents. The pre-trained model is fine-tuned using a large-scale instruction dataset, covering four core document intelligence tasks. We demonstrate that our solution outperforms SotA LLMs on 14 out of 16 datasets across all tasks, and generalizes well to 4 out of 5 previously unseen datasets.

For large language models (LLMs) to be effective in the financial domain – where each decision can have a significant impact – it is necessary to investigate realistic tasks and data. Financial professionals often interact with documents spanning hundreds of pages, but most financial research datasets only deal with short excerpts from these documents. To address this, we introduce a long-document financial QA task. We augment 7,437 questions from the existing FinQA dataset with full-document context, extending the average context length from under 700 words in FinQA to 123k words in DocFinQA. We conduct extensive experiments over retrieval-based QA pipelines and long-context language models. Based on our experiments, DocFinQA proves a significant challenge for even state-of-the-art systems. We also provide a case study on a subset of the longest documents in DocFinQA and find that models particularly struggle with these documents. Addressing these challenges may have a wide-reaching impact across applications where specificity and long-range contexts are critical, like gene sequences and legal document contract analysis. DocFinQA dataset is publicly accessible.

Fine-tuning large-scale pre-trained models is inherently a resource-intensive task. While it can enhance the capabilities of the model, it also incurs substantial computational costs, posing challenges to the practical application of downstream tasks. Existing parameter-efficient fine-tuning (PEFT) methods such as Low-Rank Adaptation (LoRA) rely on a bypass framework that ignores the differential parameter budget requirements across weight matrices, which may lead to suboptimal fine-tuning outcomes. To address this issue, we introduce the Dynamic Low-Rank Adaptation (DoRA) method. DoRA decomposes high-rank LoRA layers into structured single-rank components, allowing for dynamic pruning of parameter budget based on their importance to specific tasks during training, which makes the most of the limited parameter budget. Experimental results demonstrate that DoRA can achieve competitive performance compared with LoRA and full model fine-tuning, and outperform various strong baselines with the same storage parameter budget. Our code is available at [github](https://github.com/MIkumikumi0116/DoRA)

Recent advances in LLMs have sparked a debate on whether they understand text. In this position paper, we argue that opponents in this debate hold different definitions for understanding, and particularly differ in their view on the role of consciousness. To substantiate this claim, we propose a thought experiment involving an open-source chatbot Z which excels on every possible benchmark, seemingly without subjective experience. We ask whether Z is capable of understanding, and show that different schools of thought within seminal AI research seem to answer this question differently, uncovering their terminological disagreement. Moving forward, we propose two distinct working definitions for understanding which explicitly acknowledge the question of consciousness, and draw connections with a rich literature in philosophy, psychology and neuroscience.

In everyday language use, speakers frequently utter and interpret sentences that are semantically underspecified, namely, whose content is insufficient to fully convey their message or interpret them univocally. For example, to interpret the underspecified sentence “Don’t spend too much”, which leaves implicit what (not) to spend, additional linguistic context or outside knowledge is needed. In this work, we propose a novel Dataset of semantically Underspecified Sentences grouped by Type (DUST) and use it to study whether pre-trained language models (LMs) correctly identify and interpret underspecified sentences. We find that newer LMs are reasonably able to identify underspecified sentences when explicitly prompted. However, interpreting them correctly is much harder for any LMs. Our experiments show that when interpreting underspecified sentences, LMs exhibit little uncertainty, contrary to what theoretical accounts of underspecification would predict. Overall, our study reveals limitations in current models’ processing of sentence semantics and highlights the importance of using naturalistic data and communicative scenarios when evaluating LMs’ language capabilities.

We ask whether multilingual language models trained on unbalanced, English-dominated corpora use English as an internal pivot language—-a question of key importance for understanding how language models function and the origins of linguistic bias. Focusing on the Llama-2 family of transformer models, our study is based on carefully constructed non-English prompts with a unique correct single-token continuation. From layer to layer, transformers gradually map an input embedding of the final prompt token to an output embedding from which next-token probabilities are computed. Tracking intermediate embeddings through their high-dimensional space reveals three distinct phases, whereby intermediate embeddings (1) start far away from output token embeddings; (2) already in middle layers allow for decoding a semantically correct next token, but giving higher probability to its version in English than in the input language; (3) move into an input-language-specific region of the embedding space. We cast these results into a conceptual model where the three phases operate in ”input space”, ”concept space”, and ”output space”, respectively. Crucially, our evidence suggests that the abstract ”concept space” lies closer to English than to other input languages, which may have important consequences regarding the biases embodied by multilingual language models.

The evaluation of the problem-solving capability under incomplete information scenarios of Large Language Models (LLMs) is increasingly important, encompassing capabilities such as questioning, knowledge search, error detection, and path planning. Current research mainly focus on LLMs’ problem-solving capability such as “Twenty Questions”.However, these kinds of games do not require recognizing misleading cues which are necessary in the incomplete information scenario.Moreover, the existing game such as “Who is undercover” are highly subjective, making it challenging for evaluation.Therefore, in this paper, we introduce a novel game named BrainKing based on the “Who is undercover” and “Twenty Questions” for evaluating LLM capabilities under incomplete information scenarios. It requires LLMs to identify target entities with limited yes-or-no questions and potential misleading answers. By setting up easy, medium, and hard difficulty modes, we comprehensively assess the performance of LLMs across various aspects. Our results reveal the capabilities and limitations of LLMs in BrainKing, providing significant insights of LLM problem-solving levels.

We study whether Large Language Models (LLMs) latently perform multi-hop reasoning with complex prompts such as “The mother of the singer of ‘Superstition’ is”. We look for evidence of a latent reasoning pathway where an LLM (1) latently identifies “the singer of ‘Superstition’” as Stevie Wonder, the bridge entity, and (2) uses its knowledge of Stevie Wonder’s mother to complete the prompt. We analyze these two hops individually and consider their co-occurrence as indicative of latent multi-hop reasoning. For the first hop, we test if changing the prompt to indirectly mention the bridge entity instead of any other entity increases the LLM’s internal recall of the bridge entity. For the second hop, we test if increasing this recall causes the LLM to better utilize what it knows about the bridge entity. We find strong evidence of latent multi-hop reasoning for the prompts of certain relation types, with the reasoning pathway used in more than 80% of the prompts. However, the utilization is highly contextual, varying across different types of prompts. Also, on average, the evidence for the second hop and the full multi-hop traversal is rather moderate and only substantial for the first hop. Moreover, we find a clear scaling trend with increasing model size for the first hop of reasoning but not for the second hop. Our experimental findings suggest potential challenges and opportunities for future development and applications of LLMs.

We examine whether large language models (LLMs) exhibit race- and gender-based name discrimination in hiring decisions, similar to classic findings in the social sciences (Bertrand and Mullainathan, 2004). We design a series of templatic prompts to LLMs to write an email to a named job applicant informing them of a hiring decision. By manipulating the applicant’s first name, we measure the effect of perceived race, ethnicity, and gender on the probability that the LLM generates an acceptance or rejection email. We find that the hiring decisions of LLMs in many settings are more likely to favor White applicants over Hispanic applicants. In aggregate, the groups with the highest and lowest acceptance rates respectively are masculine White names and masculine Hispanic names. However, the comparative acceptance rates by group vary under different templatic settings, suggesting that LLMs’ race- and gender-sensitivity may be idiosyncratic and prompt-sensitive.

We explore which linguistic factors—at the sentence and token level—play an important role in influencing language model predictions, and investigate whether these are reflective of results found in humans and human corpora (Gries and Kootstra, 2017). We make use of the structural priming paradigm—where recent exposure to a structure facilitates processing of the same structure—to investigate where priming effects manifest, and what factors predict them. We find these effects can be explained via the inverse frequency effect found in human priming, where rarer elements within a prime increase priming effects, as well as lexical dependence between prime and target. Our results provide an important piece in the puzzle of understanding how properties within their context affect structural prediction in language models.

Advances in large vision-language models (LVLMs) have led to significant progress in generating natural language descriptions for visual contents. These powerful models are known for producing texts that are factually inconsistent with the visual input. While some efforts mitigate such inconsistencies in natural image captioning, the factuality of generated captions for structured visuals, such as charts, has not received as much scrutiny. This work introduces a comprehensive typology of factual errors in generated chart captions. A large-scale human annotation effort provides insight into the error patterns in captions generated by various models, ultimately forming the foundation of a dataset, CHOCOLATE. Our analysis reveals that even advanced models like GPT-4V frequently produce captions laced with factual inaccuracies. To combat this, we establish the task of Chart Caption Factual Error Correction and introduce CHARTVE, a visual entailment model that outperforms current LVLMs in evaluating caption factuality. Furthermore, we propose C2TFEC, an interpretable two-stage framework that excels at correcting factual errors. This work inaugurates a new domain in factual error correction for chart captions, presenting a novel evaluation metric, and demonstrating an effective approach to ensuring the factuality of generated chart captions. The code and data as well as the continuously updated benchmark can be found at: https://khuangaf.github.io/CHOCOLATE/.

We design probes trained on the internal representations of a transformer language model to predict its hallucinatory behavior on three grounded generation tasks. To train the probes, we annotate for span-level hallucination on both sampled (organic) and manually edited (synthetic) reference outputs. Our probes are narrowly trained and we find that they are sensitive to their training domain: they generalize poorly from one task to another or from synthetic to organic hallucinations. However, on in-domain data, they can reliably detect hallucinations at many transformer layers, achieving 95% of their peak performance as early as layer 4. Here, probing proves accurate for evaluating hallucination, outperforming several contemporary baselines and even surpassing an expert human annotator in response-level detection F1. Similarly, on span-level labeling, probes are on par or better than the expert annotator on two out of three generation tasks. Overall, we find that probing is a feasible and efficient alternative to language model hallucination evaluation when model states are available.

Videos are more informative than images becausethey capture the dynamics of the scene.By representing motion in videos, we can capturedynamic activities. In this work, we introduceGPT-4 generated motion descriptions thatcapture fine-grained motion descriptions of activitiesand apply them to three action datasets.We evaluated several video-text models on thetask of retrieval of motion descriptions. Wefound that they fall far behind human expertperformance on two action datasets, raisingthe question of whether video-text models understandmotion in videos. To address it, weintroduce a method of improving motion understandingin video-text models by utilizingmotion descriptions. This method proves tobe effective on two action datasets for the motiondescription retrieval task. The results drawattention to the need for quality captions involvingfine-grained motion information in existingdatasets and demonstrate the effectiveness ofthe proposed pipeline in understanding finegrainedmotion during video-text retrieval.

Functional Distributional Semantics (FDS) models the meaning of words by truth-conditional functions. This provides a natural representation for hypernymy but no guarantee that it can be learnt when FDS models are trained on a corpus. In this paper, we probe into FDS models and study the representations learnt, drawing connections between quantifications, the Distributional Inclusion Hypothesis (DIH), and the variational-autoencoding objective of FDS model training. Using synthetic data sets, we reveal that FDS models learn hypernymy on a restricted class of corpus that strictly follows the DIH. We further introduce a training objective that both enables hypernymy learning under the reverse of the DIH and improves hypernymy detection from real corpora.