Computation and Language (cs.CL)

Abstract: Large Language Models (LLMs) have demonstrated impressive performance in various NLP tasks, but they still suffer from challenges such as hallucination and weak numerical reasoning. To overcome these challenges, external tools can be used to enhance LLMs' question-answering abilities. However, current evaluation methods do not distinguish between questions that can be answered using LLMs' internal knowledge and those that require external information through tool use. To address this issue, we introduce a new dataset called ToolQA, which is designed to faithfully evaluate LLMs' ability to use external tools for question answering. Our development of ToolQA involved a scalable, automated process for dataset curation, along with 13 specialized tools designed for interaction with external knowledge in order to answer questions. Importantly, we strive to minimize the overlap between our benchmark data and LLMs' pre-training data, enabling a more precise evaluation of LLMs' tool-use reasoning abilities. We conducted an in-depth diagnosis of existing tool-use LLMs to highlight their strengths, weaknesses, and potential improvements. Our findings set a new benchmark for evaluating LLMs and suggest new directions for future advancements. Our data and code are freely available to the broader scientific community on GitHub.

Abstract: Knowledge graphs (KGs), containing many entity-relation-entity triples, provide rich information for downstream applications. Although extracting triples from unstructured texts has been widely explored, most of them require a large number of labeled instances. The performance will drop dramatically when only few labeled data are available. To tackle this problem, we propose the Mutually Guided Few-shot learning framework for Relational Triple Extraction (MG-FTE). Specifically, our method consists of an entity-guided relation proto-decoder to classify the relations firstly and a relation-guided entity proto-decoder to extract entities based on the classified relations. To draw the connection between entity and relation, we design a proto-level fusion module to boost the performance of both entity extraction and relation classification. Moreover, a new cross-domain few-shot triple extraction task is introduced. Extensive experiments show that our method outperforms many state-of-the-art methods by 12.6 F1 score on FewRel 1.0 (single-domain) and 20.5 F1 score on FewRel 2.0 (cross-domain).

Abstract: Text summarization is a fundamental task in natural language processing that aims to condense large amounts of textual information into concise and coherent summaries. With the exponential growth of content and the need to extract key information efficiently, text summarization has gained significant attention in recent years. In this study, LSTM and pre-trained T5, Pegasus, BART and BART-Large model performances were evaluated on the open source dataset (Xsum, CNN/Daily Mail, Amazon Fine Food Review and News Summary) and the prepared resume dataset. This resume dataset consists of many information such as language, education, experience, personal information, skills, and this data includes 75 resumes. The primary objective of this research was to classify resume text. Various techniques such as LSTM, pre-trained models, and fine-tuned models were assessed using a dataset of resumes. The BART-Large model fine-tuned with the resume dataset gave the best performance.

Abstract: The high volume of published chemical patents and the importance of a timely acquisition of their information gives rise to automating information extraction from chemical patents. Anaphora resolution is an important component of comprehensive information extraction, and is critical for extracting reactions. In chemical patents, there are five anaphoric relations of interest: co-reference, transformed, reaction associated, work up, and contained. Our goal is to investigate how the performance of anaphora resolution models for reaction texts in chemical patents differs in a noise-free and noisy environment and to what extent we can improve the robustness against noise of the model.

Abstract: Retrieval-augmented language models (LMs) have received much attention recently. However, typically the retriever is not trained jointly as a native component of the LM, but added to an already-pretrained LM, which limits the ability of the LM and the retriever to adapt to one another. In this work, we propose the Retrieval-Pretrained Transformer (RPT), an architecture and training procedure for jointly training a retrieval-augmented LM from scratch for the task of modeling long texts. Given a recently generated text chunk in a long document, the LM computes query representations, which are then used to retrieve earlier chunks in the document, located potentially tens of thousands of tokens before. Information from retrieved chunks is fused into the LM representations to predict the next target chunk. We train the retriever component with a semantic objective, where the goal is to retrieve chunks that increase the probability of the next chunk, according to a reference LM. We evaluate RPT on four long-range language modeling tasks, spanning books, code, and mathematical writing, and demonstrate that RPT improves retrieval quality and subsequently perplexity across the board compared to strong baselines.

Abstract: Image captioning aims to describe visual content in natural language. As 'a picture is worth a thousand words', there could be various correct descriptions for an image. However, with maximum likelihood estimation as the training objective, the captioning model is penalized whenever its prediction mismatches with the label. For instance, when the model predicts a word expressing richer semantics than the label, it will be penalized and optimized to prefer more concise expressions, referred to as conciseness optimization. In contrast, predictions that are more concise than labels lead to richness optimization. Such conflicting optimization directions could eventually result in the model generating general descriptions. In this work, we introduce Semipermeable MaxImum Likelihood Estimation (SMILE), which allows richness optimization while blocking conciseness optimization, thus encouraging the model to generate longer captions with more details. Extensive experiments on two mainstream image captioning datasets MSCOCO and Flickr30K demonstrate that SMILE significantly enhances the descriptiveness of generated captions. We further provide in-depth investigations to facilitate a better understanding of how SMILE works.

Abstract: Abstract Meaning Representation (AMR) is a Semantic Parsing formalism that aims at providing a semantic graph abstraction representing a given text. Current approaches are based on autoregressive language models such as BART or T5, fine-tuned through Teacher Forcing to obtain a linearized version of the AMR graph from a sentence. In this paper, we present LeakDistill, a model and method that explores a modification to the Transformer architecture, using structural adapters to explicitly incorporate graph information into the learned representations and improve AMR parsing performance. Our experiments show how, by employing word-to-node alignment to embed graph structural information into the encoder at training time, we can obtain state-of-the-art AMR parsing through self-knowledge distillation, even without the use of additional data. We release the code at \url{http://www.github.com/sapienzanlp/LeakDistill}.

Abstract: Transformer-based language models have achieved impressive success in various natural language processing tasks due to their ability to capture complex dependencies and contextual information using self-attention mechanisms. However, they are not without limitations. These limitations include hallucinations, where they produce incorrect outputs with high confidence, and alignment issues, where they generate unhelpful and unsafe outputs for human users. These limitations stem from the absence of implicit and missing context in the data alone. To address this, researchers have explored augmenting these models with external knowledge from knowledge graphs to provide the necessary additional context. However, the ad-hoc nature of existing methods makes it difficult to properly analyze the effects of knowledge infusion on the many moving parts or components of a transformer. This paper introduces a systematic method for infusing knowledge into different components of a transformer-based model. A modular framework is proposed to identify specific components within the transformer architecture, such as the self-attention mechanism, encoder layers, or the input embedding layer, where knowledge infusion can be applied. Additionally, extensive experiments are conducted on the General Language Understanding Evaluation (GLUE) benchmark tasks, and the findings are reported. This systematic approach aims to facilitate more principled approaches to incorporating knowledge into language model architectures.

Abstract: Natural language (NL) feedback contains rich information about the user experience. Existing studies focus on an instance-level approach, where feedback is used to refine specific examples, disregarding its system-wide application. This paper proposes a general framework for unlocking the system-level use of NL feedback. We show how to use feedback to formalize system-level design decisions in a human-in-the-loop-process -- in order to produce better models. In particular this is done through: (i) metric design for tasks; and (ii) language model prompt design for refining model responses. We conduct two case studies of this approach for improving search query generation and dialog response generation, demonstrating the effectiveness of the use of system-level feedback. We show the combination of system-level feedback and instance-level feedback brings further gains, and that human written instance-level feedback results in more grounded refinements than GPT-3.5 written ones, underlying the importance of human feedback for building systems.

Abstract: With the rise of Large Language Models (LLMs) and their ubiquitous deployment in diverse domains, measuring language model behavior on realistic data is imperative. For example, a company deploying a client-facing chatbot must ensure that the model will not respond to client requests with profanity. Current evaluations approach this problem using small, domain-specific datasets with human-curated labels. These evaluation sets are often sampled from a narrow and simplified distribution, and data sources can unknowingly be leaked into the training set which can lead to misleading evaluations. To bypass these drawbacks, we propose a framework for self-supervised evaluation of LLMs by analyzing their sensitivity or invariance to transformations on the input text. Self-supervised evaluation can directly monitor LLM behavior on datasets collected in the wild or streamed during live model deployment. We demonstrate self-supervised evaluation strategies for measuring closed-book knowledge, toxicity, and long-range context dependence, in addition to sensitivity to grammatical structure and tokenization errors. When comparisons to similar human-labeled benchmarks are available, we find strong correlations between self-supervised and human-supervised evaluations. The self-supervised paradigm complements current evaluation strategies that rely on labeled data.