Computation and Language (cs.CL)

Abstract: Low-rank adaptations (LoRA) are often employed to fine-tune large language models (LLMs) for new tasks. This paper investigates LoRA composability for cross-task generalization and introduces LoraHub, a strategic framework devised for the purposive assembly of LoRA modules trained on diverse given tasks, with the objective of achieving adaptable performance on unseen tasks. With just a few examples from a novel task, LoraHub enables the fluid combination of multiple LoRA modules, eradicating the need for human expertise. Notably, the composition requires neither additional model parameters nor gradients. Our empirical results, derived from the Big-Bench Hard (BBH) benchmark, suggest that LoraHub can effectively mimic the performance of in-context learning in few-shot scenarios, excluding the necessity of in-context examples alongside each inference input. A significant contribution of our research is the fostering of a community for LoRA, where users can share their trained LoRA modules, thereby facilitating their application to new tasks. We anticipate this resource will widen access to and spur advancements in general intelligence as well as LLMs in production. Code will be available at https://github.com/sail-sg/lorahub.

Abstract: Chain-of-thought (CoT) prompting has been shown to empirically improve the accuracy of large language models (LLMs) on various question answering tasks. While understanding why CoT prompting is effective is crucial to ensuring that this phenomenon is a consequence of desired model behavior, little work has addressed this; nonetheless, such an understanding is a critical prerequisite for responsible model deployment. We address this question by leveraging gradient-based feature attribution methods which produce saliency scores that capture the influence of input tokens on model output. Specifically, we probe several open-source LLMs to investigate whether CoT prompting affects the relative importances they assign to particular input tokens. Our results indicate that while CoT prompting does not increase the magnitude of saliency scores attributed to semantically relevant tokens in the prompt compared to standard few-shot prompting, it increases the robustness of saliency scores to question perturbations and variations in model output.

Abstract: Recently, with the emergence of numerous Large Language Models (LLMs), the implementation of AI has entered a new era. Irrespective of these models' own capacity and structure, there is a growing demand for LLMs to possess enhanced comprehension of longer and more complex contexts with relatively smaller sizes. Models often encounter an upper limit when processing sequences of sentences that extend beyond their comprehension capacity and result in off-topic or even chaotic responses. While several recent works attempt to address this issue in various ways, they rarely focus on "why models are unable to compensate or strengthen their capabilities on their own". In this paper, we thoroughly investigate the nature of information transfer within LLMs and propose a novel technique called Attention Transition. This technique empowers models to achieve longer and better context comprehension with minimal additional training or impact on generation fluency. Our experiments are conducted in XSum and achieve substantial improvement compared with the original generation results.

Abstract: It is a well-known fact that current AI-based language technology -- language models, machine translation systems, multilingual dictionaries and corpora -- focuses on the world's 2-3% most widely spoken languages. Recent research efforts have attempted to expand the coverage of AI technology to `under-resourced languages.' The goal of our paper is to bring attention to a phenomenon that we call linguistic bias: multilingual language processing systems often exhibit a hardwired, yet usually involuntary and hidden representational preference towards certain languages. Linguistic bias is manifested in uneven per-language performance even in the case of similar test conditions. We show that biased technology is often the result of research and development methodologies that do not do justice to the complexity of the languages being represented, and that can even become ethically problematic as they disregard valuable aspects of diversity as well as the needs of the language communities themselves. As our attempt at building diversity-aware language resources, we present a new initiative that aims at reducing linguistic bias through both technological design and methodology, based on an eye-level collaboration with local communities.

Abstract: Ambiguity is ubiquitous in natural language. Resolving ambiguous meanings is especially important in information retrieval tasks. While word embeddings carry semantic information, they fail to handle ambiguity well. Transformer models have been shown to handle word ambiguity for complex queries, but they cannot be used to identify ambiguous words, e.g. for a 1-word query. Furthermore, training these models is costly in terms of time, hardware resources, and training data, prohibiting their use in specialized environments with sensitive data. Word embeddings can be trained using moderate hardware resources. This paper shows that applying DBSCAN clustering to the latent space can identify ambiguous words and evaluate their level of ambiguity. An automatic DBSCAN parameter selection leads to high-quality clusters, which are semantically coherent and correspond well to the perceived meanings of a given word.

Abstract: In this paper, we conduct a holistic exploration of the Universal Decompositional Semantic (UDS) Parsing. We first introduce a cascade model for UDS parsing that decomposes the complex parsing task into semantically appropriate subtasks. Our approach outperforms the prior models, while significantly reducing inference time. We also incorporate syntactic information and further optimized the architecture. Besides, different ways for data augmentation are explored, which further improve the UDS Parsing. Lastly, we conduct experiments to investigate the efficacy of ChatGPT in handling the UDS task, revealing that it excels in attribute parsing but struggles in relation parsing, and using ChatGPT for data augmentation yields suboptimal results. Our code is available at https://github.com/hexuandeng/HExp4UDS.

Abstract: The Zero-Shot Learning (ZSL) task pertains to the identification of entities or relations in texts that were not seen during training. ZSL has emerged as a critical research area due to the scarcity of labeled data in specific domains, and its applications have grown significantly in recent years. With the advent of large pretrained language models, several novel methods have been proposed, resulting in substantial improvements in ZSL performance. There is a growing demand, both in the research community and industry, for a comprehensive ZSL framework that facilitates the development and accessibility of the latest methods and pretrained models.In this study, we propose a novel ZSL framework called Zshot that aims to address the aforementioned challenges. Our primary objective is to provide a platform that allows researchers to compare different state-of-the-art ZSL methods with standard benchmark datasets. Additionally, we have designed our framework to support the industry with readily available APIs for production under the standard SpaCy NLP pipeline. Our API is extendible and evaluable, moreover, we include numerous enhancements such as boosting the accuracy with pipeline ensembling and visualization utilities available as a SpaCy extension.

Abstract: The emergence of generative pre-trained models has facilitated the synthesis of high-quality text, but it has also posed challenges in identifying factual errors in the generated text. In particular: (1) A wider range of tasks now face an increasing risk of containing factual errors when handled by generative models. (2) Generated texts tend to be lengthy and lack a clearly defined granularity for individual facts. (3) There is a scarcity of explicit evidence available during the process of fact checking. With the above challenges in mind, in this paper, we propose FacTool, a task and domain agnostic framework for detecting factual errors of texts generated by large language models (e.g., ChatGPT). Experiments on four different tasks (knowledge-based QA, code generation, mathematical reasoning, and scientific literature review) show the efficacy of the proposed method.

Abstract: Recently, diffusion models have excelled in image generation tasks and have also been applied to neural language processing (NLP) for controllable text generation. However, the application of diffusion models in a cross-lingual setting is less unexplored. Additionally, while pretraining with diffusion models has been studied within a single language, the potential of cross-lingual pretraining remains understudied. To address these gaps, we propose XDLM, a novel Cross-lingual diffusion model for machine translation, consisting of pretraining and fine-tuning stages. In the pretraining stage, we propose TLDM, a new training objective for mastering the mapping between different languages; in the fine-tuning stage, we build up the translation system based on the pretrained model. We evaluate the result on several machine translation benchmarks and outperformed both diffusion and Transformer baselines.

Abstract: Financial analysis is an important tool for evaluating company performance. Practitioners work to answer financial questions to make profitable investment decisions, and use advanced quantitative analyses to do so. As a result, Financial Question Answering (QA) is a question answering task that requires deep reasoning about numbers. Furthermore, it is unknown how well pre-trained language models can reason in the financial domain. The current state-of-the-art requires a retriever to collect relevant facts about the financial question from the text and a generator to produce a valid financial program and a final answer. However, recently large language models like GPT-3 have achieved state-of-the-art performance on wide variety of tasks with just a few shot examples. We run several experiments with GPT-3 and find that a separate retrieval model and logic engine continue to be essential components to achieving SOTA performance in this task, particularly due to the precise nature of financial questions and the complex information stored in financial documents. With this understanding, our refined prompt-engineering approach on GPT-3 achieves near SOTA accuracy without any fine-tuning.

Abstract: This thesis work falls within the framework of question answering (QA) in the biomedical domain where several specific challenges are addressed, such as specialized lexicons and terminologies, the types of treated questions, and the characteristics of targeted documents. We are particularly interested in studying and improving methods that aim at finding accurate and short answers to biomedical natural language questions from a large scale of biomedical textual documents in English. QA aims at providing inquirers with direct, short and precise answers to their natural language questions. In this Ph.D. thesis, we propose four contributions to improve the performance of QA in the biomedical domain. In our first contribution, we propose a machine learning-based method for question type classification to determine the types of given questions which enable to a biomedical QA system to use the appropriate answer extraction method. We also propose an another machine learning-based method to assign one or more topics (e.g., pharmacological, test, treatment, etc.) to given questions in order to determine the semantic types of the expected answers which are very useful in generating specific answer retrieval strategies. In the second contribution, we first propose a document retrieval method to retrieve a set of relevant documents that are likely to contain the answers to biomedical questions from the MEDLINE database. We then present a passage retrieval method to retrieve a set of relevant passages to questions. In the third contribution, we propose specific answer extraction methods to generate both exact and ideal answers. Finally, in the fourth contribution, we develop a fully automated semantic biomedical QA system called SemBioNLQA which is able to deal with a variety of natural language questions and to generate appropriate answers by providing both exact and ideal answers.

Abstract: With the development of pre-trained models and the incorporation of phonetic and graphic information, neural models have achieved high scores in Chinese Spelling Check (CSC). However, it does not provide a comprehensive reflection of the models' capability due to the limited test sets. In this study, we abstract the representative model paradigm, implement it with nine structures and experiment them on comprehensive test sets we constructed with different purposes. We perform a detailed analysis of the results and find that: 1) Fusing phonetic and graphic information reasonably is effective for CSC. 2) Models are sensitive to the error distribution of the test set, which reflects the shortcomings of models and reveals the direction we should work on. 3) Whether or not the errors and contexts have been seen has a significant impact on models. 4) The commonly used benchmark, SIGHAN, can not reliably evaluate models' performance.

Abstract: Large Language Models (LLMs) have demonstrated remarkable performance on various quantitative reasoning and knowledge benchmarks. However, many of these benchmarks are losing utility as LLMs get increasingly high scores, despite not yet reaching expert performance in these domains. We introduce ARB, a novel benchmark composed of advanced reasoning problems in multiple fields. ARB presents a more challenging test than prior benchmarks, featuring problems in mathematics, physics, biology, chemistry, and law. As a subset of ARB, we introduce a challenging set of math and physics problems which require advanced symbolic reasoning and domain knowledge. We evaluate recent models such as GPT-4 and Claude on ARB and demonstrate that current models score well below 50% on more demanding tasks. In order to improve both automatic and assisted evaluation capabilities, we introduce a rubric-based evaluation approach, allowing GPT-4 to score its own intermediate reasoning steps. Further, we conduct a human evaluation of the symbolic subset of ARB, finding promising agreement between annotators and GPT-4 rubric evaluation scores.

Abstract: The rise of large language models (LLMs) has marked a pivotal shift in the field of natural language processing (NLP). LLMs have revolutionized a multitude of domains, and they have made a significant impact in the medical field. Large language models are now more abundant than ever, and many of these models exhibit bilingual capabilities, proficient in both English and Chinese. However, a comprehensive evaluation of these models remains to be conducted. This lack of assessment is especially apparent within the context of radiology NLP. This study seeks to bridge this gap by critically evaluating thirty two LLMs in interpreting radiology reports, a crucial component of radiology NLP. Specifically, the ability to derive impressions from radiologic findings is assessed. The outcomes of this evaluation provide key insights into the performance, strengths, and weaknesses of these LLMs, informing their practical applications within the medical domain.