Objective: To explore and compare the performance of ChatGPT and other state-of-the-art LLMs on domain-specific NER tasks covering different entity types and domains in TCM against COVID-19 literature. Methods: We established a dataset of 389 articles on TCM against COVID-19, and manually annotated 48 of them with 6 types of entities belonging to 3 domains as the ground truth, against which the NER performance of LLMs can be assessed. We then performed NER tasks for the 6 entity types using ChatGPT (GPT-3.5 and GPT-4) and 4 state-of-the-art BERT-based question-answering (QA) models (RoBERTa, MiniLM, PubMedBERT and SciBERT) without prior training on the specific task. A domain fine-tuned model (GSAP-NER) was also applied for a comprehensive comparison. Results: The overall performance of LLMs varied significantly in exact match and fuzzy match. In the fuzzy match, ChatGPT surpassed BERT-based QA models in 5 out of 6 tasks, while in exact match, BERT-based QA models outperformed ChatGPT in 5 out of 6 tasks but with a smaller F-1 difference. GPT-4 showed a significant advantage over other models in fuzzy match, especially on the entity type of TCM formula and the Chinese patent drug (TFD) and ingredient (IG). Although GPT-4 outperformed BERT-based models on entity type of herb, target, and research method, none of the F-1 scores exceeded 0.5. GSAP-NER, outperformed GPT-4 in terms of F-1 by a slight margin on RM. ChatGPT achieved considerably higher recalls than precisions, particularly in the fuzzy match. Conclusions: The NER performance of LLMs is highly dependent on the entity type, and their performance varies across application scenarios. ChatGPT could be a good choice for scenarios where high recall is favored. However, for knowledge acquisition in rigorous scenarios, neither ChatGPT nor BERT-based QA models are off-the-shelf tools for professional practitioners.

Sampling-based decoding strategies have been widely adopted for Large Language Models (LLMs) in numerous applications, which target a balance between diversity and quality via temperature tuning and tail truncation (e.g., top-k and top-p sampling). Considering the high dynamic range of the candidate next-token given different prefixes, recent studies propose to adaptively truncate the tail of LLM's predicted distribution. Although improved results haven been reported with these methods on open-ended text generation tasks, the results are highly dependent on the curated truncation parameters and exemplar text. In this paper, we propose a systematic way to estimate the intrinsic capacity of a truncation sampling method by considering the trade-off between diversity and risk at each decoding step, based on our collected prefix tree which preserves the context of a full sentence. Our work provides a comprehensive comparison between existing truncation sampling methods, as well as their recommended parameters as a guideline for users.

This research seeks to obviate the need for creating QA datasets and grading (chatbot) LLM responses when comparing LLMs' knowledge in specific topic domains. This is done in an entirely end-user centric way without need for access to any inner workings of the LLM, so long as it can be prompted and given a random seed to create different generations to the same prompt. The paper does this by, for a given topic domain, defining the "response dispersion" of an LLM by repeatedly asking an LLM the same opinion question about that topic domain. Namely, the response dispersion is the count of singular values needed to explain 95% of the variance in the embedding matrix of the LLM's responses. It is found that the response dispersion is inversely correlated with accuracy on relevant QA evaluations (average spearman rank correlation stronger than -.59). A use-case analysis shows that when comparing two different LLMs on the same topic domain, comparing their response dispersion is a suitable replacement for comparing their QA accuracy between 74% and 89% of the time, the range depending on certain reasonable accuracy-difference tolerances that may be acceptable to an end-user in exchange for the labor being saved using response dispersion instead of QA accuracy for comparison. Two response embeddings are studied for creating the embedding matrix in this study, one is from OpenAI's APIs and one is a novel embedding, here named reference sentence similarity embeddings, that can be computed locally and performs very nearly as well in calculating response dispersion. Also in this research, a pre-existing dataset called the IRC-Wiki Trivia dataset, originally developed for trivia games, has been re-purposed, curated, and the curation, called IRC-WikiTriviaQA, is made available for the purpose of this research.

Retrieval-Augmented Generation (RAG) has emerged as a crucial method for addressing hallucinations in large language models (LLMs). While recent research has extended RAG models to complex noisy scenarios, these explorations often confine themselves to limited noise types and presuppose that noise is inherently detrimental to LLMs, potentially deviating from real-world retrieval environments and restricting practical applicability. In this paper, we define seven distinct noise types from a linguistic perspective and establish a Noise RAG Benchmark (NoiserBench), a comprehensive evaluation framework encompassing multiple datasets and reasoning tasks. Through empirical evaluation of eight representative LLMs with diverse architectures and scales, we reveal that these noises can be further categorized into two practical groups: noise that is beneficial to LLMs (aka beneficial noise) and noise that is harmful to LLMs (aka harmful noise). While harmful noise generally impairs performance, beneficial noise may enhance several aspects of model capabilities and overall performance. Our analysis offers insights for developing more robust, adaptable RAG solutions and mitigating hallucinations across diverse retrieval scenarios.

Sign language translation has historically been peripheral to mainstream machine translation research. In order to help converge the fields, we introduce FLEURS-ASL, an extension of the multiway parallel benchmarks FLORES (for text) and FLEURS (for speech) to support their first sign language (as video), American Sign Language, translated by 5 Certified Deaf Interpreters. FLEURS-ASL can be used to evaluate a variety of tasks -- primarily sentence- and discourse-level translation -- between ASL and 200 other languages as text, or 102 languages as speech. We provide baselines for tasks from ASL to English text using a unified modeling approach that incorporates timestamp tokens and previous text tokens in a 34-second context window, trained on random video clips from YouTube-ASL. This model meets or exceeds the performance of phrase-level baselines while supporting a multitude of new tasks. We also use FLEURS-ASL to show that multimodal frontier models have virtually no understanding of ASL, underscoring the importance of including sign languages in standard evaluation suites.

Biases inherent in human endeavors pose significant challenges for machine learning, particularly in supervised learning that relies on potentially biased "ground truth" data. This reliance, coupled with models' tendency to generalize based on statistical maximal likelihood, can propagate and amplify biases, exacerbating societal issues. To address this, our study proposes a reflective methodology utilizing multiple Large Language Models (LLMs) engaged in a dynamic dialogue to uncover diverse perspectives. By leveraging conditional statistics, information theory, and divergence metrics, this novel approach fosters context-dependent linguistic behaviors, promoting unbiased outputs. Furthermore, it enables measurable progress tracking and explainable remediation actions to address identified biases.

Speech-language multi-modal learning presents a significant challenge due to the fine nuanced information inherent in speech styles. Therefore, a large-scale dataset providing elaborate comprehension of speech style is urgently needed to facilitate insightful interplay between speech audio and natural language. However, constructing such datasets presents a major trade-off between large-scale data collection and high-quality annotation. To tackle this challenge, we propose an automatic speech annotation system for expressiveness interpretation that annotates in-the-wild speech clips with expressive and vivid human language descriptions. Initially, speech audios are processed by a series of expert classifiers and captioning models to capture diverse speech characteristics, followed by a fine-tuned LLaMA for customized annotation generation. Unlike previous tag/templet-based annotation frameworks with limited information and diversity, our system provides in-depth understandings of speech style through tailored natural language descriptions, thereby enabling accurate and voluminous data generation for large model training. With this system, we create SpeechCraft, a fine-grained bilingual expressive speech dataset. It is distinguished by highly descriptive natural language style prompts, containing approximately 2,000 hours of audio data and encompassing over two million speech clips. Extensive experiments demonstrate that the proposed dataset significantly boosts speech-language task performance in stylist speech synthesis and speech style understanding.

Schema matching is the process of identifying correspondences between the elements of two given schemata, essential for database management systems, data integration, and data warehousing. The inherent uncertainty of current schema matching algorithms leads to the generation of a set of candidate matches. Storing these results necessitates the use of databases and systems capable of handling probabilistic queries. This complicates the querying process and increases the associated storage costs. Motivated by GPT-4 outstanding performance, we explore its potential to reduce uncertainty. Our proposal is to supplant the role of crowdworkers with GPT-4 for querying the set of candidate matches. To get more precise correspondence verification responses from GPT-4, We have crafted Semantic-match and Abbreviation-match prompt for GPT-4, achieving state-of-the-art results on two benchmark datasets DeepMDatasets 100% (+0.0) and Fabricated-Datasets 91.8% (+2.2) recall rate. To optimise budget utilisation, we have devised a cost-aware solution. Within the constraints of the budget, our solution delivers favourable outcomes with minimal time expenditure. We introduce a novel framework, Prompt-Matcher, to reduce the uncertainty in the process of integration of multiple automatic schema matching algorithms and the selection of complex parameterization. It assists users in diminishing the uncertainty associated with candidate schema match results and in optimally ranking the most promising matches. We formally define the Correspondence Selection Problem, aiming to optimise the revenue within the confines of the GPT-4 budget. We demonstrate that CSP is NP-Hard and propose an approximation algorithm with minimal time expenditure. Ultimately, we demonstrate the efficacy of Prompt-Matcher through rigorous experiments.

Large Language Models (LLMs) are increasingly serving as evaluators in Natural Language Generation (NLG) tasks. However, the capabilities of LLMs in scoring NLG quality remain inadequately explored. Current studies depend on human assessments and simple metrics that fail to capture the discernment of LLMs across diverse NLG tasks. To address this gap, we propose the Discernment of Hierarchical Perturbation (DHP) benchmarking framework, which provides quantitative discernment scores for LLMs utilizing hierarchically perturbed text data and statistical tests to measure the NLG evaluation capabilities of LLMs systematically. We have re-established six evaluation datasets for this benchmark, covering four NLG tasks: Summarization, Story Completion, Question Answering, and Translation. Our comprehensive benchmarking of five major LLM series provides critical insight into their strengths and limitations as NLG evaluators.

Machine Translation of Culture-Specific Items (CSIs) poses significant challenges. Recent work on CSI translation has shown some success using Large Language Models (LLMs) to adapt to different languages and cultures; however, a deeper analysis is needed to examine the benefits and pitfalls of each method. In this paper, we introduce the ChineseMenuCSI dataset, the largest for Chinese-English menu corpora, annotated with CSI vs Non-CSI labels and a fine-grained test set. We define three levels of CSI figurativeness for a more nuanced analysis and develop a novel methodology for automatic CSI identification, which outperforms GPT-based prompts in most categories. Importantly, we are the first to integrate human translation theories into LLM-driven translation processes, significantly improving translation accuracy, with COMET scores increasing by up to 7 points.