Extractive reading comprehension question answering (QA) datasets are typically evaluated using Exact Match (EM) and F1-score, but these metrics often fail to fully capture model performance. With the success of large language models (LLMs), they have been employed in various tasks, including serving as judges (LLM-as-a-judge). In this paper, we reassess the performance of QA models using LLM-as-a-judge across four reading comprehension QA datasets. We examine different families of LLMs and various answer types to evaluate the effectiveness of LLM-as-a-judge in these tasks. Our results show that LLM-as-a-judge is highly correlated with human judgments and can replace traditional EM/F1 metrics. By using LLM-as-a-judge, the correlation with human judgments improves significantly, from 0.17 (EM) and 0.36 (F1-score) to 0.85. These findings confirm that EM and F1 metrics underestimate the true performance of the QA models. While LLM-as-a-judge is not perfect for more difficult answer types (e.g., job), it still outperforms EM/F1, and we observe no bias issues, such as self-preference, when the same model is used for both the QA and judgment tasks.
While large language models (LLMs) have revolutionized text-to-speech (TTS) synthesis through discrete tokenization paradigms, current architectures exhibit fundamental tensions between three critical dimensions: 1) irreversible loss of acoustic characteristics caused by quantization of speech prompts; 2) stringent dependence on precisely aligned prompt speech-text pairs that limit real-world deployment; and 3) catastrophic forgetting of the LLM's native text comprehension during optimization for speech token generation. To address these challenges, we propose an LLM-based text-to-speech Generation approach Optimized via a novel dual-branch ArchiTecture (GOAT-TTS). Our framework introduces two key innovations: (1) The modality-alignment branch combines a speech encoder and projector to capture continuous acoustic embeddings, enabling bidirectional correlation between paralinguistic features (language, timbre, emotion) and semantic text representations without transcript dependency; (2) The speech-generation branch employs modular fine-tuning on top-k layers of an LLM for speech token prediction while freezing the bottom-n layers to preserve foundational linguistic knowledge. Moreover, multi-token prediction is introduced to support real-time streaming TTS synthesis. Experimental results demonstrate that our GOAT-TTS achieves performance comparable to state-of-the-art TTS models while validating the efficacy of synthesized dialect speech data.
The ability of cross-lingual context retrieval is a fundamental aspect of cross-lingual alignment of large language models (LLMs), where the model extracts context information in one language based on requests in another language. Despite its importance in real-life applications, this ability has not been adequately investigated for state-of-the-art models. In this paper, we evaluate the cross-lingual context retrieval ability of over 40 LLMs across 12 languages to understand the source of this ability, using cross-lingual machine reading comprehension (xMRC) as a representative scenario. Our results show that several small, post-trained open LLMs show strong cross-lingual context retrieval ability, comparable to closed-source LLMs such as GPT-4o, and their estimated oracle performances greatly improve after post-training. Our interpretability analysis shows that the cross-lingual context retrieval process can be divided into two main phases: question encoding and answer retrieval, which are formed in pre-training and post-training, respectively. The phasing stability correlates with xMRC performance, and the xMRC bottleneck lies at the last model layers in the second phase, where the effect of post-training can be evidently observed. Our results also indicate that larger-scale pretraining cannot improve the xMRC performance. Instead, larger LLMs need further multilingual post-training to fully unlock their cross-lingual context retrieval potential. Our code and is available at https://github.com/NJUNLP/Cross-Lingual-Context-Retrieval
Large language models (LLMs) demonstrate remarkable text comprehension and generation capabilities but often lack the ability to utilize up-to-date or domain-specific knowledge not included in their training data. To address this gap, we introduce KEDiT, an efficient method for fine-tuning LLMs for knowledge-grounded dialogue generation. KEDiT operates in two main phases: first, it employs an information bottleneck to compress retrieved knowledge into learnable parameters, retaining essential information while minimizing computational overhead. Second, a lightweight knowledge-aware adapter integrates these compressed knowledge vectors into the LLM during fine-tuning, updating less than 2\% of the model parameters. The experimental results on the Wizard of Wikipedia and a newly constructed PubMed-Dialog dataset demonstrate that KEDiT excels in generating contextually relevant and informative responses, outperforming competitive baselines in automatic, LLM-based, and human evaluations. This approach effectively combines the strengths of pretrained LLMs with the adaptability needed for incorporating dynamic knowledge, presenting a scalable solution for fields such as medicine.
Despite the steady progress in machine translation evaluation, existing automatic metrics struggle to capture how well meaning is preserved beyond sentence boundaries. We posit that reliance on a single intrinsic quality score, trained to mimic human judgments, might be insufficient for evaluating translations of long, complex passages, and a more ``pragmatic'' approach that assesses how accurately key information is conveyed by a translation in context is needed. We introduce TREQA (Translation Evaluation via Question-Answering), a framework that extrinsically evaluates translation quality by assessing how accurately candidate translations answer reading comprehension questions that target key information in the original source or reference texts. In challenging domains that require long-range understanding, such as literary texts, we show that TREQA is competitive with and, in some cases, outperforms state-of-the-art neural and LLM-based metrics in ranking alternative paragraph-level translations, despite never being explicitly optimized to correlate with human judgments. Furthermore, the generated questions and answers offer interpretability: empirical analysis shows that they effectively target translation errors identified by experts in evaluated datasets. Our code is available at https://github.com/deep-spin/treqa
The ability of machines to grasp spatial understanding within narrative contexts is an intriguing aspect of reading comprehension that continues to be studied. Motivated by the goal to test the AI's competence in understanding the relationship between characters and their respective locations in narratives, we introduce two new datasets: Andersen and Persuasion. For the Andersen dataset, we selected fifteen children's stories from "Andersen's Fairy Tales" by Hans Christian Andersen and manually annotated the characters and their respective locations throughout each story. Similarly, for the Persuasion dataset, characters and their locations in the novel "Persuasion" by Jane Austen were also manually annotated. We used these datasets to prompt Large Language Models (LLMs). The prompts are created by extracting excerpts from the stories or the novel and combining them with a question asking the location of a character mentioned in that excerpt. Out of the five LLMs we tested, the best-performing one for the Andersen dataset accurately identified the location in 61.85% of the examples, while for the Persuasion dataset, the best-performing one did so in 56.06% of the cases.
In modern online streaming platforms, the comments section plays a critical role in enhancing the overall user experience. Understanding user behavior within the comments section is essential for comprehensive user interest modeling. A key factor of user engagement is staytime, which refers to the amount of time that users browse and post comments. Existing watchtime prediction methods struggle to adapt to staytime prediction, overlooking interactions with individual comments and their interrelation. In this paper, we present a micro-video recommendation dataset with video comments (named as KuaiComt) which is collected from Kuaishou platform. correspondingly, we propose a practical framework for comment staytime prediction with LLM-enhanced Comment Understanding (LCU). Our framework leverages the strong text comprehension capabilities of large language models (LLMs) to understand textual information of comments, while also incorporating fine-grained comment ranking signals as auxiliary tasks. The framework is two-staged: first, the LLM is fine-tuned using domain-specific tasks to bridge the video and the comments; second, we incorporate the LLM outputs into the prediction model and design two comment ranking auxiliary tasks to better understand user preference. Extensive offline experiments demonstrate the effectiveness of our framework, showing significant improvements on the task of comment staytime prediction. Additionally, online A/B testing further validates the practical benefits on industrial scenario. Our dataset KuaiComt (https://github.com/lyingCS/KuaiComt.github.io) and code for LCU (https://github.com/lyingCS/LCU) are fully released.
The release of top-performing open-weight LLMs has cemented China's role as a leading force in AI development. Do these models support languages spoken in China? Or do they speak the same languages as Western models? Comparing multilingual capabilities is important for two reasons. First, language ability provides insights into pre-training data curation, and thus into resource allocation and development priorities. Second, China has a long history of explicit language policy, varying between inclusivity of minority languages and a Mandarin-first policy. To test whether Chinese LLMs today reflect an agenda about China's languages, we test performance of Chinese and Western open-source LLMs on Asian regional and Chinese minority languages. Our experiments on Information Parity and reading comprehension show Chinese models' performance across these languages correlates strongly (r=0.93) with Western models', with the sole exception being better Mandarin. Sometimes, Chinese models cannot identify languages spoken by Chinese minorities such as Kazakh and Uyghur, even though they are good at French and German. These results provide a window into current development priorities, suggest options for future development, and indicate guidance for end users.
This paper explores the effectiveness of Multimodal Large Language models (MLLMs) as assistive technologies for visually impaired individuals. We conduct a user survey to identify adoption patterns and key challenges users face with such technologies. Despite a high adoption rate of these models, our findings highlight concerns related to contextual understanding, cultural sensitivity, and complex scene understanding, particularly for individuals who may rely solely on them for visual interpretation. Informed by these results, we collate five user-centred tasks with image and video inputs, including a novel task on Optical Braille Recognition. Our systematic evaluation of twelve MLLMs reveals that further advancements are necessary to overcome limitations related to cultural context, multilingual support, Braille reading comprehension, assistive object recognition, and hallucinations. This work provides critical insights into the future direction of multimodal AI for accessibility, underscoring the need for more inclusive, robust, and trustworthy visual assistance technologies.
Large Language Models (LLMs) have shown remarkable proficiency in Machine Reading Comprehension (MRC) tasks; however, their effectiveness for low-resource languages like Vietnamese remains largely unexplored. In this paper, we fine-tune and evaluate two state-of-the-art LLMs: Llama 3 (8B parameters) and Gemma (7B parameters), on ViMMRC, a Vietnamese MRC dataset. By utilizing Quantized Low-Rank Adaptation (QLoRA), we efficiently fine-tune these models and compare their performance against powerful LLM-based baselines. Although our fine-tuned models are smaller than GPT-3 and GPT-3.5, they outperform both traditional BERT-based approaches and these larger models. This demonstrates the effectiveness of our fine-tuning process, showcasing how modern LLMs can surpass the capabilities of older models like BERT while still being suitable for deployment in resource-constrained environments. Through intensive analyses, we explore various aspects of model performance, providing valuable insights into adapting LLMs for low-resource languages like Vietnamese. Our study contributes to the advancement of natural language processing in low-resource languages, and we make our fine-tuned models publicly available at: https://huggingface.co/iaiuet.
In this study, it was investigated whether AI evaluators assess the content validity of B1-level English reading comprehension test items in a manner similar to human evaluators. A 25-item multiple-choice test was developed, and these test items were evaluated by four human and four AI evaluators. No statistically significant difference was found between the scores given by human and AI evaluators, with similar evaluation trends observed. The Content Validity Ratio (CVR) and the Item Content Validity Index (I-CVI) were calculated and analyzed using the Wilcoxon Signed-Rank Test, with no statistically significant difference. The findings revealed that in some cases, AI evaluators could replace human evaluators. However, differences in specific items were thought to arise from varying interpretations of the evaluation criteria. Ensuring linguistic clarity and clearly defining criteria could contribute to more consistent evaluations. In this regard, the development of hybrid evaluation systems, in which AI technologies are used alongside human experts, is recommended.
Machine Reading Comprehension (MRC) is an essential task in evaluating natural language understanding. Existing MRC datasets primarily assess specific aspects of reading comprehension (RC), lacking a comprehensive MRC benchmark. To fill this gap, we first introduce a novel taxonomy that categorizes the key capabilities required for RC. Based on this taxonomy, we construct MRCEval, an MRC benchmark that leverages advanced Large Language Models (LLMs) as both sample generators and selection judges. MRCEval is a comprehensive, challenging and accessible benchmark designed to assess the RC capabilities of LLMs thoroughly, covering 13 distinct RC skills with a total of 2.1K high-quality multi-choice questions. We perform an extensive evaluation of 28 widely used open-source and proprietary models, highlighting that MRC continues to present significant challenges even in the era of LLMs.
With the rapid development in Transformer-based language models, the reading comprehension tasks on short documents and simple questions have been largely addressed. Long documents, specifically the scientific documents that are densely packed with knowledge discovered and developed by humans, remain relatively unexplored. These documents often come with a set of complex and more realistic questions, adding to their complexity. We present a zero-shot pipeline framework that enables social science researchers to perform question-answering tasks that are complex yet of predetermined question formats on full-length research papers without requiring machine learning expertise. Our approach integrates pre-trained language models to handle challenging scenarios including multi-span extraction, multi-hop reasoning, and long-answer generation. Evaluating on MLPsych, a novel dataset of social psychology papers with annotated complex questions, we demonstrate that our framework achieves strong performance through combination of extractive and generative models. This work advances document understanding capabilities for social sciences while providing practical tools for researchers.
An Achilles heel of Large Language Models (LLMs) is their tendency to hallucinate non-factual statements. A response mixed of factual and non-factual statements poses a challenge for humans to verify and accurately base their decisions on. To combat this problem, we propose Highlighted Chain-of-Thought Prompting (HoT), a technique for prompting LLMs to generate responses with XML tags that ground facts to those provided in the query. That is, given an input question, LLMs would first re-format the question to add XML tags highlighting key facts, and then, generate a response with highlights over the facts referenced from the input. Interestingly, in few-shot settings, HoT outperforms vanilla chain of thought prompting (CoT) on a wide range of 17 tasks from arithmetic, reading comprehension to logical reasoning. When asking humans to verify LLM responses, highlights help time-limited participants to more accurately and efficiently recognize when LLMs are correct. Yet, surprisingly, when LLMs are wrong, HoTs tend to make users believe that an answer is correct.
Extracting causal relationships from a medical case report is essential for comprehending the case, particularly its diagnostic process. Since the diagnostic process is regarded as a bottom-up inference, causal relationships in cases naturally form a multi-layered tree structure. The existing tasks, such as medical relation extraction, are insufficient for capturing the causal relationships of an entire case, as they treat all relations equally without considering the hierarchical structure inherent in the diagnostic process. Thus, we propose a novel task, Causal Tree Extraction (CTE), which receives a case report and generates a causal tree with the primary disease as the root, providing an intuitive understanding of a case's diagnostic process. Subsequently, we construct a Japanese case report CTE dataset, J-Casemap, propose a generation-based CTE method that outperforms the baseline by 20.2 points in the human evaluation, and introduce evaluation metrics that reflect clinician preferences. Further experiments also show that J-Casemap enhances the performance of solving other medical tasks, such as question answering.
Reading comprehension is a key for individual success, yet the assessment of question difficulty remains challenging due to the extensive human annotation and large-scale testing required by traditional methods such as linguistic analysis and Item Response Theory (IRT). While these robust approaches provide valuable insights, their scalability is limited. There is potential for Large Language Models (LLMs) to automate question difficulty estimation; however, this area remains underexplored. Our study investigates the effectiveness of LLMs, specifically OpenAI's GPT-4o and o1, in estimating the difficulty of reading comprehension questions using the Study Aid and Reading Assessment (SARA) dataset. We evaluated both the accuracy of the models in answering comprehension questions and their ability to classify difficulty levels as defined by IRT. The results indicate that, while the models yield difficulty estimates that align meaningfully with derived IRT parameters, there are notable differences in their sensitivity to extreme item characteristics. These findings suggest that LLMs can serve as the scalable method for automated difficulty assessment, particularly in dynamic interactions between learners and Adaptive Instructional Systems (AIS), bridging the gap between traditional psychometric techniques and modern AIS for reading comprehension and paving the way for more adaptive and personalized educational assessments.
As neural language models achieve human-comparable performance on Machine Reading Comprehension (MRC) and see widespread adoption, ensuring their robustness in real-world scenarios has become increasingly important. Current robustness evaluation research, though, primarily develops synthetic perturbation methods, leaving unclear how well they reflect real life scenarios. Considering this, we present a framework to automatically examine MRC models on naturally occurring textual perturbations, by replacing paragraph in MRC benchmarks with their counterparts based on available Wikipedia edit history. Such perturbation type is natural as its design does not stem from an arteficial generative process, inherently distinct from the previously investigated synthetic approaches. In a large-scale study encompassing SQUAD datasets and various model architectures we observe that natural perturbations result in performance degradation in pre-trained encoder language models. More worryingly, these state-of-the-art Flan-T5 and Large Language Models (LLMs) inherit these errors. Further experiments demonstrate that our findings generalise to natural perturbations found in other more challenging MRC benchmarks. In an effort to mitigate these errors, we show that it is possible to improve the robustness to natural perturbations by training on naturally or synthetically perturbed examples, though a noticeable gap still remains compared to performance on unperturbed data.
Fine-tuning Large Language Models (LLMs) on specific datasets is a common practice to improve performance on target tasks. However, this performance gain often leads to overfitting, where the model becomes too specialized in either the task or the characteristics of the training data, resulting in a loss of generalization. This paper introduces Selective Self-to-Supervised Fine-Tuning (S3FT), a fine-tuning approach that achieves better performance than the standard supervised fine-tuning (SFT) while improving generalization. S3FT leverages the existence of multiple valid responses to a query. By utilizing the model's correct responses, S3FT reduces model specialization during the fine-tuning stage. S3FT first identifies the correct model responses from the training set by deploying an appropriate judge. Then, it fine-tunes the model using the correct model responses and the gold response (or its paraphrase) for the remaining samples. The effectiveness of S3FT is demonstrated through experiments on mathematical reasoning, Python programming and reading comprehension tasks. The results show that standard SFT can lead to an average performance drop of up to $4.4$ on multiple benchmarks, such as MMLU and TruthfulQA. In contrast, S3FT reduces this drop by half, i.e. $2.5$, indicating better generalization capabilities than SFT while performing significantly better on the fine-tuning tasks.
Despite the impressive performance of multilingual large language models (mLLMs) in various natural language processing tasks, their ability to understand procedural texts, particularly those with culture-specific content, remains largely unexplored. Texts describing cultural procedures, including rituals, traditional craftsmanship, and social etiquette, require an inherent understanding of cultural context, presenting a significant challenge for mLLMs. In this work, we introduce CAPTex, a benchmark designed to evaluate mLLMs' ability to process and reason about culturally diverse procedural texts across multiple languages using various methodologies to assess their performance. Our findings indicate that (1) mLLMs face difficulties with culturally contextualized procedural texts, showing notable performance declines in low-resource languages, (2) model performance fluctuates across cultural domains, with some areas presenting greater difficulties, and (3) language models exhibit better performance on multiple-choice tasks within conversational frameworks compared to direct questioning. These results underscore the current limitations of mLLMs in handling culturally nuanced procedural texts and highlight the need for culturally aware benchmarks like CAPTex to enhance their adaptability and comprehension across diverse linguistic and cultural landscapes.
Recent progress in Learning by Reading and Machine Reading systems has significantly increased the capacity of knowledge-based systems to learn new facts. In this work, we discuss the problem of selecting a set of learning requests for these knowledge-based systems which would lead to maximum Q/A performance. To understand the dynamics of this problem, we simulate the properties of a learning strategy, which sends learning requests to an external knowledge source. We show that choosing an optimal set of facts for these learning systems is similar to a coordination game, and use reinforcement learning to solve this problem. Experiments show that such an approach can significantly improve Q/A performance.