LinguaLens: Towards Interpreting Linguistic Mechanisms of Large Language Models via Sparse Auto-Encoder
DCST, BNRist（DCST、BNRist；KIRC、人工智能研究院、清华大学，中国）；KIRC, Institute for Artificial Intelligence, Tsinghua University, China（中文语言文学系、北京大学，中国）；Department of Chinese Language and Literature, Peking University, China
Yi Jing, Zijun Yao, Hongzhu Guo, Lingxu Ran, Xiaozhi Wang, Lei Hou, Juanzi Li
Large language models (LLMs) demonstrate exceptional performance on tasks requiring complex linguistic abilities, such as reference disambiguation and metaphor recognition/generation. Although LLMs possess impressive capabilities, their internal mechanisms for processing and representing linguistic knowledge remain largely opaque. Prior research on linguistic mechanisms is limited by coarse granularity, limited analysis scale, and narrow focus. In this study, we propose LinguaLens, a systematic and comprehensive framework for analyzing the linguistic mechanisms of large language models, based on Sparse Auto-Encoders (SAEs). We extract a broad set of Chinese and English linguistic features across four dimensions (morphology, syntax, semantics, and pragmatics). By employing counterfactual methods, we construct a large-scale counterfactual dataset of linguistic features for mechanism analysis. Our findings reveal intrinsic representations of linguistic knowledge in LLMs, uncover patterns of cross-layer and cross-lingual distribution, and demonstrate the potential to control model outputs. This work provides a systematic suite of resources and methods for studying linguistic mechanisms, offers strong evidence that LLMs possess genuine linguistic knowledge, and lays the foundation for more interpretable and controllable language modeling in future research.
Department of Computer & Information Science & Engineering, University of Florida（佛罗里达大学计算机与信息科学与工程系）；School of Computer Science, University of Oklahoma（俄克拉荷马大学计算机科学学院）；Department of Electrical Engineering and Computer Science, United States Military Academy（美国军事学院电子工程与计算机科学系）
Rubric-Aligned Disentangled Evaluation of Human Simultaneous Interpreting
School of Data Science, The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳）数据科学学院）；School of Artificial Intelligence, The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳）人工智能学院）；Shenzhen Loop Area Institute（深圳河套学院）
Do they mean 'us'? Interpreting Referring Expressions in Intergroup Bias
Department of Computer Science, Ithaca College（伊萨卡学院计算机科学系）；Department of Computer Science, Brown University（布朗大学计算机科学系）；Department of Linguistics, The University of Texas at Austin（德克萨斯大学奥斯汀分校语言学系）
SAEExplainer: Interpreting SAE Features with Activation-Guided Preference Optimization
Shanghai Jiao Tong University（上海交通大学）；NJIT（新泽西理工学院）；Jilin University（吉林大学）；Institute of Computing Technology, CAS（中国科学院计算技术研究所）；The Chinese University of Hong Kong, Shenzhen（香港中文大学（深圳））
Interpreters facilitate multi-lingual meetings but the affordable set of languages is often smaller than what is needed. Automatic simultaneous speech translation can extend the set of provided languages. We investigate if such an automatic system should rather follow the original speaker, or an interpreter to achieve better translation quality at the cost of increased delay. To answer the question, we release Europarl Simultaneous Interpreting Corpus (ESIC), 10 hours of recordings and transcripts of European Parliament speeches in English, with simultaneous interpreting into Czech and German.…
SAE-V: Interpreting Multimodal Models for Enhanced Alignment
Institute for AI, Peking University, Beijing, China（人工智能研究院，北京大学，北京，中国）；State Key Laboratory of General Artificial Intelligence, Institute for AI, Peking University, Beijing, China（通用人工智能国家重点实验室，人工智能研究院，北京大学，北京，中国）
Chongqing University of Posts and Telecommunications（重庆邮电大学）
Currently end-to-end deep learning based open-domain dialogue systems remain black box models, making it easy to generate irrelevant contents with data-driven models. Specifically, latent variables are highly entangled with different semantics in the latent space due to the lack of priori knowledge to guide the training. To address this problem, this paper proposes to harness the generative model with a priori knowledge through a cognitive approach involving mesoscopic scale feature disentanglement. Particularly, the model integrates the macro-level guided-category knowledge and micro-level op…
IIIT-Delhi（德里英迪拉·甘地信息技术学院）；Adobe（奥多比公司）；State University of New York at Buffalo（纽约州立大学布法罗分校）；University of Texas at Austin（德克萨斯大学奥斯汀分校）
Significant progress has been made in deep-learning based Automatic Essay Scoring (AES) systems in the past two decades. However, little research has been put to understand and interpret the black-box nature of these deep-learning based scoring models. Recent work shows that automated scoring systems are prone to even common-sense adversarial samples. Their lack of natural language understanding capability raises questions on the models being actively used by millions of candidates for life-changing decisions. With scoring being a highly multi-modal task, it becomes imperative for scoring mode…
Alzheimer's disease (AD) is an irreversible and progressive brain disease that can be stopped or slowed down with medical treatment. Language changes serve as a sign that a patient's cognitive functions have been impacted, potentially leading to early diagnosis. In this work, we use NLP techniques to classify and analyze the linguistic characteristics of AD patients using the DementiaBank dataset. We apply three neural models based on CNNs, LSTM-RNNs, and their combination, to distinguish between language samples from AD and control patients. We achieve a new independent benchmark accuracy for…
Columbia University（哥伦比亚大学）；Second Language Testing, Inc.（第二语言测试公司）；Indraprastha Institute of Information Technology, Delhi（德里印多尔信息科技学院）；SUNY at Buffalo（布法罗纽约州立大学）；Adobe Media Data Science Research（奥多比媒体数据科学研究中心）
English proficiency assessments have become a necessary metric for filtering and selecting prospective candidates for both academia and industry. With the rise in demand for such assessments, it has become increasingly necessary to have the automated human-interpretable results to prevent inconsistencies and ensure meaningful feedback to the second language learners. Feature-based classical approaches have been more interpretable in understanding what the scoring model learns. Therefore, in this work, we utilize classical machine learning models to formulate a speech scoring task as both a cla…
Language carries implicit human biases, functioning both as a reflection and a perpetuation of stereotypes that people carry with them. Recently, ML-based NLP methods such as word embeddings have been shown to learn such language biases with striking accuracy. This capability of word embeddings has been successfully exploited as a tool to quantify and study human biases. However, previous studies only consider a predefined set of biased concepts to attest (e.g., whether gender is more or less associated with particular jobs), or just discover biased words without helping to understand their me…
Instituto Superior Técnico（里斯本高等理工学院）；Technical University of Denmark（丹麦技术大学）
To improve the traveling experience, researchers have been analyzing the role of attitudes in travel behavior modeling. Although most researchers use closed-ended surveys, the appropriate method to measure attitudes is debatable. Topic Modeling could significantly reduce the time to extract information from open-ended responses and eliminate subjective bias, thereby alleviating analyst concerns. Our research uses Topic Modeling to extract information from open-ended questions and compare its performance with closed-ended responses. Furthermore, some respondents might prefer answering questions…
