SkillMentor: LLM Agent Self-Evolution via Learning Blind-Spot Diagnosis

Agent self-evolution has primarily focused on learning how to act, while overlooking an equally important capability: learning to discover what an agent does not know. Existing approaches typically assume that failure discovery is given, focusing on how to repair failures once they are identified. We ask whether blind-spot diagnosis itself can be learned. We thus study diagnosis as an agent capability separate from execution, and exclude two alternative sources of progress: executor adaptation and human supervision. Under these constraints, performance cannot improve through executor updates or annotated examples, forcing all improvements to originate from the learned diagnostic capability. We propose SkillMentor, which trains a Mentor policy via reinforcement learning to generate diagnostic tasks, identify recurrent failure modes, and curate them into reusable corrective skills. Across AppWorld and BFCLv3, SkillMentor improves executor performance by an average of 44.2%. These results suggest that blind-spot diagnosis is a learnable capability, enabling self-evolution without updating executor weights or relying on human-curated data.

Chinese sensorimotor and embodiment norms for 3,000 lexicalized concepts

Understanding how conceptual knowledge is grounded in bodily experience, and to what extent machine systems can acquire such knowledge without direct sensorimotor experience, are central questions in both cognitive science and embodied artificial intelligence research. Large-scale normative resources are essential for investigating these questions empirically, yet such resources remain sparse for non-Indo-European languages. We present a novel normative database for 3,000 lexicalized concepts in Mandarin Chinese, comprising 11-dimensional sensorimotor ratings and unidimensional embodiment ratings collected from 378 native Mandarin speakers. The ratings demonstrate high reliability and strong cross-norm validity with existing Chinese resources, each of which covers fewer words and a subset of the 11 sensorimotor dimensions. In a validation study, we tested new variables derived from a theoretically motivated metric, Perceptual Strength of Embodiment (PSE) (Huang et al., 2025), together with seven common composite variables, on lexical decision tasks. The results suggest that PSE-Sensorimotor and Minkowski-3 are the strongest composite predictors of lexical decision performance, capturing the facilitatory effects of sensorimotor information on lexical processing. A further exploratory study showed that sensorimotor ratings are substantially recoverable from purely linguistic representations using simple regression models (mean Spearman r = .62 across dimensions), though recovery varied markedly: visual and auditory dimensions yielded higher correspondence than chemosensory ones. Representational similarity analysis further showed that the relational geometry of the sensorimotor space is also partially recoverable (r = .540), consistent with the view that distributional language use encodes aspects of embodied conceptual structure.

Do LLMs Capture Embodied Cognition and Cultural Variation? Cross-Linguistic Evidence from Demonstratives

Do large language models (LLMs) truly acquire embodied cognition and cultural conventions from text? We introduce demonstratives, fundamental spatial expressions like "this/that" in English and "zhè/nà" in Chinese, as a novel probe for grounded knowledge. Using 6,400 responses from 320 native speakers, we establish a human baseline: English speakers reliably distinguish proximal-distal referents but struggle with perspective-taking, while Chinese speakers switch perspectives fluently but tolerate distal ambiguity. In contrast, five state-of-the-art LLMs fail to inherently understand the proximal-distal contrast and show no cultural differences, defaulting to English-centric reasoning. Our study contributes (i) a new task, based on demonstratives, as a new lens for evaluating embodied cognition and cultural conventions; (ii) empirical evidence of cross-cultural asymmetries in human interpretation; (iii) a new perspective on the egocentric-sociocentric debate, showing both orientations coexist but vary across languages; and (iv) a call to address individual variation in future model design.

Discourse-Aware Dual-Track Streaming Response for Low-Latency Spoken Dialogue Systems

Achieving human-like responsiveness is a critical yet challenging goal for cascaded spoken dialogue systems. Conventional ASR-LLM-TTS pipelines follow a strictly sequential paradigm, requiring complete transcription and full reasoning before speech synthesis can begin, which results in high response latency. We propose the Discourse-Aware Dual-Track Streaming Response (DDTSR) framework, a low-latency architecture that enables listen-while-thinking and speak-while-thinking. DDTSR is built upon three key mechanisms: (1) connective-guided small-large model synergy, where an auxiliary small model generates minimal-committal discourse connectives while a large model performs knowledge-intensive reasoning in parallel; (2) streaming-based cross-modal collaboration, which dynamically overlaps ASR, LLM inference, and TTS to advance the earliest speakable moment; and (3) curriculum-learning-based discourse continuity enhancement, which maintains coherence and logical consistency between early responses and subsequent reasoning outputs. Experiments on two spoken dialogue benchmarks demonstrate that DDTSR reduces response latency by 19%-51% while preserving discourse quality. Further analysis shows that DDTSR functions as a plug-and-play module compatible with diverse LLM backbones, and remains robust across varying utterance lengths, indicating strong practicality and scalability for real-time spoken interaction.

Sparse Brains are Also Adaptive Brains: Cognitive-Load-Aware Dynamic Activation for LLMs

Dense large language models(LLMs) face critical efficiency bottlenecks as they rigidly activate all parameters regardless of input complexity. While existing sparsity methods(static pruning or dynamic activation) address this partially, they either lack adaptivity to contextual or model structural demands or incur prohibitive computational overhead. Inspired by human brain's dual-process mechanisms - predictive coding (N400) for backbone sparsity and structural reanalysis (P600) for complex context - we propose CLADA, a \textit{\textbf{C}ognitive-\textbf{L}oad-\textbf{A}ware \textbf{D}ynamic \textbf{A}ctivation} framework that synergizes statistical sparsity with semantic adaptability. Our key insight is that LLM activations exhibit two complementary patterns: 1) \textit{Global statistical sparsity} driven by sequence-level prefix information, and 2) \textit{Local semantic adaptability} modulated by cognitive load metrics(e.g., surprisal and entropy). CLADA employs a hierarchical thresholding strategy: a baseline from offline error-controlled optimization ensures 40\%+ sparsity, dynamically adjusted by real-time cognitive signals. Evaluations across six mainstream LLMs and nine benchmarks demonstrate that CLADA achieves \textbf{~20\% average speedup with <2\% accuracy drop}, outperforming Griffin (5\%+ degradation) and TT (negligible speedup). Crucially, we establish the first formal connection between neurolinguistic event-related potential (ERP) components and LLM efficiency mechanisms through multi-level regression analysis ($R^2=0.17$ for sparsity-adaptation synergy). Requiring no retraining or architectural changes, CLADA offers a deployable solution for resource-aware LLM inference while advancing biologically-inspired AI design. Our code is available at \href{https://github.com/Oldify/CLADA}{CLADA}.

Cross-strait Variations on Two Near-synonymous Loanwords xie2shang1 and tan2pan4: A Corpus-based Comparative Study

This study attempts to investigate cross-strait variations on two typical synonymous loanwords in Chinese, i.e. xie2shang1 and tan2pan4, drawn on MARVS theory. Through a comparative analysis, the study found some distributional, eventual, and contextual similarities and differences across Taiwan and Mainland Mandarin. Compared with the underused tan2pan4, xie2shang1 is significantly overused in Taiwan Mandarin and vice versa in Mainland Mandarin. Additionally, though both words can refer to an inchoative process in Mainland and Taiwan Mandarin, the starting point for xie2shang1 in Mainland Mandarin is somewhat blurring compared with the usage in Taiwan Mandarin. Further on, in Taiwan Mandarin, tan2pan4 can be used in economic and diplomatic contexts, while xie2shang1 is used almost exclusively in political contexts. In Mainland Mandarin, however, the two words can be used in a hybrid manner within political contexts; moreover, tan2pan4 is prominently used in diplomatic contexts with less reference to economic activities, while xie2sahng1 can be found in both political and legal contexts, emphasizing a role of mediation.

Automatic Analysis of Linguistic Features in Journal Articles of Different Academic Impacts with Feature Engineering Techniques

English research articles (RAs) are an essential genre in academia, so the attempts to employ NLP to assist the development of academic writing ability have received considerable attention in the last two decades. However, there has been no study employing feature engineering techniques to investigate the linguistic features of RAs of different academic impacts (i.e., the papers of high/moderate citation times published in the journals of high/moderate impact factors). This study attempts to extract micro-level linguistic features in high- and moderate-impact journal RAs, using feature engineering methods. We extracted 25 highly relevant features from the Corpus of English Journal Articles through feature selection methods. All papers in the corpus deal with COVID-19 medical empirical studies. The selected features were then validated of the classification performance in terms of consistency and accuracy through supervised machine learning methods. Results showed that 24 linguistic features such as the overlapping of content words between adjacent sentences, the use of third-person pronouns, auxiliary verbs, tense, emotional words provide consistent and accurate predictions for journal articles with different academic impacts. Lastly, the random forest model is shown to be the best model to fit the relationship between these 24 features and journal articles with high and moderate impacts. These findings can be used to inform academic writing courses and lay the foundation for developing automatic evaluation systems for L2 graduate students.

Predicting gender and age categories in English conversations using lexical, non-lexical, and turn-taking features

This paper examines gender and age salience and (stereo)typicality in British English talk with the aim to predict gender and age categories based on lexical, phrasal and turn-taking features. We examine the SpokenBNC, a corpus of around 11.4 million words of British English conversations and identify behavioural differences between speakers that are labelled for gender and age categories. We explore differences in language use and turn-taking dynamics and identify a range of characteristics that set the categories apart. We find that female speakers tend to produce more and slightly longer turns, while turns by male speakers feature a higher type-token ratio and a distinct range of minimal particles such as "eh", "uh" and "em". Across age groups, we observe, for instance, that swear words and laughter characterize young speakers' talk, while old speakers tend to produce more truncated words. We then use the observed characteristics to predict gender and age labels of speakers per conversation and per turn as a classification task, showing that non-lexical utterances such as minimal particles that are usually left out of dialog data can contribute to setting the categories apart.

A Structured Distributional Model of Sentence Meaning and Processing

Most compositional distributional semantic models represent sentence meaning with a single vector. In this paper, we propose a Structured Distributional Model (SDM) that combines word embeddings with formal semantics and is based on the assumption that sentences represent events and situations. The semantic representation of a sentence is a formal structure derived from Discourse Representation Theory and containing distributional vectors. This structure is dynamically and incrementally built by integrating knowledge about events and their typical participants, as they are activated by lexical items. Event knowledge is modeled as a graph extracted from parsed corpora and encoding roles and relationships between participants that are represented as distributional vectors. SDM is grounded on extensive psycholinguistic research showing that generalized knowledge about events stored in semantic memory plays a key role in sentence comprehension. We evaluate SDM on two recently introduced compositionality datasets, and our results show that combining a simple compositional model with event knowledge constantly improves performances, even with different types of word embeddings.

A realistic and robust model for Chinese word segmentation

A realistic Chinese word segmentation tool must adapt to textual variations with minimal training input and yet robust enough to yield reliable segmentation result for all variants. Various lexicon-driven approaches to Chinese segmentation, e.g. [1,16], achieve high f-scores yet require massive training for any variation. Text-driven approach, e.g. [12], can be easily adapted for domain and genre changes yet has difficulty matching the high f-scores of the lexicon-driven approaches. In this paper, we refine and implement an innovative text-driven word boundary decision (WBD) segmentation model proposed in [15]. The WBD model treats word segmentation simply and efficiently as a binary decision on whether to realize the natural textual break between two adjacent characters as a word boundary. The WBD model allows simple and quick training data preparation converting characters as contextual vectors for learning the word boundary decision. Machine learning experiments with four different classifiers show that training with 1,000 vectors and 1 million vectors achieve comparable and reliable results. In addition, when applied to SigHAN Bakeoff 3 competition data, the WBD model produces OOV recall rates that are higher than all published results. Unlike all previous work, our OOV recall rate is comparable to our own F-score. Both experiments support the claim that the WBD model is a realistic model for Chinese word segmentation as it can be easily adapted for new variants with the robust result. In conclusion, we will discuss linguistic ramifications as well as future implications for the WBD approach.