We introduce a new computational framework for segmenting Classical Tibetan texts into autochthonous and allochthonous regions, distinguishing between indigenous Tibetan compositions and translated materials, primarily from Sanskrit sources. To support this task, we release the first annotated Tibetan corpus for ALLO/AUTO segmentation and evaluate several multilingual encoders, including mBERT and XLM-R, fine-tuned for sequence labeling. Our best model achieves strong alignment with expert annotations, showing that multilingual representations can effectively capture philological boundaries in low-resource settings. This work contributes new resources and methods for computational philology and sheds light on the linguistic markers that trace the intercultural transmission of Buddhist thought in Tibet.

A fable is a short narrative whose purpose is to convey a moral lesson. While semantically related, morals and fables differ fundamentally in form: morals are abstract and propositional, while fables express the same idea through characters and plot. In this work, we ask how well this conceptual relationship is captured in the latent representations of embedding models. To study this, we introduce MoraLink, a benchmark for moral-to-fable retrieval in which models must retrieve the correct fable for a given moral. We find that off-the-shelf embedding models struggle with this task, and that augmenting fable documents with LLM-generated moral-style summaries substantially improves retrieval, identifying document representation as the primary bottleneck. Contrastive fine-tuning further improves performance, and we release MoraLink as a challenging testbed for probing how well embedding models capture the abstraction gap between a moral principle and the narrative that embodies it.

Large language models (LLMs) achieve strong performance on idiom identification benchmarks, yet their robustness to misleading contextual signals remains largely untested. We introduce ID10M-JAM, an adversarial extension of the ID10M dataset designed to jam model understanding by injecting coherent but conflicting context before each target sentence. For every sentence containing a potential idiomatic expression (PIE), we construct variants that deliberately invert contextual expectations: placing literal cues before idiomatic uses and idiomatic cues before literal ones. All variants are validated by human annotators to ensure naturalness and unambiguous interpretation for human readers. ID10M-JAM exposes systematic vulnerabilities in LLMs’ contextual reasoning, pushing idiom identification to its breaking point.

Evaluating conversational therapist competence in Motivational Interviewing (MI) requires multi-session assessment, yet current benchmarks rely on simulated patients that fail to model progression from session to session. We present MuSe-MI (Multiple Sessions Motivational Interviewing), a simulated patient framework grounded in Motivational Enhancement Therapy (MET), a multi-session clinical protocol of MI. MuSe-MI regulates dialogue generation through four evolving internal states from the MET clinical manual (anger, readiness to change, problem recognition, and self-efficacy), paired with an inter-session life event simulator and structured memory. Blind evaluations by licensed clinical psychologists confirm that patient dialogue reflects these internal states with high inter-rater agreement. Experiments show that MuSe-MI reliably distinguishes conversational therapist MI adherence levels based on patient language progression, establishing an automated multi-session evaluation benchmark.

Idioms pose a fundamental challenge for language models, as their meaning cannot be inferred from surface form alone. Understanding such expressions, therefore, requires semantic abstraction beyond lexical overlap. We introduce IdioLink, a retrieval benchmark designed to test whether models can link idiomatic expressions to conceptually equivalent meanings expressed in literal or paraphrased forms. IdioLink comprises 10,700 documents and 2,140 queries, spanning 107 idioms with both literal and figurative uses. Each document and query is annotated with spans that convey the core meaning. Evaluating strong embedding baselines (e.g., BGE, E5, Contriever, and Qwen), we show that current models struggle to retrieve equivalent meanings across divergent surface realizations, relying instead on topical and shallow semantic cues. IdioLink exposes key gaps in idiom-aware semantic retrieval and provides a challenging testbed for future models.

Large language models (LLMs) default to figurative interpretations of idiomatic expressions when context supports them. We investigate whether this bias can be directly controlled via activation steering: asking whether idiomaticity is encoded as a manipulable direction in residual space. We introduce IdioSteer, a small controlled benchmark of 100 sentences across 20 idioms designed to expose interpretation-dependent generation. Using a steering vector applied to five LLMs from two model families, ranging from 2B to 12B parameters, we show that steering systematically shifts model outputs toward literal interpretations while preserving fluency. Three findings stand out: cross-layer steering, vectors from mid-to-late layers injected at early layers - consistently outperforms same-layer interventions; steering vectors transfer to held-out idioms; and effects replicate across architectures and model sizes using the same intervention protocol.

We investigate cross-lingual fine-tuning for idiomatic expression identification, addressing the limited availability of annotated data in many languages. We evaluate encoder and generative decoder models to examine their ability to generalize idiom identification across languages. Additionally, we conduct an explainability study using linear probing and LogitLens to analyze how idiomatic meaning is represented across model layers. Results show consistent cross-lingual transfer, with English emerging as a strong source language. All code and models are released to support future research.

We assess the capabilities of large language models on tasks involving Buddhist texts written in Sanskrit and Classical Tibetan—two typologically distinct, low-resource historical languages. To this end, we introduce DharmaBench, a benchmark suite comprising 13 classification and detection tasks grounded in Buddhist textual traditions: six in Sanskrit and seven in Tibetan, with four shared across both. The tasks are curated from scratch, tailored to the linguistic and cultural characteristics of each language. We evaluate a range of models, from proprietary systems like GPT-4o to smaller, domain-specific open-weight models, analyzing their performance across tasks and languages. All datasets and code are publicly released, under the CC-BY-4 License and the Apache-2.0 License respectively, to support research on historical language processing and the development of culturally inclusive NLP systems.

We investigate the identification of idiomatic expressions—a semantically non-compositional subclass of multiword expressions (MWEs)—in running text using large language models (LLMs) without any fine-tuning. Instead, we adopt a prompt-based approach and evaluate a range of prompting strategies, including zero-shot, few-shot, and chain-of-thought variants, across multiple languages, datasets, and model types. Our experiments show that, with well-crafted prompts, LLMs can perform competitively with supervised models trained on annotated data. These findings highlight the potential of prompt-based LLMs as a flexible and effective alternative for idiomatic expression identification.