The paper investigates how metaphoric concepts are encoded in large language models (LLMs) by applying mechanistic interpretability methods to identify causal subspaces across various architectures and sizes. It demonstrates that manipulating these subspaces allows steering of metaphor-related outputs on both classification and generation tasks, effectively controlling the metaphoricity of LLM responses.
arXiv:2510. 04120v2 Announce Type: replace-cross Abstract: Large language models (LLMs) achieve strong performance on metaphor detection and interpretation tasks, yet it remains unclear what such behavioral success reveals about metaphor processing.
arXiv:2601. 03388v3 Announce Type: replace-cross Abstract: Earlier research has shown that metaphors influence human decision-making, raising the question of whether metaphors also influence large language models (LLMs)' reasoning pathways, given that their training data contain a large number of metaphors.
arXiv:2608. 15828v1 Announce Type: cross Abstract: Current evaluation of metaphor explanations relies mainly on holistic quality ratings, revealing little about how explanation quality is structured or where human judgments agree and diverge.
arXiv:2607. 28683v1 Announce Type: cross Abstract: Large language models benefit from elements in natural language, such as metaphors and analogies in training data and inference input to achieve generalisability across different domains.
arXiv:2606. 18922v1 Announce Type: cross Abstract: Figurative language and negation are two areas that challenge current language models, however, both are widely used throughout written and spoken language.
Similes provide a compact and expressive way to describe visual characteristics in text prompts. Recent text-to-image models (t2i models) can produce visually compelling outputs from simile prompts, yet even frontier models frequently misinterpret the metaphorical vehicle and confuse it with the object.
The paper introduces VMetaphor-Bench, a benchmark for evaluating visual metaphor generation in text-to-image models, comprising 1,500 curated metaphors across three levels and ten categories, each paired with two prompts of varying specificity. It proposes a hybrid evaluation framework using a multiple-choice question protocol and dimension-based scoring to assess metaphorical fidelity. Experiments on 11 T2I models show that even top proprietary models struggle with compositional structuring and cross-domain mapping, underscoring the need for further research in this area.
arXiv:2605.05593v2 Announce Type: replace Abstract: Despite the remarkable success of Multimodal Large Language Models (MLLMs) across diverse tasks, the internal mechanisms governing how they encode...
The paper introduces VMetaphor-Bench, a benchmark for assessing visual metaphor generation in text-to-image models. It contains 1,500 curated metaphors across three levels and ten categories, each paired with two prompts of varying specificity. The authors evaluate 11 T2I models using a hybrid MLLM-as-judge framework that combines a large multiple-choice question set with dimension-based scoring, finding that even top proprietary models struggle with compositional structuring and cross-domain mapping.
MetaHOPE is an error‑severity‑aware annotation framework designed to evaluate how well machine translation (MT) and large language models (LLMs) translate metaphors. The authors applied MetaHOPE to three state‑of‑the‑art systems—GoogleMT, GPT5.4, and Hunyuan‑7b—using two human‑annotated metaphor corpora (VUAMC and PSUCMC) for English‑to‑Chinese and Chinese‑to‑English translation. They also produced a bilingual post‑edited gold reference, creating a new resource for metaphor translation research.
The paper investigates whether language models possess latent rhetorical abilities, specifically analogy reasoning, before instruction tuning. By extracting an "Analogy Vector" from the residual stream, the authors demonstrate that this vector can steer explanation style across several instruction-tuned models, improving performance metrics. Their findings reveal a continuous gradient of behavioral accessibility tied to the amount of instruction exposure during training, challenging the binary base‑vs‑instruct distinction and suggesting that instruction exposure shapes rhetorical knowledge in activation space.
arXiv:2603. 19066v2 Announce Type: replace-cross Abstract: Four-term word analogies (A:B::C:D) are classically modeled geometrically as parallelograms: adding the vector B-A+C produces D.