SYNTHESIS NOTE
Language, Text, and Discourse Reasoning, Retrieval, and Evaluation

Can one model handle all types of figurative language?

Does treating metaphor, idioms, and irony as a single pragmatic reasoning task—rather than separate classification problems—offer a more unified and effective approach to figurative language understanding in LLMs?

Synthesis note · 2026-03-26
Where exactly do language models fail at structural language tasks?

The standard NLP approach to figurative language treats each device as a separate problem: metaphor detection, idiom classification, sarcasm identification, irony recognition. Each gets its own dataset, its own benchmark, its own model architecture. But this fragmentation misses a structural unity: all figurative language involves the same underlying operation — recovering literal meaning from non-literal expression.

The Diplomat dataset formalizes this reframing. Across 4,177 dialogues with 6,494 human-annotated answers, it treats metaphors, idioms, and puns as instances of a single pragmatic reasoning task: given a figurative expression in a situated context, what is the speaker actually communicating? (Diplomat: A Dialogue Dataset for Situated PragMATic Reasoning)

This matters because it changes what we think LLMs need to do. If figurative language is a collection of separate classification problems, then the solution is more training data per category. If figurative language is a unified pragmatic reasoning task, then the solution is better pragmatic inference — the ability to reason about what a speaker intends versus what they literally say.

Since Do large language models reason symbolically or semantically?, metaphor is a particularly interesting test case. Metaphor is decoupled semantics: using one domain's vocabulary (vehicles, journeys, combat) to illuminate another (relationships, careers, arguments). The decoupling that causes LLM reasoning collapse is the defining feature of metaphorical language. If the unified pragmatic framing holds, then improving LLM performance on figurative language requires improving their ability to handle semantic decoupling — which is precisely the dimension where current architectures struggle most.

The practical implication for literary analysis: rather than building separate tools for metaphor extraction, irony detection, and idiom interpretation, a unified pragmatic reasoner could approach the full range of figurative devices in literary texts through a single inference mechanism. Whether current LLM architectures can support this remains an open question — but the Diplomat dataset provides the evaluation framework.

Inquiring lines that use this note as a source 12

This note is a source for these synthesized inquiries. Follow a line forward into its question, or open it to trace back to all of its sources.

Related concepts in this collection 4

This note in its neighbourhood — explore the map, then jump to a related concept in the list below.

Concept map
13 direct connections · 98 in 2-hop network ·medium cluster Open in graph ↗

Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph

your link semantically near linked from elsewhere

Related papers in this collection 8

Papers most semantically related to this note, ranked by cosine similarity in the embedding space.

Original note title

figurative language can be treated as a unified pragmatic reasoning task rather than separate classification problems for metaphor idiom and irony