idiomil

Idiomil is a hypothetical linguistic concept used to quantify the idiomaticity of multiword expressions in natural language. The term is not part of an established metric, but appears in discussions of figurative language and computational modeling as a way to compare how entrenched or conventionalized an expression is across languages or corpora.

Definition and scope: An idiomil is defined as a dimensionless score on a 0 to 1 scale. A score near 0 indicates that the expression is largely compositional, with its meaning deducible from its parts. A score near 1 indicates high idiomaticity or opacity, where the figurative meaning cannot be reliably inferred from the literal meanings of the components.

Measurement: Proposals for estimating idiomil combine corpus-based statistics, human judgments, and model-based estimates. Common approaches include crowd-sourced ratings of figurativeness, contextual surprisal analyses, and comparisons of semantic vectors for parts versus whole expressions. Because idiomil is not standardized, different studies may adopt different scoring procedures and thresholds.

Applications: Idiomil concepts are used to evaluate language models on idiom handling, to study cross-linguistic variation of idioms, and to inform translation systems about potential translation risk. It can help identify expressions that require special treatment in natural language processing pipelines.

Limitations: The concept faces challenges such as subjectivity in judgments, dependence on context, domain and register variation, and limited cross-language data. As a relative, not absolute, measure, idiomil is best used for comparative analysis rather than absolute classification.

See also: Idiom, multiword expression, figurative language, compositionality, lexicalized expression.