Idioms have generally played a supporting rather than a leading role in research on figurative language. In Cognitive Linguistics for instance idioms have been understood against how they are embedded in conceptual metaphors (Lakoff 1987, Women, fire, and dangerous things . Chicago: University of Chicago Press; Clausner and Croft 1997, Productivity and schematicity in metaphors. Cognitive Science 21. 247–282) while in the experimental psycholinguistic tradition their role has been to challenge the basis of conceptual metaphor in “priming” figurative language (Glucksberg et al. 1993, Conceptual metaphors are not automatically accessed during idiom comprehension. Memory and Cognition 21. 711–719; McGlone 2007, What is the explanatory value of a conceptual metaphor? Language and Communication 27. 109–206). It is, moreover, broadly assumed that criteria defining grammatical properties of idioms are limited to their morphological and syntactic behavior (Nunberg et al. 1994, Idioms. Language 70. 491–538). While the pragmatic properties of idioms have been described informally (Glucksberg 2001. Understanding figurative language: From metaphors to idioms (Oxford psychology series 36). Oxford: OUP), there are few studies which systematically contrast the behavior of nouns in literal vs. idiomatic expressions in discourse. Using a battery of criteria which has been developed to study discourse properties of subjects in spoken Arabic (Owens et al. 2013. Subject expression and discourse embeddedness in Emirati Arabic. Language Variation and Change 25. 255–285), we show that keyword nouns in Nigerian Arabic are significantly different according to whether they are idiomatic or literal. The basis of the conclusion is the statistical analysis of 1403 tokens derived from a large corpus of natural Nigerian Arabic texts. Nouns in idiomatic expressions are opaque to discourse in a way those in literal ones are not. To explain the statistical results we argue that idioms partake in a ‘semantic mapping’ which incorporates the noun and its collocate in the idiom into a word-like unit, rendering it largely invisible to subsequent discourse. Since Nigerian Arabic idiomatic nouns, as is shown, display no clause-internal syntactic constraints, exhibit no cross-clausal syntactic dependencies, and show no significant interactions with possessive pronouns which ostensibly appear to mark the discourse argument of the keyword they are suffixed to, it is concluded that the mapping is of semantic nature. Other than exemplifying basic facts obtained via elicitation, the entire argument hinges on an examination of nouns in actual spoken discourse. The article establishes that large corpora coupled with multivariate statistical treatment contribute directly to understanding semantic factors difficult to evaluate via direct elicitation or examination of individual examples, in this case the sensitivity of cross-clausal referentiality to idiomatic contextualization.