Collocation, known as words that commonly co-occur, is a major category of formulaic language. There is now general consensus among language researchers that collocation is essential to effective language use in real-world communication situations (Ellis, 2008; Nesselhauf, 2005; Schmitt, 2010; Wray, 2002). Although a number of contemporary speech-processing theories assume the importance of formulaic language to spontaneous speaking (Bygate, 1987; de Bot, 1992; Kormos, 2006; Levelt, 1999), none of them gives an adequate explanation of the role that collocation plays in speech communication. In the practices of L2 speaking assessment, a test taker’s collocational performance is usually not separately scored mainly because human raters can only focus on a limited range of speech characteristics (Luoma, 2004).

This paper argues for the centrality of collocation evaluation to communication-oriented L2 oral assessment. Based on a logical analysis of the conceptual connections among collocation, speech-processing theories, and rubrics for oral language assessment, the author formulated a new construct called Spoken Collocational Competence (SCC). In light of Skehan’s (1998, 2009) trade-off hypothesis, he developed a series of measures for SCC, namely Operational Collocational Performance Measures (OCPMs), to cover three dimensions of learner collocation performance in spontaneous speaking: collocation accuracy, collocation complexity, and collocation fluency. He then investigated the empirical performance of these measures with 2344 lexical collocations extracted from sixty adult English as a second language (ESL) learners’ oral assessment data collected in two distinctive contexts of language use: conversing with an interlocutor on daily-life topics (or the SPEAK exam) and giving an academic lecture (or the TEACH exam). Multiple regression and logistic regression were performed on criterion measures of these learners’ oral proficiency (i.e., human holistic scores and oral proficiency certification decisions) as a function of the OCPMs.

The study found that the participants generally achieved higher collocation accuracy and complexity in the TEACH exam than in the SPEAK exam. In addition, the OCPMs as a whole predicted the participants’ oral proficiency certification status (certified or uncertified) with high accuracy (Negelkerke R2 = .968). However, the predictive power of OCPMs for human holistic scores seemed to be higher in the SPEAK exam (adjusted R2 = .678) than in the TEACH exam (adjusted R2 = .573). These findings suggest that L2 learners’ collocational performance in free speech deserve examiners’ closer attention and that SCC may contribute to the construct of oral proficiency somewhat differently across speaking contexts. Implications for L2 speaking theory, automated speech evaluation, and teaching and learning of oral communication skills are discussed.