This paper explores the speech acts most frequently performed by learners of English as a foreign language (EFL) through the Dialogue Annotation and Research Tool (DART), a semi-automatic annotation tool. The study used learner corpus research and an English as the first language (L1) speaker corpus as a reference to compare the speech acts performed by the EFL learners. The study involved 90 EFL learners of four nationalities. A spoken learner corpus was built with the dyadic interlanguage English conversation register. The most frequently performed speech acts were generally different for the English L1 speakers and the EFL learners. The speech acts labeled as correctSelf, referProcess, expressNonAwareness, and stateReason were in the top 10 for EFL learners, whereas they were not in the top 10 for the English L1 speakers. This difference was caused by three factors: the English proficiency of the EFL learners, the frequently used formulaic phrases, and the task requirements. Moreover, the corpus annotation has a problem with unrecognized speech acts, primarily due to the redundancies and fragments in the learner discourses. The findings of this study reveal the complementary relationship between grammatical and pragmatic performances.