This study examines frequency, contextual diversity, and contextual distinctiveness effects in predicting produced versus not-produced frequent nouns and verbs by early second language (L2) learners of English. The study analyzes whether word frequency is the strongest predictor of early L2 word production independent of contextual diversity and distinctiveness and whether differences exist in the lexical properties of nouns and verbs that can help explain beginning-level L2 word production. The study uses machine learning algorithms to develop models that predict produced and unproduced words in L2 oral discourse. The results demonstrate that word frequency is the strongest classifier of whether a noun is produced or not produced in beginning L2 oral discourse, whereas contextual diversity is the strongest classifier of whether a verb is produced or not produced. Post hoc tests reveal that nouns are more concrete, meaningful, imageable, specific, and unambiguous than verbs, which indicates that lexical properties may explain differences in noun and verb production. Thus, whereas distributional properties of nouns may allow lexical acquisition on the basis of association through exposure alone (i.e., nouns may adhere to frequency effects), the abstractness and ambiguity found in verbs make them difficult to acquire based solely on repetition. Therefore, verb acquisition may follow a principle of likely need characterized by contextual diversity effects.