We have implemented an incremental lexical acquisition mechanism that learns the meanings of previously unknown words from the context in which they appear, as a part of the process of parsing and semantically interpreting sentences. Implement at ion of this algorithm brought to light a fundamental difference between learning verbs and learning nouns. Specifically, because verbs typically play the predicate role in English sentences, whereas nouns typically function as arguments, we found that different mechanisms were required to learn verbs and nouns. Because of this difference in usage, our learning algorithm formulates the most specific hypotheses possible, consistent with the data, for verb meanings, but the most general hypotheses possible for nouns. Subsequent examples may falsify a current hypothesis, causing verb meanings to be generalized and noun meanings to be made more specific. This paper describes the two approaches used to learn verbs and nouns in the system, and reports on the system’s performance in substantial empirical testing.