Semantic similarity between noun or verb meanings is known to matter in the organization of the lexicon, in the interface between syntax and semantics, and in sentence processing. Over the last ten years, my students, colleagues, and I have explored in detail the role that semantic similarity plays across various areas of language structure and use, from lexical access to discourse processing, to event categorization, stressing the differences in the kinds of similarity involved or the role similarity plays in each of these areas. In this seminar, I intend to cover each of these areas; each lecture will center on a particular research question, but as I cover our lab’s research, I will introduce students to various techniques we have used, how semantic similarity was measured, as well as the required background knowledge.
« Polysemy vs. homonymy in lexical access and the mental lexicon »
The various meanings associated with a single word form can either be unrelated (in which case one talks of homonymy) or similar (in which case one talks of polysemy). The literature is unclear in whether the two kinds of words are represented in the same way in the mental lexicon. Part of the reason for this uncertainty stems from the fact that experiments exploring this issue do not always control for the joint effect of two critical factors (1) semantic relatedness and (2) meaning dominance (whether the two meanings are equally frequent). In this lecture, I present the results of several experiments that controlled for both factors and use a variety of techniques (continuous priming, masked priming, eye-tracked sentence reading) which suggest that polysemous words share semantic representations in a way homonyms do not.
« Semantic similarity and argument structure constructions »
It has been claimed that verbs that occur in the same syntactic environments tend to have similar meanings and that one verb tends to characterize that meaning (e.g., give for the ditransitive construction in English). In this lecture, I will look at the relation between semantic similarity and syntactic similarity (in terms of frequency of occurrence in argument structure constructions) using both corpus studies and syntactic priming studies and will present corpus studies that explore whether all argument structure constructions behave the same way (they do not!).
« Semantic similarity in grammar vs. sentence processing »
In this lecture, I will survey the dimensions of semantic similarity that matter for grammar (including lexical meaning) and sentence processing by comparing rules that map meaning onto syntax (alluded to in the second lecture) and various experiments that show how detailed knowledge of events or entropy affect predictions upcoming material. Possible explanations for these differences in the kind of semantic similarity that affects grammatical and processing behavior will be explored.
« Semantic similarity and event categorization »
This lecture looks at our recent research on the categorization of events below the verb sense (what we call micro-senses). We have used rating studies and experimental techniques to determine (1) whether speakers can categorize events on the basis of similarity of event participant properties below lexicographic entries (they do!) and (2) make hypotheses about what the bases of these micro-sense event categories are. We have tried to determine whether clustering (of sentences) based on computational measures of similarity of event participants (e.g., Latent Semantic Analysis) approximate human classification (they kinda do!).