JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR 17, 375-401 (1978) Semantic Composition in Sentence Verification LANCE J. Rips University of Chicago EDWARD E. SMITH Stanford University AND EDWARD J. SHOBEN University of Illinois Verifying simple sentences generally involves a process wherein the meanings of individual words are combined to form the meaning of the entire sentence. In this paper we investigated the combination process by asking subjects to decide whether sentences of the form S-V-Adj-O (e.g., Pines have green needles) were true or false. Such sentences can be factored into three com- ponent propositions: (i) a proposition formed from the adjective and object noun (e.g., Needles can be green); (2) a proposition formed from the subject noun, verb, and object noun (Pines have needles); and (3) a proposition formed from the subject noun, verb, and adjective (Pines have green [parts]). In Experiment 1, we varied the truth value and semantic relatedness of Propositions (1) and (2) in an effort to determine which of these components plays a role in the semantic composition process. Reaction time results indicated reliable effects of truth and relatedness for both propositions. Experiment 2 employed the same task and extended it by testing all three of the above propositions. This time, the results exhibited significant effects of the truth of Propositions (2) and (3). Finally, Experiment 3 applied the same method to sentences of a new syntactic type: cleft sentences of the form It's-Adj-O-that-S-V (e.g., It's green needles that pines have). The effects of the truth value of Propositions (2) and (3) were once more in evidence, although syntactic form differentially influenced the size of these effects. To understand a novel sentence we must ordinarily access the meanings of its con- stituent words and combine them in a com- posite meaning for the sentence as a whole. The present paper focuses on this combination We are grateful for the advice and assistance of N. Adams, J. Angiolillo, D. Burke, H, Clark, J. Greeno, J. Huttenlocher, T. Landauer, S. Marcus, W. Marslen- Wilson, M. Nissen, M. Reiser, S. Schacht, and T, Schumacher. The research was supported by U.S. Public Health Service Grant MH-19705 and National Science Foundation Grant BNS76-0337. Address correspond- ence to: Lance J. Rips, Department of Behavioral Sciences, University of Chicago, 5848 S. University Avenue, Chicago, Illinois, 60637. 375 process, which we will hereafter call "semantic composition." We note that this process must be a critical part of language understanding; although it may sometimes be short-circuited by context (as when context alone provides the interpretation of an upcoming phrase), still the very fact that we can understand isolated sen- tences implies its use. For sentences that we have not yet encoded, the composition pro- cess must construct the underlying semantic representation in order for us to be able to store it in memory. But even in dealing with sentences whose meaning we already know--for example, in answering questions or 0022-5371/78/0174-0375 $02.00/0 Copyright (~ 1978 by Academic Press, Inc. All rights of reproduction in any form reserved. Printed in Great Britain