Multiple semantic encoding of homophones and homographs in contexts biasing dominant or subordinate meanings

ROBERT E. WARREN 0 1 NATALIE TERBOVIC WARREN 0 1 JOHN P. GREEN 0 1 JAN H. BRESNICK 0 1 0 Requests for reprints should be sent to Robert E. Warren, Bell Laboratories , 6200 East Broad Street, Columbus, Ohio 43213 . Some of the data from Experiment 1 were reported at the meeting of the Psychonomic Society, St. Louis, Missouri , November 1976. The authors wish to express their appreciation for access to the facilities of the Center for Cognitive Studies, University of Oregon , during testing for Experiment I 1 Columbia University , New York, New York 10027 The resolution of lexical ambiguity was studied in two experiments. While subsequent selection would appear to limit longer term storage to one meaning of an ambiguous word, multiple initial encoding of homophones and homographs was indicated by the intrusion of their alternative meanings across trials in a Brown-Peterson paradigm. In Experiment I subjects heard four words and then engaged in a IO-sec distractor task. Written recall was then demanded. On context trials a homophone with two distinct alternative meanings was presented with other words biasing either a dominant or subordinate meaning, for example, look, stare, peer, glance. On the next trial words related to an alternative meaning were presented, for example, dock, wharf, quay, jetty. Intrusions of the graphic forms of alternative meanings (e.g., PIER) into critical trial recall occurred whether the dominant or subordinate meaning of the homophone had been biased by context. Experiment 2 employed visual presentation of homographs, for example, GOLD, IRON, LEAD, and oral recall, but was logically similar to Experiment 1. Intrusions of the alternative codings of homographs across trials again occurred, for example, GUIDE, DIRECT, lead, regardless of the meaning originally biased by context. - Becker, 1976) assert that only one meaning of a word is ever active during encoding and that this meaning is completely determined by the context in which the word occurs. Second, multiple encoding models (e.g., Conrad, 1974; Foss & Jenkins, 1973; Warren & Warren, 1976) suggest that at some early stage of encoding all meanings of a word are active, and only at some later point is a selection based on context made. Finally, ordered activation models (Hogaboam & Perfetti, 1975) assert that when an ambiguous word is encountered its meanings are activated in order of their relative fre quency of usage (dominance) until a meaning that matches context is located. The most frequent (dom inant) meaning is examined first and if a match is made no further meanings become active. If, on the other hand, a less frequent (subordinate) meaning is the one to match context, it and all meanings more frequent than it will have been activated. When evaluating the various encoding hypotheses, one must distinguish three possible facets of the encod ing process: the activation of meaning representations, the processes which lead to a selection of one meaning over others, and the storage or other subsequent use of such selected representations. Evidence that only a single meaning of a word is available some time after process ing is complete does not preclude the possibility that multiple semantic representations of the word were active and available earlier. Work on the recognition of nouns presented in context by Light and Carter Sobell (1970), for example, shows that only a single representation of a word can be retrieved 6-7 min after presentation. These investigators presented words such as JAM in one context (e.g., strawberry JAM) and tested their recognition in other contexts (e.g., traffic JAM). The presence of only one meaning representation was indicated by reduced performance under these condi tions compared to those in which recognition context biased the same meaning (e.g., grape JAM). This, how ever, can indicate only the end state of the encoding process and does not necessarily imply an equally restrictive representation of alternative meanings during initial encoding. Similarly, evidence that the level of general processing capacity available for other tasks is affected by the presence or absence of ambiguity may only reflect changes in those portions of the encoding process that require such capacity. It might be the case, for example, that the activation of the meaning representations of words in memory does not require capacity, whereas the selection of one meaning from a number which are activated simultaneously does take capacity. Studies by Foss and his colleagues (Foss, 1970; Foss & Jenkins, 1973) have used a phoneme monitoring task as a measure of the processing demands of lexical items embedded in sentences. Subjects in the experiments listened to sentences in which a target phoneme followed either an ambiguous or unambiguous word. Reaction time to detect the phoneme was longer following ambig uous words, indicating that such words increased local processing demands. Foss argued that the increased processing demands of ambiguous words were a by product of the activation of their multiple meaning representations, and demonstrated that such effects were not changed by the introduction of contextual material that resolved the lexical ambiguity. Subsequent research using the same paradigm, however, has shown that sufficiently strong contexts eliminate the ambiguity effect (Swinney & Hakes, 1976). In any case, these results may only reflect changes in the ease with which selection proceeds and leave open the question of whether single or multiple meanings are activated initially. In an attempt to study the very early stages of the encoding of ambiguous words, Conrad (1974) presented subjects with sentences in which the final word was a homonym. Words preceding the homonym were either neutral with respect to its various possible meanings or related to one particular meaning, for example, "The sailors enjoyed the port" or "The sailors drank the port." Immediately following the last word in the sentence, subjects were required to name the color of the ink in which a visual test word, such as WINE or HARBOR, was printed. Warren (1972, 1974) has shown that the latency of color naming in this situation increases if subjects have encountered material semantically related to the word serving as the carrier for the color just prior to its presentation, relative to latencies observed when it is tested after unrelated material. Conrad (1974) found that both WINE and HARBOR produced such delays in color naming when tested after either the ambiguous or unambiguous presentation of PORT. Since associates of both meanings of PORT appear to have been affected equally, regardless of contextual information, Conrad suggested that, at least initially, both meanings of a homonym are available during its encoding. On the other hand, Schvaneveldt, Meyer, and Becker (1976) found no indication that multiple meanings of homonyms were activated during encoding. These investigators presented t (...truncated)