The advantages and disadvantages of semantic ambiguity

The Advantages and Disadvantages of Semantic Ambiguity Jennifer Rodd (jenni.rodd@mrc-cbu.cam.ac.uk) MRC Cognition and Brain Sciences Unit 15 Chaucer Road, Cambridge, UK Gareth Gaskell (g.gaskell@psych.york.ac.uk) Department of Psychology University of York, York, UK William Marslen-Wilson (william.marslen-wilson@mrc-cbu.cam.ac.uk) MRC Cognition and Brain Sciences Unit 15 Chaucer Road, Cambridge, UK Abstract There have been several reports of faster lexical decisions for words that have many meanings (e.g., ring) compared with words with few meanings (e.g., hotel). However, it is not clear whether this advantage for ambiguous words arises because they have multiple unrelated meanings, or because they have a large number of highly related word senses. All current ac- counts of the ambiguity advantage assume that it is unrelated meanings that produce the processing benefit. We report two experiments that challenge this assumption; in visual and audi- tory lexical decision experiments we found that while multiple senses did produce faster responses, multiple meanings pro- duced a disadvantage. We discuss how models of word recog- nition could accommodate this new pattern of results. Introduction Many words are semantically ambiguous, and can refer to more than one concept. For example, bark can refer either to a part of a tree, or to the sound made by a dog. To under- stand such words, we must disambiguate between these dif- ferent interpretations, normally on the basis of the context in which the word occurs. However, ambiguous words can also be recognised in isolation; when presented with a word like bark we are able to identify an appropriate meaning rapidly, and are often unaware of any other meanings. Words can be ambiguous in different ways. The two mean- ings of a word like bark are semantically unrelated, and seem to share the same written and spoken form purely by chance. Other words are ambiguous between highly related senses, which are systematically related to each other. For example, the word twist can refer to a bend in a road, an unexpected ending to a story, a type of dance, and other related concepts. The linguistic literature makes a distinction between these two types of ambiguity, and refers to them as homonymy and polysemy (Lyons, 1977; Cruse, 1986). Homonyms, such as the two meanings of bark, are said to be different words that by chance share the same orthographic and phonologi- cal form. On the other hand, a polysemous word like twist is considered to be a single word that has more than one sense. All standard dictionaries respect this distinction between word meanings and word senses; lexicographers routinely de- cide whether different usages of the same spelling should cor- respond to different lexical entries or different senses within a single entry. Many criteria (e.g., etymological, semantic and syntactic) have been suggested to operationalise this distinc- tion between senses and meanings. However, it is generally agreed that while the distinction appears easy to formulate, it is difficult, to apply with consistency and reliability. People will often disagree about whether two usages of a word are sufficiently related that they should be taken as senses of a single meaning rather than different meanings. This suggests that these two types of ambiguity may be best viewed as the end points on a continuum. However, even if there is not a clear distinction between these two different types of ambigu- ity, it is important to remember that words that are described as ambiguous can vary between these two extremes. In this paper we will review the evidence on how lexical ambiguity affects the recognition of isolated words, and will argue that the distinction been these two qualitatively dif- ferent types of ambiguity has not been addressed. We then report two experiments that confirm the importance of the sense-meaning distinction, and show that in both the visual and the auditory domains the effects of word meanings and word senses are very different. The Ambiguity Advantage In early studies of semantic ambiguity, Rubenstein, Garfield, and Millikan (1970) and Jastrzembski (1981) reported faster visual lexical decisions for semantically ambiguous words than for unambiguous words. However, these studies did not control for the subjective familiarity of the words, and Gerns- bacher (1984) found no effect of ambiguity over and above familiarity. Since then, however, Kellas, Ferraro, and Simp- son (1988), Borowsky and Masson (1996) and Azuma and Van Orden (1997) have all reported an ambiguity advantage in visual lexical decision experiments using stimuli that were controlled for familiarity. Although there does seem to a consensus in the litera- ture that lexical ambiguity can produce faster lexical decision times, it is not at all clear what type of ambiguity is produc- ing the effect. Is it multiple meanings, or multiple senses that produces the advantage? One way of trying to answer this question is to examine the dictionary entries of the words used in these experiments. As described above, dictionaries make a distinction between words whose meanings are sufficiently unrelated that they are given multiple entries, and those that