Processing Semantic Ambiguity: Different Loci for Meanings and Senses

Processing Semantic Ambiguity: Different Loci for Meanings and Senses Jakke Tamminen (jjt2@york.ac.uk) Alexandra A. Cleland (aac5@york.ac.uk) Philip T. Quinlan (ptq1@york.ac.uk) M. Gareth Gaskell (mgg5@york.ac.uk) University of York, Department of Psychology York, YO10 5DD, United Kingdom Abstract Semantic ambiguity can occur either as a consequence of ambiguity between unrelated meanings (e.g. bank) or ambiguity between related senses (e.g. hook). Early research did not distinguish between the two, finding that ambiguous words were recognized faster than unambiguous words. More recently it has been shown that words with many meanings suffer from a disadvantage in terms of recognition times, and words with many senses enjoy an advantage over their unambiguous counterparts. We report an auditory lexical decision study in which we apply the Psychological Refractory Period (PRP) logic to investigate the loci of the two types of ambiguity, and argue that they operate at different levels in the word recognition system. Keywords: Psycholinguistics; speech; semantic access Introduction Many words have multiple meanings, that is, they are semantically ambiguous. For example, bank can refer equally to a financial institution or to the margin of a river. In such cases the context in which the word occurs can be used to disambiguate the meaning. Evidence has been accumulating in the past decades that, when presented alone, ambiguous words are recognized faster than unambiguous words (see e.g. Rubenstein, Garfield, & Millikan, 1970, for an early demonstration of this with visual lexical decision). Accounts of this effect have relied on the assumption that different meanings of an ambiguous word have separate representations in the lexicon. One such theory was put forward by Jastrzembski (1981), who suggested that, since an ambiguous word has several representations, one of them will be likely to reach a recognition threshold before the single representation of an unambiguous word. There is, however, some recent evidence suggesting that the issue may be more complicated. While the two meanings of bank are clearly semantically unrelated, the majority of ambiguous words have multiple meanings that are closely related. To put it differently, it is common for a word to have many related senses (also referred to as polysemous words, as opposed to homonymous words which have many unrelated meanings). An example of a word with many senses would be hook, which can refer to a piece of fishing equipment, as well as to a sharp metal bend, or the act of connecting something. Most past research into semantic ambiguity has not made a distinction between unrelated meanings and related senses, yet the difference between the two concepts seems important. Thus it is possible that the two types of semantic ambiguity have different consequences for word recognition. Rodd, Gaskell, and Marslen-Wilson (2002) reported visual and auditory lexical decision experiments where they orthogonally manipulated both variables in a 2 x 2 design. Surprisingly, these authors found an ambiguity disadvantage, in that words with many unrelated meanings were recognized more slowly that words with few unrelated meanings. Furthermore, words with many related senses resulted in faster recognition times than words with few senses. Beretta, Fiorentino, and Poeppel (2005) sought to replicate these findings, and to shed more light on their underlying processes by using magnetoencephalography (MEG). Their behavioral data replicated those of Rodd et al., by showing a senses advantage and a meanings disadvantage. The MEG data showed an interesting effect on the M350 component. This is a component that peaks about 350 ms after stimulus onset, and has been demonstrated to be sensitive to the initial activation of lexical hypotheses. For example, the latency of the M350 has been found to be shorter for high-frequency words than to low-frequency words (Embick, Hackl, Schaeffer, Kelepir, & Marantz, 2001). The M350 was found to be slower to words with many unrelated meanings than few meanings. The opposite pattern was found for senses; words with many senses had shorter M350 latencies than words with few senses. These results suggest that previous research has confounded the number of unrelated meanings and related senses, and that the ambiguity advantage observed in previous studies may have in fact been caused by items with many senses. Rodd, Gaskell, and Marslen-Wilson (2004) proposed a distributed connectionist model that could explain the senses advantage and the meanings disadvantage. This model explains the meanings disadvantage by postulating separate representations for the different meanings within semantic space. An ambiguous word results in a blend state of activation between the meanings, and the network then needs to move away from the blend state and settle on one meaning. It is this process of competition between the semantic representations that accounts for longer recognition times for ambiguous words.