Running Head: Phonological Inference in Korean PHONOLOGICAL INFERENCE IN WORD RECOGNITION: EVIDENCE FROM KOREAN OBSTRUENT NASALIZATION

Gaskell and Marslen-Wilson (1996) use data from cross-modal priming to show that word recognition involves phonological inference: listeners more readily recognize a word that is changed from its canonical form if that change is conditioned by a phonological process. Subsequent research has questioned whether word recognition does in fact involve phonological inference, based on evidence that perceptual compensation for assimilation can involve universal, rather than language-specific mechanisms (Gow 2003) and on evidence that changes are accepted even outside of the context in which they are phonologically conditioned (Wheeldon and Waksler 2004). We present new evidence for phonological inference based on a cross-modal priming study with Korean listeners, which shows that they accept an obstruent-tonasal change (e.g. [sok] as [soŋ] ‘inner part’) only when that change occurs in the context of a following nasal (e.g. [soŋmaɨm] ‘innermost feelings’), the environment in which it occurs in Korean phonology. We propose that listeners perform a phonological viability check, in which lexical hypotheses are submitted to the phonological grammar to determine if they are viable given the phonological context. We further suggest that a probabilistic viability check can explain apparent differences in robustness of the inference effect across different types of phonological process. Phonological Inference in Korean INTRODUCTION Gaskell and Marslen-Wilson (1996) (henceforth GM-W) present results from cross-modal priming experiments showing that English listeners accurately recognize a word whose final consonant’s place of articulation has been altered to have same place of articulation as a following consonant. For example, [grim], with a final labial replacing the coronal of [grin], is recognized as a variant of green when it is followed by another labial (e.g. in [grimbin] green bean). GM-W’s results indicate that this effect is contingent on phonological context; [grim] would be less likely to be recognized as green when a non-labial consonant follows (e.g. [grimgræs] for green grass). Since the conditions under which changes in place of articulation are tolerated in word recognition correspond to the conditions under which the phonological process of place assimilation occurs in English (coronals assimilate to following non-coronals), GM-W take these results as evidence that listeners perform phonological inference in word recognition. That is, the word recognition process involves a calculation of whether a discrepancy from the canonical form of a word is due to the application of a phonological change. Subsequent research has questioned whether word recognition involves phonological inference. Gow (2003) argues that listeners cope with this sort of variation in the form of words with universal (non-language-specific) perceptual mechanisms. Under Gow’s feature parsing theory, the listener attributes the presence of acoustic cues to labiality in [grim] to the presence of the following labial consonant in [grimbin]. Gow supports this account by showing that partially assimilated segments are interpreted on the basis of the following consonant: a consonant intermediate between [n] and [m], which can be interpreted as either [n] or [m] in isolation, is perceived as [m] before a word like tables because the coronal cues are attributed to the [t], and as [n] before a word like buns whose initial consonant provides the source of labiality. The interpretation of the ambiguous consonant as [m] before tables cannot be the undoing of a phonological process, since English does not have labial-to-coronal assimilation (see further Gow and Im 2004 and Mitterer et al. 2006ab). A different challenge is provided by Wheeldon and Waksler (2004), who argue for a phonological underspecification account of listeners’ tolerance for variation in the phonetic shape of words (following Lahiri and Marslen-Wilson 1991). In this view, [grimbin] is acceptable because the lexical representation of the final coronal in green lacks a specification Phonological Inference in Korean for place of articulation (i.e. /griN/, where N is a nasal unspecified for place). They support this account with results from a new version of GM-W’s priming study, which replicated G-MW’s finding that tolerance for place changes is limited to word-final coronals, but failed to replicate the context effect: their results indicate that listeners would equally accept [grim] as green in both [grimgræs] and [grimbin]. They argue that these findings support an analysis in which English (word-final) coronals are underspecified for place, since under this account the acceptability of [grim] for /griN/ does not depend on whether there is a following labial (see also Lahiri and Reetz 2002). There has also been research subsequent to GM-W that favors the phonological inference account of their findings (see Gaskell and Snoeren in press for a recent summary). For example, Darcy et al. (2007) tested subjects whose native language phonologies contain different processes: English subjects whose phonology includes the above-mentioned place assimilation, and French subjects whose phonology includes voicing assimilation, in which an obstruent takes on the voicing specification of the following consonant (e.g. [rob] robe vs. [ropsæl] robe sale). French lacks place assimilation, and English lacks voicing assimilation. The results from a wordspotting task provide evidence that tolerance for changes in the shape of words is languagespecific and sensitive to phonological context (though they do note that there is also some evidence for language-independent compensation strategies). Language specificity is not expected under Gow’s perceptual compensation account, and context-sensitivity is not expected under the “pure” version of the underspecification account that Wheeldon and Waksler (2004) adopt. Because the evidence for contextand language-specificity is inconsistent across studies, and because Darcy et al. used a relatively off-line word-spotting task, the role of phonological inference in word recognition remains to be firmly established. In this paper, we discuss further evidence of phonological inference involving a phonological process in Korean. In Korean obstruent nasalization, obstruents become nasals in the presence of a following nasal. Examples of this process are provided in (1). The underlying forms in slashes show the forms as they would be produced in isolation (e.g. [pap] ‘rice’). The forms in phonetic square brackets show how the words are produced when concatenated into a compound: the word final obstruents are produced as nasals due to the presence of the following word-initial nasals (e.g. [pam]). Note that Phonological Inference in Korean if they do not undergo the independent process of place assimilation (see section 2 below), the nasal-assimilated segments retain their original place of articulation, as shown especially in (1d). (1) a. /pap + mul/ [pammul] ‘rice water’ b. /pat + noŋsa/ [pannoŋsa] ‘field farming’ c. /os + noŋ/ [onnoŋ] ‘a clothes chest’ d. /kuk + mul/ [kuŋmul] ‘broth’ In an earlier study reported in part in Lee (2005ab), we used a word-spotting methodology adapted from Darcy et al. (2007), and found that Korean speakers, but not English speakers, accept the realization of an obstruent as a nasal in the context of a following nasal, but not in the context of other following consonants. We discuss the results of that study in the next section, “Word Spotting and Korean Nasalization”. While these results do suggest that phonological inference is at work in Korean speakers’ recognition of word-final obstruent-nasal alternations, this interpretation is subject to the same caveat as the similar interpretation of Darcy et al.’s results from English and French. Word spotting, especially the variant we employed, is a relatively off-line task, and it could be that listeners are able to conduct phonological inference under those conditions, but do not employ it in on-line speech recognition. We therefore conducted a cross-modal priming study using a methodology adapted from GM-W. Again, we found that Korean subjects recognize words whose final obstruents are realized as nasals if and only if the following nasal context is present. These new results are presented in the section entitled “Cross-modal priming and Korean nasalization”. Given what seems to be incontrovertible evidence of phonological inference in word recognition, an adequate model of speech recognition must include a mechanism by which listeners assess the phonological viability of lexical hypotheses. In the “Discussion” section, we propose that listeners determine the probability of hypothesized lexical representations given the perceived phonological string by submitting them to their phonological grammar. We formalize this phonological viability check in terms of Optimality Theory (OT: Prince and Smolensky 1993/2004), and show how a probabilistic version of this theory may explain the apparently Phonological Inference in Korean greater robustness of phonological inference in the case of Korean obstruent nasalization relative to English place assimilation. We conclude that word recognition involves both perceptual compensation of the type documented by Gow (2003) and others, as well as a mechanism of phonological inference as proposed in GM-W. WORD-SPOTTING AND KOREAN NASALIZATION In our earlier study, we investigated not only obstruent nasalization, but also place assimilation. As in English, in Korean a consonant often takes on the place of articulation of the immediately following one. Also as in English, coronal consonants assimilate to following velars (2a,b) and labials (2c,d). Korean, however, has an assimilation pattern missing from English: labials assimilate to following velars (2e,f). 1 (2) a. /pat + ko/ [pakko] ‘receive and’ b. /han + kaŋ/ [haŋgaŋ] ‘the Han river’ c. /kot + palo/ [koppalo] ‘straight’ d. /han + pən/ [hambən] ‘once’ e. /əp+ko/ [əkko] ‘bear on the back + conj.’ f. /namkɨk/ [naŋgɨk] ‘the South Pole’ Our experiment tested nine types of assimilation process, shown schematically in (3), where T = coronal, K = velar, P = labial, and N = nasal. The rules in (3a-c) schematize the Korean place assimilation processes in (2). Those in (3d-f) correspond to place assimilation processes that do not occur in Korean. The rules in (3g-j) correspond to the obstruent nasalization patterns shown in (1). 1 In these and further examples there are also instances of other phonological processes that we will not discuss. Korean has coda neutralization: obstruents in coda position are produced as unreleased lax stops (these can be underlyingly plain (e.g. /p/), aspirated (/p/) or tensed (/p’/). Further, obstruents are realized as voiced between