Background: phonotactic gaps The study Results Discussion Summary

Hidden knowledge of gap wellformedness

James Kirby and Alan C. L. Yu

Phonology Lab Department of University of Chicago

Experimental Approaches to Optimality Theory Ann Arbor 19 May 2007 Background: phonotactic gaps The study Results Discussion Summary Outline

1 Background: phonotactic gaps

2 The study

3 Results

4 Discussion

5 Summary Background: phonotactic gaps The study Results Discussion Summary Not all gaps are the same

Phonologists traditionally recognize a distinction between accidental and systematic gaps in the (Fischer-Jørgensen 1952; Halle 1962) A is deemed descriptively and explanatorily adequate if and only if it ... 1 admits possible (attested words and accidental gaps), and 2 rules out impossible words (systematic gaps). Background: phonotactic gaps The study Results Discussion Summary Not all gaps are the same

Accidental gaps don’t violate any phonotactic restrictions Systematic gaps entirely absent from the language because of well-defined phonotactic constraints...... or whatever the phenomenon of interest happens to be (e.g. OCP-PLACE violations: Frisch et. al. 2004) Background: phonotactic gaps The study Results Discussion Summary Not all gaps are the same

Traditional grammatical approaches presume a categorical distinction between systematic vs. accidental gaps. all systematic gaps are equally ill-formed; all accidental gaps are equally well-formed. Predicts wellformedness judgments to be categorical as well. Background: phonotactic gaps The study Results Discussion Summary Not all gaps are the same

But: not all unattested words are judged identically. Acceptability of unattested words is gradient (Ohala & Ohala 1986; Coleman & Pierrehumbert 1997; Frisch et. al. 2000; Albright & Hayes 2003; Hay et. al. 2004, etc.) Acceptability reflected in statistical properties of the lexicon (n-gram probabilities, neighborhood density, O/E ratios, etc.) Background: phonotactic gaps The study Results Discussion Summary Inner structure of systematic gaps?

Accidental gap acceptability is gradient What about systematic gaps? Prevailing view seems to be that systematic gaps are categorically ill-formed. Background: phonotactic gaps The study Results Discussion Summary Systematic gap → categorical illformedness

Certain OT theorists have proposed nonce wellformedness is determined by the relative ranking of markedness constraints (Boersma 1997; Hayes 2000; Zuraw 2000, 2002; Boersma & Hayes 2001; cf. Anttila 1997a, 1997b; Anttila & Andrus 2006). Statistical patterns in the lexicon may be reflected in the analysis in the form of stochastically ranked constraints. Gradient well-formedness reflects probability of the winning output. Background: phonotactic gaps The study Results Discussion Summary Systematic gap → categorical illformedness

Gradedness is not possible without variation “Our basic premise, then, is that intermediate well-formedness judgments often result from grammatically encodeable patterns in the learning data that are rare, but not vanishingly so, with the degree of ill-formedness related monotonically to the rarity of the pattern.” (Boersma & Hayes 2001:73) Background: phonotactic gaps The study Results Discussion Summary Systematic gap → categorical illformedness

Frisch et. al. (2004); Coetzee & Pater (2006): strength of OCP-PLACE restriction correlated with segment similarity (operationalized as observed/expected values) Zero-frequency items are assigned O/E = 0.0 Doesn’t make a prediction about the range of acceptability within the set of zero-frequency items (but cf. Wilson & Hayes 2006, Pater 2007) Background: phonotactic gaps The study Results Discussion Summary Systematic gap → categorical illformedness

Difference in judgments are task-specific (Berent & Shimron 1997; Coetzee to appear) Gradience predicted only when comparing two nonwords In non-comparative wordlikeness judgment tasks, judgments were categorical, with no intra-group variation Within-group acceptability variation was observed only in the comparative task Prediction: gradienc only in comparative wordlikeness tasks. Background: phonotactic gaps The study Results Discussion Summary Systematic gap → categorical illformedness

Other studies focus on gaps which do not violate language Frisch, Large, & Pisoni (2000); Bailey & Hahn (2001): designed not to contain any over phonotactic violations Albright & Hayes (2003): contained only one obvious phonotactic violation [bzarSk] Iverson & Salmons (2005): investigate subtypes of accidental gaps These studies thus have very little to say about acceptability of systematic gaps. Background: phonotactic gaps The study Results Discussion Summary Research questions

Few (no) studies have examined variation within systematic gaps.

1 Is systematic gap acceptability gradient? 2 Is it task-specific? 3 Is it supported by lexical statistics? Background: phonotactic gaps The study Results Discussion Summary Our study: Cantonese

Why Cantonese? Well understood historical phonotactic gaps. Highly restricted syllable phonotactics Related work on Mandarin (Myers 2002; Myers & Tsay 2004, 2005) Background: phonotactic gaps The study Results Discussion Summary Cantonese phonotactics

Yue dialect spoken in Hong Kong, Guangdong province, diaspora. (C)(G)V(V)(C) syllable structure 19 onsets: /p ph t th ts tsh k kh kw kwh m n N f s h l j w/ 6 codas: /p t k m n N/ 8 monophthongs: /a: a E: i: O: ø: u: y:/ 11 diphthongs: /ai 5i au 5u ei Eu 8y Oi ui iu ou/ 6 tones: /55 25 33 21 23 22/ Due to various documented sound changes, several syllable combinations are not possible Background: phonotactic gaps The study Results Discussion Summary Systematic gaps

Labial gaps

Labial onsets do not occur in with labial codas (*pap) Labial codas do not occur with rounded vowels (*um) Labial onsets do not occur with front rounded vowels (*my)

Onset-tone gaps

Aspirated onsets do not occur with 22 tone (*pha22) Unaspirated onsets do not occur with 11 or 23 tones (*pa11)

Coronal-vowel gaps

Coronal onsets and codas may not co-occur with the nuclei /o, u/ (*ton) Coronal onsets also do not occur with the vowel /u/ (*tup) Background: phonotactic gaps The study Results Discussion Summary Experimental corpus

432 items conforming to a CV(C) template, derived from all possible combination of eight onset phonemes /f, p, ph, m, s, t, th, n/; three vowel phonemes /a:, i:, u:/; three codas /m, n/ and ∅; six tones /55, 25, 33, 21, 23, 22/ This resulted in 162 attested syllables and 270 nonwords. Background: phonotactic gaps The study Results Discussion Summary Nonwords

Of the nonwords, 61 fill labial dissimilation gaps; 36 fill onset-tone gaps; 42 fill coronal gaps; 27 syllables filled two types simultaneously, and 1 all three. The remaining 103 syllables were judged to be accidental gaps because they did not violate any phonotactic constraintsper se Background: phonotactic gaps The study Results Discussion Summary Procedure

Ten native speakers of Cantonese were presented with a randomized series of items from the corpus and given two tasks per stimulus:

Lexical decision task “Is this a word of Cantonese?” (yes/no)

Wordlikeness rating task “How good a word of Cantonese is this?” (1-7)

Items judged on a 7-point scale, with 1 indicating “very poor - highly unlikely to be a real word of Cantonese” and 7 indicating “very good - a highly prototypical Cantonese word” Background: phonotactic gaps The study Results Discussion Summary Results 0.6 Cor=Coronal Mul=Multiple Lab=Labial

0.4 Acc=Accidental Tone=Onset−Tone Lex=Lexical 0.2 0.0 Zscorearcsin −0.2 −0.4 −0.6

Att Tone Acc Lab Mul Cor

gap type Background: phonotactic gaps The study Results Discussion Summary Wilcoxon rank sum tests (α = 0.016) 0.6 Lex=Lexical Tone=Onset−Tone Acc=Accidental

0.4 Lab=Labial Mul=Multiple Onset-tone Cor=Coronal

0.2 µ = 0.018 σ = 0.578 0.0 Accidental Zscorearcsin −0.2 µ = 0.28

−0.4 σ = 0.46

−0.6 Significant

Att Tone Acc Lab Mul Cor p = 0.0015

gap type Background: phonotactic gaps The study Results Discussion Summary Wilcoxon rank sum tests (α = 0.016) 0.6 Lex=Lexical Tone=Onset−Tone Acc=Accidental

0.4 Lab=Labial Mul=Multiple Accidental Cor=Coronal

0.2 µ = 0.28 σ = 0.46 0.0 Labial Zscorearcsin −0.2 µ = −0.302

−0.4 σ = 0.367

−0.6 Not significant

Att Tone Acc Lab Mul Cor p = 0.0825

gap type Background: phonotactic gaps The study Results Discussion Summary Wilcoxon rank sum tests (α = 0.016) 0.6 Lex=Lexical Tone=Onset−Tone Acc=Accidental

0.4 Lab=Labial Mul=Multiple Labial Cor=Coronal

0.2 µ = −0.302 σ = 0.367 0.0 Coronal Zscorearcsin −0.2 µ = −0.713

−0.4 σ = 0.24

−0.6 Significant

Att Tone Acc Lab Mul Cor p < 0.001

gap type Background: phonotactic gaps The study Results Discussion Summary Lexical statistics

Phonotactic probability Speakers are sensitive to sequential segment probabilities Nonwords with high phonotactic probability should be judged as “more wordlike” Operationalized as average conditional probability, joint n-gram probability, n-gram frequency stated in terms of natural classes... Neighorhood density Hearing novel words activates a set of stored exemplars The more exemplars activated, the more the nonword resembles an existing word Operationalized as edit distance, weighted edit distance (GNM), others? Background: phonotactic gaps The study Results Discussion Summary Phonotactic probability

Phonotactic probability operationalized as average bigram log probability: length(W ) X P(W ) ≈ −log2 p(wi |wi−1) i=1 Zero-frequency bigrams were assigned a small non-zero probability Background: phonotactic gaps The study Results Discussion Summary Lexical neighorhood density

Neighborhood density (Greenberg & Jenkins 1964): number of lexical neighbors which differ by k changes (substitution, deletion, addition) Calculated using the Chinese Character Database (Kwan et. al. 2003) which covers a total of 13,060 character entries Weighted by token frequency in Hong Kong Cantonese Adult Language Corpus (HKCAC: Leung & Law 2001), containing 140,000 monosyllables drawn from around 8 hours of speech Background: phonotactic gaps The study Results Discussion Summary Phonotactic probability

Demonstrated as predictive of nonword acceptability judgments (Coleman & Pierrehumbert 1997; Frisch, Large, & Pisoni 2000; Vitevich & Luce 1998; etc.) Vitevitch & Luce 1998: claimed as dominant predictor Background: phonotactic gaps The study Results Discussion Summary Phonotactic probability

Not significant ● ● ● ● ● ● ● ● ●● ● ● ●●● ● ● ● ● ● ● ● ●● ●● 2 ● ●●● ● ●●● ● ● ● ●●● ● ●● ● ● Ra : -0.002 ●● ● ●● ● ● ●● ● ● words 1.0 ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● ● nonwords ● ● ● ● ●● ● ● ● ●●●● ● F(1, 430) = 0.1387 ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● p = 0.71

0.5 ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●●● ●● ● ● ●●● ●● ●● ● ● ● ●● ●● ● ● ● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● 0.0 ● ●● Words ● ● ●● ● ● ●●● ●●● ● ● ● ● Zscorearcsin ●● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●● ● ● ●● ● ● µ = 12.34 ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ●●● ●● ● ●● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● ● ● −0.5 ● ●●● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● σ = . ● ● ● ● ● ●● ● ● 2 79 ●● ● ● ● ● ● ● ● ● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ●●●● ●● ● ● ●● ●● ● ● ●● ● ● ● ● ● ●● −1.0 ● Nonwords 8 ● 10 12 14 16 18 20 µ = 13.11 ● average logP σ = 3.83 Background: phonotactic gaps The study Results Discussion Summary Neighborhood density

Lexical neighborhood density: dominant in the processing of real words (Vitevitch & Luce 1998; Bailey & Hahn 2001) Less effective a predictor for nonword wordlikeness judgments? Background: phonotactic gaps The study Results Discussion Summary Neighborhood density

Significant ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●●● ●● 2 ●● ●● ● ●● ●● ● = . ●● ●● ●● ● ● Ra 0 277 ● ●● ●●● ● ● ● 1.0 ● ● ●●●● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ●●●●● F(1, 430) = 166 ●● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ●●● ● ● ● ● ● ● p < 0.001

0.5 ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ●●●●● ● ● ●● ● ● ●● ● ● ●● ● ●●● ● ● ● 0.0 ● ● ● Words ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Zscorearcsin ● ● ● ● ● ●● ●●● ● ● ● ● ● ● ● ●● ● ● ●● ●● ●● ● ● ● ● ● ● ●● ● ●●● ●●●● ● µ = 179.45 ● ● ●● ● ● ● ● ● ● ●●● ● ● ●● ●●● ● ● ● ● ●● ● ●● ● ●● ●● ●● ●●●● ● ● ● ●● ● ●● ● ●●● ● ● −0.5 ● ●●●● ●● ● ●●● ●●●● ● ● ● ● σ = . ● ● ● ● ●● ● words 67 32 ● ● ●● ●● ● ● ● ● ● ● ● ● ●●●●● ●●● ● nonwords ● ● ● ● ●●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●● ● ●● ● ● ● ● ● −1.0 ● Nonwords ● 100 200 300 400 µ = 96.57 ● weighted neighborhood density σ = 63.07 Background: phonotactic gaps The study Results Discussion Summary Multiple regressions

Entire corpus 2 Ra = 0.3429 F(2, 429) = 113.4 p < 0.001

Words Nonwords 2 2 Ra = 0.052 Ra = 0.214 F(2, 159) = 4.43 F(2, 267) = 37.71 p = 0.013 p < 0.001 Only density is significant Both factors significant Background: phonotactic gaps The study Results Discussion Summary Phonotactics and neighborhood density

5 10 15 20 25 0 100 200 300 400 LAB MUL TONE LAB MUL TONE

● 1.0 ● 1.0 ● ● ● ● ● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● 0.5 ●● ● 0.5 ●● ● ●●● ● ● ● ● ●●● ●● ● ● ●●● ● ● ●● ●●●●●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ●●●●●● 0.0 ● ●●●● 0.0 ●● ● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ●● ●● ● ● ● ● ● ●● ●●● ●● ● ● ● ●● ●● ●●●●● ●● ● ● ● ●●●● ● ● ● ● ● ● ● ● ● ● ● −0.5 ●●● ●● ● −0.5 ● ●● ● ● ● ● ●●●● ●● ● ●● ●●●● ● ●● ● ● ● ● ● ● ● ●● ●●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ●● ● ●● ● ● −1.0 ● ● ● ● ● −1.0 ● ● −1.5 −1.5 ACC ATT COR ACC ATT COR ●● ●●●●● ● ●●●●● ●● ●●●●●●●●● ● ● ●●● ●●●●●●●● ● ●●●●●●●●●●● ●●●● ●●●●●●●●● 1.0 ● ●●●●●●●● ●●●● 1.0 ● ●●●●●●● ●● ●●●● ● ● ●●●●●●● ● ●● ●●● ●● Zscorearcsin ● ●● ●●●● Zscorearcsin ●● ● ● ●●● ● ● ● ●●●●●●●● ● ● ●● ●●●●●●●●●●●●● ●● ●●● ● ● ●●● ● ● ●● ●● ● ● ● ●● ● ● ● ● ●●● ● ● ● 0.5 ● ●●● 0.5 ●● ● ● ●● ●●●● ●●●● ● ● ●● ●●● ●● ●● ● ● ● ●● ● ●●● ● ●●●● ● ● ●●● ● ● ● ● ●●●● ● ● ● ●●● ● ● ●● ● ● ● ● ●●●● ●●● ●●● ● ● ●● ●●●●● ● 0.0 ● ●● ●●● ● ● ● 0.0 ●● ●●● ● ● ●● ● ●●●● ●●● ● ●● ● ● ●●●● ●● ● ● ● ●●● ●● ●● ● ● ● ● ● ● ●●● ● ● ●● ● ●● ●●●●● ●● ● ● ●● ● ●●●●●● ●● ●● ● ● ●●●●● ●●● ● ●● ●●●●● ● ● ● ●●● ●● ●● ● ●●●●● −0.5 ●●●● ●●●● ●● ● ● ●● ● −0.5 ●●●●●● ● ● ● ● ●●● ● ● ●● ●●●●●● ●● ● ● ● ● ●● ● ●●● ● ●●●● ● ● ● ● ● ●● ● ●●●●●●●●●● ● ●●●●●●● ●●● ● ● ● ●● ● ● ● ● ●● ● ● ● ●●●●● ● ● ● ● ● ● ●● −1.0 ●● ● ● −1.0 ● ● ● ● ● −1.5 ● −1.5 ●

5 10 15 20 25 5 10 15 20 25 0 100 200 300 400 0 100 200 300 400 ptpnt_token nnd

Phonotactics/wordlikeness trend weaker overall than density/wordlikeness trend. Background: phonotactic gaps The study Results Discussion Summary Discussion

There appears to be intra-group variation (cf. Frisch & Zawaydeh) There appears to be intra-group variation in a non-comparative wordlikeness task (cf. Coetzee) There appears to be degrees of ill-formedness among zero-freqency items (cf. Boersma & Hayes) Background: phonotactic gaps The study Results Discussion Summary Towards an explanation

Why did we observe intra-group variation? Why is neighborhood density apparently such a good regressor, and phonotactic probability such a poor one? Background: phonotactic gaps The study Results Discussion Summary Towards an explanation

Maybe still a function of task type? Inclusion of lexical fillers may cause lexical density to emerge as the dominant cue of wordlikness (Bailey & Hahn 2001; Shademan 2006) However, Frisch et. al. 2000 found that, in the presence of grammatical probability and lexical similarity effects, grammatical probability was a better predictor of well-formedness judgments... and Shademan (2006) claims phonotactic log probability was an invariantly good predictor; lexical density just emerged as a better predictor when real-word fillers were introduced. Not so for Cantonese - why? Background: phonotactic gaps The study Results Discussion Summary Towards an explanation

English, which permits complex onsets and codas, allows for a far greater number of logically possible monosyllables (n > 158, 000) than does Cantonese (n = 5, 130 [19 initials × 45 finals × 6 tones]) English also makes use of a much smaller proportion of these possibilities (10,000 monosyllables ≈ 6%) than does Cantonese (1,900 monosyllables, ≈ 36%) Background: phonotactic gaps The study Results Discussion Summary Towards an explanation

This may explain why neighborhood density correlates with wordlikeness so well in Cantonese: most nonwords are similar in at least one segment/toneme to existing lexemes The fact that most nonwords have lexical neighbors may underlie the emergence of lexical neighborhood density as a predictor of wordlikeness. Background: phonotactic gaps The study Results Discussion Summary Towards an explanation

Why wasn’t phonotactic probability a good regressor? Most sequences violate phonotactics, so expected probabilities will be low Might improve with a better model? Background: phonotactic gaps The study Results Discussion Summary Future directions

Wilson & Hayes (2006) adjust O/E values to allow the size of E to reflect the acceptability of non-occurring segment pairs Pater (2007) outlines a mechanisms making use of the GLA in a Harmonic Grammar framework that can assign gradient acceptability values to zero-frequency sequences Background: phonotactic gaps The study Results Discussion Summary Future directions

Explore more sophisticated neighborhood models: Bybee suggests high type frequency neighbors might actually NOT count as part the neighborhood for purposes of the lexical density calculation (dissociation effect) Explore more sophisticated phonotactic models (backoff, smoothing) Replicate study with stimulus set balanced for lexical statistics Background: phonotactic gaps The study Results Discussion Summary Summary

Gradient acceptability effects emerge even among nonwords which roundly violate phonotactic constraints. In Cantonese, acceptability seems to be correlated most strongly with lexical neighborhood density. Correlation of lexical statistics to wordlikeness influenced not only by task, but also by the phonotactic and lexical properties of a given language Appendix For Further Reading I

Albright, A. 2006. Gradient phonotactic effects: lexical? grammatical? both? neither? Talk presented at the 80th annual meeting of Linguistic Society of America, 5-8 January 2006.

Albright, A. and B. Hayes. 2003. Rules vs. analogy in English past tenses: A computational/experimental study. Cognition 90:119-161.

Anttila, A. 1997a. Deriving variation from grammar. In F. Hinskens, R. van Hout and L. Wetzels (eds.), Variation, Change and Phonological Theory, 35-68. Amsterdam: John Benjamins.

Anttila, A. 1997b. Variation in Finnish and . Stanford Ph.D dissertation.

Antttila, A. and C. Andrus. 2006. T-orders. Stanford University ms.

Bailey, T. and U. Hahn. 2001. Determinants of wordlikeness: phonotactics or lexical neighborhoods? Journal of Memory and Language 44:568-591.

Boersma, P. 1997. How we learning variation, optionality, and probability. Proc. Inst. of Phonetic Sciences of the University of Amsterdam 21:43-58.

Boersma, P. and B. Hayes. 2001. Empirical tests of the Gradual Learning Algorithm. Linguistic Inquiry 32: 45–86.

Coetzee, A. To appear. Grammar is both categorical and gradient. In S. Parker (ed.), Phonological argumentation. London: Equinox.

Coetzee, A and J. Pater. 2006. Lexically-ranked OCP-Place constraints in Muna. University of Michican and University of Massachusetts-Amherts ms.

Coleman, J. and J. Pierrehumbert. 1997. Stochastic phonological and acceptability. In Computation Phonology: Third Meeting of the ACL Special Interest Group in Computational Phonology, pages 49-56, Somerset, NJ: Association of Computational Linguistics.

Fischer-Jørgensen, E. 1952. On the definition of phoneme categories on a distributional basis. Acta Linguistica 7:8-39.

Frisch, S. A., N. R. Large, and D. B. Pisoni. 2000. Perception of wordlikeness: effects of segment probability and length on the processing of nonwords. Journal of Memory and Language 42:481-496.

Frisch, S. A., J. B. Pierrhumbert, and M. B. Broe. 2004. Similarity avoidance and the OCP. Journal of Memory and Language 22:179-228.

Frisch. S. A. and B. Zawaydeh. 2001. The psychological reality of OCP-Place in Arabic. Language 77:91-106. Appendix For Further Reading II

Greenberg, J. H. and Jenkins, J. J. 1964. Studies in the psychological correlates of the sound system of American English. Word 20:157-177.

Halle, M. 1962. Phonology in generative grammar. Word 18:54-72.

Hay, J., J. Pierrehumbert, and M. Beckman. 2004. Speech perception, well-formedness and the statistics of the lexicon. In J. Local, R. Ogden, and R. Temple (eds.), Phonetic Interpretation: Papers in Laboratory Phonology VI. Cambridge: Cambridge University Press.

Hayes, B. 2000. Gradient well-formedness in Optimality Theory. In J. Dekkers, F. van der Leeuw, and J. van de Weijer (eds.), Optimality Theory: Phonology, Syntax, and Acquisition, 88-120. Oxford:Oxford University Press.

Iverson, G. and J. Salmons. 2005. Filling the Gap: English tense vowel plus final /š/. Journal of English Linguistics 33:207-221.

Kwan, T.-W., W.-S. Tang, T.-M. Chiu, L.-Y. Wong, D. Wong, and L. Zhong. 2003.Chinese character database with word-formations phonologically disambiguated according to the Cantonese dialect. Research Centre for Humanities Computing, City University of Hong Kong.

Leung, M.-T. and S.-P. Law. 2001. HKCAC: the Hong Kong Cantonese adult language corpus. International Journal of Corpus Linguistics 6:305-326.

Moreton, E. 2002. Structural constraints in the perception of English stop-sonorant clusters. Cognition 84:55-71.

Myers, J. 2002. An analogical approach to the Mandarin syllabary. J. Chinese Phonology 11:163-190.

Myers, J. and J. Tsay. 2004. Exploring performance-based predictors of phonological acceptability judgements in Mandarin. Poster presented at LabPhon 9, UIUC.

Myers, J. and J. Tsay. 2005. The processing of phonological acceptability judgements. In Proceedings of Symposium on 90-92 NSC Projects, Taipei.

Ohala, J. and M. Ohala. 1986. Testing hypotheses regarding the psychological manifestation of structure constraints. In J. J. Ohala and J. J. Jager, editors, Experimental Phonology, pages 239-252. Academic Press, Florida.

Pater, J. 2007. Gradual learning and gradient phonotactics. Handout, Workshop: Acquisition of Phonology, Radboud University, Nijmegen, March 13, 2007. Appendix For Further Reading III

Shademan, S. 2006. Is phonotactic knowledge grammatical knowledge? In D. Baumer, D. Montero, and M. Scanlon (eds.), Proc. WFCCL 25. Somerville, MA: Cascadilla Proceedings Project.

Vitevitch, M. and Luce, P.A. 1998. When words compete: Levels of processing in perception of spoken words. Psychological Science 9:325-329.

Vitevitch, M.S., P. A. Luce, and D. Kemmerer. 1997. Phonotactics and syllable stress: implications for the processing of spoken nonsense words. Language and Speech 40(1):47-62.

Wilson, C. and B. Hayes. To appear. A Maximum Entropy model of phonotactics and phonotactic learning. To appear in Linguistic Inquiry.

Zuraw, K. 2000. Patterned exceptions in phonology. UCLA Ph.D dissertation.

Zuraw, K. 2002. Aggressive reduplication. Phonology 19:395-439.