Variation and mental representation Gregory R. Guy New York - - PowerPoint PPT Presentation

Variation and mental representation

Gregory R. Guy New York University

Words in the mind

• Abstract representations, composed of strings
• f phonemes.
• Morphological structure either stored in

representation or generated by derivation and /ænd/ band /bænd/ banned /bæn#d/

Words in speech

• Surface forms reflect allophony, phrasal

phonology, etc.

• Show variable realizations reflecting variable

phonological processes.

– e.g., final coronal stop deletion and ~ an’ band ~ ban’ banned ~ ban’

What’s in the mental lexicon?

Variable processes are conditioned by:

• Lexical frequency
• Morphological structure
• Lexical exceptions
• Priming

• I. Frequency effects on variation
• coronal stop deletion in English
• -ing/-in alternation in English
• coda –s lenition in Caribbean Spanish

Coronal stop deletion increases with lexical frequency

Guy, Hay & Walker 2008. p=.0005

• ing > -in: More frequent words show more

alveolar [ɪn] pronunciation

Source: Laturnus, de Vilchez, Chaves & Guy, 2016.

0.1 0.2 0.3 0.4 0.5 0.6 0.7 <8 9-10 11 12+ log10 freq

p [ɪn] Figure 1. [ɪn] by lexical frequency

Spanish -s lenition (Erker 2008) No significant frequency effect

• Correlation between frequency and:

–Spectral center of gravity

N=318 r = -.02 p = .74, n.s. – Duration N=453 r = .07 p = .136, n.s.

Bybee on lexical frequency effects (Bybee 2000)

• “The more a word is used, the more it

is exposed to the reductive effect of articulatory automation…”

• “Sound change affects stored

representations incrementally each time a word is used…”

• 2. Morphological constraints on

variation

• coda –s deletion in Brazilian Portuguese
• -ing/-in alternation in English
• coronal stop deletion in English

Morphological constraint on coda /s/ deletion in Brazilian Portuguese (Guy 1981) Coda /s/ deletion in unstressed final syllables N % deleted Monomorphemes 1392 53%

e.g. menos

Nominal plurals* 5247 5%

e.g. eles, todos

*NB: first position in NP only

Morphological constraint on –in/-ing alternation in English

Laturnus et al. 2016

12

N % [n] weight Lexical class verbs 489 55% .53 something, nothing 81 67% .51 nouns 79 38% .32

Morphological constraint on coronal stop deletion in English

N % deleted

Monomorphemes 658 38.1% e.g., guest Irregular Past 56 33.9 e.g., lost Regular Past 181 16.0 e.g., guessed

(Guy 1992 corpus)

Morphological structure interacts with lexical frequency: coronal stop deletion (Myers & Guy 1997) Monomorphemic words Significant frequency effect N Deletions % Del Low frequency 151 28 18.5 High frequency 573 194 33.9 p < .01

Morphology interacts with frequency: coronal stop deletion (Myers & Guy 1997) Regular Past Tense Verbs No significant frequency effect

N Deletions % Del Low frequency 96 7 7.3 High frequency 220 18 8.2 chi-sq (1df) = .073, p > .70

Morphology interacts with frequency: coronal stop deletion (Myers & Guy 1997)

Frequency effect by morphology

5 10 15 20 25 30 35 40 low freq high freq Lexical frequency % -t,d deletion monomorphemes regular past

• t,d morphology: Fruehwald

(probabilities of /t,d/ retention)

Variation and morphology: conclusions

• Variable processes are sensitive to morphological

structure; they can be conditioned by derived environments.

• Hence variable phonological processes ‘see’

mental representations that incorporate morphological information, in addition to the phonological content.

• Morphological constraints on phonology interact

with lexical frequency: derived forms are less affected by frequency

Morphology and lexical frequency

• Bybee’s model of lenition fed by frequency is

phonetically motivated

• This should be independent of and orthogonal to

morphological constraints

• But other models make other predictions. Pinker

and others argue that regular derived forms are generated by rule; only roots and irregular forms are stored

• Hence, regular derived forms lack independent

mental representations to accumulate frequency

• r collocational information

• 3. Lexical exceptions
• Coronal stop deletion in English:

exceptional and

• Coda /s/ deletion in Brazilian Portuguese:

exceptional 1pl morpheme –mos

• Coda /s/ deletion in Salvadoran Spanish:

exceptional entonces, digamos

• /ay/ monophthongization in Southern US

English: exceptional I, my

Lexical exceptions in variation

Many variable processes are known to exhibit unusually high rates of occurrence in particular lexical items.

• e.g., coronal stop deletion in English is

exceptionally frequent in ‘and’ (Exceptional because deletion occurs significantly more often in and than in phonologically comparable words like sand, band, hand, etc.)

Exceptional and in the NZE corpus:

N Deletion rate and 597 80%

• ther words

3348 29%

Lexical exceptions are not just high frequency words

• The second-highest frequency word in the

ONZE corpus was just; it showed significant following context effects and did not behave like and

• Spanish menos is higher in frequency than

entonces and digamos, but does not behave exceptionally

Following context effect in the NZE corpus

Other words and Following N % del N % del Context: __C 1339 58.3 315 87.9 __V 1477 10.4 182 75.3 Range: 47.9% > 12.6%

(18 speakers from the ONZE corpus at U Canterbury)

Contextual effects on Brazilian Portuguese –s deletion

• verbs ending in 1plural –mos suffix omit final

/s/ at exceptionally high rate.

• Coda /s/ deletion mainly occurs in

preconsonantal position, and is strongly constrained by place manner and voicing of the following C

• Do these constraints affect exceptional –mos

words just like other words?

Lexical exceptions in Brazilian Portuguese: coda -s deletion [factor weights]

Features of following C Non-exceptions Lexical exceptions (-mos forms) Voice/Manner: sonorant .69 .49 voiced obstruent .44 .58 voiceless obstruent .36 .44 Range .33 > .14 Place: labial .32 .58 coronal .61 .53 velar .44 .39 Range .29 > .19 N: 5880 1225 Goodness of fit (log likelihood) -704.8

• 791.5

Following context effect appears significantly weaker in exceptional -mos

• Range of probabilities is smaller for both the

place effect and the manner/voicing effect

• The goodness of fit measure is significantly

worse for the exceptional forms, suggesting that they aren’t as well explained by the contextual conditions

Contextual effects on Salvadoran Spanish –s deletion

• El Salvador has variable final –s deletion, like
• ther Caribbean dialects
• Hoffman 2004 finds strong constraint effects
• n deletion; more deletion in stressed

syllables, more deletion before consonants, than before vowels

• Three discourse markers show exceptionally

high rates of deletion: entonces, digamos, pues

• s deletion in Salvadoran Spanish (Hoffman 2004)

Non-exceptional words Lexical exceptions Following context: (entonces, digamos, pues) sonorant .60 .63 voiced obstruent .75 .55 voiceless obstruent .33 .38 vowel .36 .38 pause .44 .56 Range .42 > .25 Syllable Stress: stressed .38 .42 unstressed .62 .58 Range .24 > .16

Another variable: monophthongal /ay/ in Southern American English (SoAmEng)

• The English diphthong /ay/ is variably

monophthongized to /a/ in Southern American English

• More monophthongs are found in pre-voiced

contexts (ride vs. right), in phonetically shorter syllables, and among lower status speakers

• I and my are lexical exceptions, with very high

rates of monophthongization, even before voiceless consonants (cf: ‘my time’)

/ay/ monophthongization in SoAmEng: following context effect (Woods 2008)

Other words I, my

% monophthong 34% 53%

• Fol. Context:

__C[+vce] .76 (.51) __V or G .41 (.49) __C[-vce] .17 (.48) Range: .59 > .03 (n.s.)

/ay/ monophthongization in Southern AmEng: duration effect

Other words I, my

Duration: shorter .89 .68 longer .49 .45 Range: .40 > .23

(Data from Woods 2008)

Contextual effects are much weaker on exceptional I, my in SoAmEng

• Following context effect is not significant for

I, my

• Duration effect is much weaker
• Monophthongization occurs much more often

in these two words, and is relatively insensitive to context.

Conclusion: Exceptional words have alternate mental representations

• Lexical exceptions to variable processes reveal

allomorphy in the mental representation

• They have additional lexical entries that

incorporate the output of the variable process.

– and ~ an’ or ‘n’

• I, my ~ [a], [ma]

– -mos ~ -mo – -entonces, digamos ~ entonce, digamo

• When the ‘reduced’ allomorph is selected,

context has no effect, and apparent frequency of

• ccurrence of the variable phonological process is

increased.

• 4. Priming
• Specific variants tend to occur in clusters: e.g.,

in BP –s deletion, one deleted form is likely to favor subsequent deleted forms, an undeleted form favors subsequent undeleted forms.

• in/-ing alternation: priming
• Priming is a significant predictor

– Alveolars favor subsequent alveolars – Velars favor subsequent velars – But less than 40% of tokens occur in priming contexts

36

Priming context N % [n] weight

• n

129 79% .64

• ŋ

119 21% .27 ∅ 401 54% .53

Priming: same or different?

prime form [n] prime form [ŋ] no prime

prime same as target V to V, N to N

.70 .25

prime different from target: V to N, N to V

.52 .25

no prime

.53

Goldvarb factor weights for /in/

Same or similar?

• The strongest priming effect is like-to-like (verb

to verb, noun to noun): .72 vs .25

• Cross-category priming (V to N, N to V) is also

strong: .52 vs .25

• No significant improvement in model by

distinguishing whether priming category is same

• r different from target.

Summary

• Mental lexical representations

– accumulate frequency information – variably incorporate morphological information – permit allomorphy – permit partial, overlapping identities

• i.e., they have quantified, continuous properties

The fuzzy lexicon

• Lexical entries are associated with probability

functions

• Stable components (segments or features) of the

morpheme are assigned high probabilities of realization

• Components that are variably realized have

lowered probabilities.

• Exceptional allomorphs are entries in which the

probability of some component has reached zero.

Frequency

• Probability functions are updated with

experience

• This incorporates information about lexical

frequency without requiring storage of exemplars

• Derived forms without lexical entries do not

accumulate frequency information.

• Frequency is an ‘elsewhere’ constraint

Probabilistic representations

banned: ban + ed band: ban and: an ~ an

Thank you!

Gracias Obrigado Merci Arigato

Dank je wel!