the dynamic interaction between frequency and internal ... · the dynamic interaction between...
TRANSCRIPT
The Dynamic Interaction between Frequency and Internal Constraints: A Case Study of
(ING)
Jon Forrest
SVALP 2016 Blacksburg, VA
Introduction
• The role of lexical frequency in linguistic variation and change
• Hybrid models of phonology (Pierrehumbert 2016)
• The importance of context of occurrence for word-specific experiences
Frequency and Lenition Bias
• Physiologically-motivated sound changes affect more frequent words before less frequent words, and vice versa (Phillips 1984)
• Lexical frequency has an effect on the rate of lenition processes (Bybee 2002; Dinkin 2008; Pierrehumbert 2001, 2002)
• Further study of t/d deletion finds no frequency effect overall, only frequency in certain environments (Walker 2012)
Lenition, Frequency, and (ING)
• (ING) found not to have frequency effects (Abramowicz 2007) – Implementation of frequency
• –ing to –in as (partially) a lenition process • Vowel cues? (Forrest 2014; Yuan and
Lieberman 2011) • /ŋ/ to /n/ as a leniting process (Abramowicz
2007)
Context of Occurrence
• Word-specific effects of frequent environment of occurrence (Guy, Hay, and Walker 2009)
Context of Occurrence
• Word-specific effects of frequent environment of occurrence (Guy, Hay, and Walker 2009)
[lɛft]
[lɛft]
[lɛft]
[lɛft]
[lɛf] [lɛf]
[lɛf]
[lɛf] [lɛf]
[lɛf] [lɛf]
[lɛf]
[lɛf]
[lɛft]
[lɛf] [lɛf]
[lɛf]
[lɛf]
[lɛf]
[lɛf]
[lɛf]
[lɛf]
[lɛf]
[lɛft]
[lɛft]
[lɛft]
Allexemplarsof“le.”
Exemplarsof“le.”inenvironmentsdisfavoringdele7on
Exemplarsof“le.”inenvironmentsfavoringdele7on
Research Questions
• Does frequency play a role in the realization of (ING) (Abramowicz 2007)?
• Do collocations explain either frequency effects or phonological/morphological effects (Bybee 2006; Seyfarth 2014; Walker 2012)?
• Does frequency interact with year of birth, reflecting the change in rates of –in and –ing over time?
• Does a word’s frequent occurrence in environments that (dis)favor –in have an effect independent of frequency (Guy, Hay, and Walker 2009)?
Data
• 121 European-American speakers from Raleigh, NC – Roughly balanced for Sex and Occupation – Dialect contact in Raleigh – Conversational interview from the Raleigh Corpus
• 12052 tokens of (ING) coded overall
Coding Methods
• Impressionistically coded as –in or –ing – Coding following Forrest (2015)
• Assimilated tokens (“gonna”, “tryna”), pronouns, and prepositions excluded
• Following Place of Articulation (coronal, velar, coda, other)
• Lexical Category (verbal, gerund, noun, adjective)
Coding of Frequency Variables
• Word frequency measured two ways (both log10 transformed): – SUBTLEX (Brysbaert and New 2009) – Corpus frequency
• Two other scaled frequency variables for each word measuring occurrence in specific environments – Percent occurrence in high-IN grammatical environments
(verbal) – Percent occurrence in low-IN phonological environments (pre-
velar and pre-pausal) • Collinearity issues
– .01 to .04 for grammatical measures – .02 to .16 for phonological measures.
Statistical Methods
• Logistic mixed-effects models with –in vs. –ing as outcome variable – Random intercept for word – Random slope for following place of articulation
by speaker • Addition of frequency variables stepwise for
model fitting
Model Comparison Statistics
1. Base Model (Internal Constraints and Social Factors) 2. Frequency 3. Frequency/Age Interaction 4. Grammatical Environment Frequency 5. Phonological Environment Frequency 6. Frequency/Grammatical Environment Interaction 7. Frequency/Phonological Environment Interaction 8. Frequency/Grammatical Environment/Phonological Environment Interaction
• Model selection determined by AIC comparison
• Models:
Full Model Results
Gerund 0.917**(0.328)Verbal 1.333***(0.333)Noun 0.485(0.365)FollowingPause -0.819***(0.121)FollowingCoronal 0.137(0.077)FollowingVelar -0.603***(0.181)Male 0.758*(0.369)DOB(Scaled) -1.192***(0.236)College -3.076***(0.712)Graduate -3.726***(0.813)NoCollege -1.714*(0.797)SUBTLEXFrequency(Logged) 0.581***(0.070)PercentHigh-INGramma7cal(Scaled) 0.126(0.144)PercentHigh-INPhonological(Scaled) -0.107*(0.049)Age/FrequencyInterac7on 0.114**(0.036)
Frequency/Gramma7calEnvironmentInterac7on 0.114*(0.057)
(Intercept) -1.394*(0.798)
ObservaNons 12,052LogLikelihood -4,627.739AkaikeInf.Crit. 9,311.478BayesianInf.Crit. 9,518.594
Note: *p<0.05;**p<0.01;***p<0.001
Variable Effects of Collocations
• Four of top-10 most frequent words in corpus – Three most common
following articulation environments
• Frequency of the “verb-ing + to” collocation
• Informaticity (Seyfarth 2014)
Frequency Matters Less for Older Speakers
• Frequency: – Mean = 3.56 – SD = 1.01
• Age: – Mean = 1962 – SD = 17.37 yrs
Frequency Matters More for Words in Verbal Environments
• Percentage that a word occurs in a verbal context – Mean = 50.77% – Standard Dev. = 30.56%
Frequency Matters More for Words in Verbal Environments
• Snowballing effect of exemplars in verbal environments
Discussion
• Frequency plays a role in the realization of (ING)
• Collocations do not primarily drive any of the frequency or internal constraint effects
• Support for exemplar-based accounts of (ING) – Two things for age effects (lots of –in exemplars
around; ing becomes base form)
Full Model Results
Gerund 0.917**(0.328)Verbal 1.333***(0.333)Noun 0.485(0.365)FollowingPause -0.819***(0.121)FollowingCoronal 0.137(0.077)FollowingVelar -0.603***(0.181)Male 0.758*(0.369)DOB(Scaled) -1.192***(0.236)College -3.076***(0.712)Graduate -3.726***(0.813)NoCollege -1.714*(0.797)SUBTLEXFrequency(Logged) 0.581***(0.070)PercentHigh-INGrammaNcal(Scaled) 0.126(0.144)PercentHigh-INPhonological(Scaled) -0.107*(0.049)Age/FrequencyInteracNon 0.114**(0.036)
Frequency/GrammaNcalEnvironmentInteracNon 0.114*(0.057)
(Intercept) -1.394*(0.798)
ObservaNons 12,052LogLikelihood -4,627.739AkaikeInf.Crit. 9,311.478BayesianInf.Crit. 9,518.594
Note: *p<0.05;**p<0.01;***p<0.001
Following Place of Articulation by Speaker
• Individually adjusted coefficients for following place of articulation – Positive coefficients
favor –in, negative favor –ing
Conclusion
• Frequency matters, and it matters in a complex way
• Points towards a conception of variation as a combination of grammar constraint rules and word-specific knowledge – Underlying forms vs. word-specific phonetics?
• Look at (ING) in terms of connected speech processes and articulatory movement (Temple 2014)
References
• Abramowicz, Ł. (2007). Sociolinguistics meets exemplar theory: Frequency and recency effects in (ing). University of Pennsylvania Working Papers in Linguistics, 13(2), 3.
• Bybee, J. (2002). Word frequency and context of use in the lexical diffusion of phonetically conditioned sound change. Language variation and change, 14(03), 261-290.
• Bybee, J. (2006). From usage to grammar: The mind's response to repetition. Language, 711-733. • Dinkin, A. J. (2008). The real effect of word frequency on phonetic variation. University of
Pennsylvania Working Papers in Linguistics, 14(1), 8. • Hazen, K. (2011). Flying high above the social radar: Coronal stop deletion in modern Appalachia.
Language Variation and Change, 23(01), 105-137. • Labov, W. (2011). Principles of Linguistic Change, Cognitive and Cultural Factors (Vol. 3). John
Wiley & Sons. • Pierrehumbert, J. B. (2001). lenition and contrast. Frequency and the emergence of linguistic
structure, 45, 137. • Pierrehumbert, J. (2002). Word-specific phonetics. Laboratory phonology, 7, 101-139. • Tagliamonte, S., & Temple, R. (2005). New perspectives on an ol'variable:(t, d) in British English.
Language Variation and Change, 17(03), 281-302. • Walker, J. A. (2012). Form, function, and frequency in phonological variation. Language Variation
and Change, 24(03), 397-415.
Full Model Comparison
ModelFitUsingCorpusFrequencyModels K AICc Delta_AICc LogLikelihood
Frequency/Phon.Env.InteracNon 29 9320.04 0 -4630.95
Frequency/Gram.Env.InteracNon 28 9322.78 2.75 -4633.32
Freq./Gram.Env./Phon.Env.InteracNon 31 9323.78 3.74 -4630.81
PhonologicalEnvironmentFrequency 27 9334.04 14 -4639.95
GrammaNcalEnvironmentFrequency 26 9337.24 17.2 -4642.56
Age/FrequencyInteracNon 25 9392.51 72.47 -4671.2
Frequency 24 9399.88 79.84 -4675.89
BaseModel 23 9441.54 121.5 -4697.72ModelFitUsingSUBTLEXFrequency
Models K AICc Delta_AICc LogLikelihood
Frequency/Phon.Env.InteracNon 29 9311.85 0 -4626.86
Frequency/Gram.Env.InteracNon 28 9311.98 0.13 -4627.92
PhonologicalEnvironmentFrequency 27 9313.78 1.92 -4629.83
Freq./Gram.Env./Phon.Env.InteracNon 31 9314.07 2.21 -4625.95
GrammaNcalEnvironmentFrequency 26 9315.92 4.06 -4631.9
Age/FrequencyInteracNon 25 9354.41 42.55 -4652.15
Frequency 24 9361.83 49.97 -4656.86
BaseModel 23 9441.54 129.68 -4697.72