Effects of phonological contrast on phonetic variation in Hindi and - PowerPoint PPT Presentation

Effects of phonological contrast on phonetic variation in Hindi and English stops Ivy Hauser University of Massachusetts Amherst blogs.umass.edu/ihauser Workshop on Phonological Variation and its Interfaces November 22, 2018 0

Introduction Big picture: What is the typology of variation? How does variation itself vary across languages? What are the limits? 1

Introduction Big picture: What is the typology of variation? How does variation itself vary across languages? What are the limits? ◮ What varies 1

Introduction Big picture: What is the typology of variation? How does variation itself vary across languages? What are the limits? ◮ What varies ◮ Amount of variation 1

Introduction Big picture: What is the typology of variation? How does variation itself vary across languages? What are the limits? ◮ What varies ◮ Amount of variation ◮ Sources of structure in variation 1

Introduction → Question: How do different systems of phonological contrast affect patterns of phonetic variation? 2

Introduction → Question: How do different systems of phonological contrast affect patterns of phonetic variation? ◮ Hypothesis: “Systems with more phonological contrasts should exhibit less within-category variation than systems with fewer contrasts” (Lindblom, 1986). 2

Introduction → Question: How do different systems of phonological contrast affect patterns of phonetic variation? ◮ Hypothesis: “Systems with more phonological contrasts should exhibit less within-category variation than systems with fewer contrasts” (Lindblom, 1986). ◮ Hypothesis: Variation predicted by number of phonemes in an inventory. 2

Introduction → Question: How do different systems of phonological contrast affect patterns of phonetic variation? ◮ Hypothesis: “Systems with more phonological contrasts should exhibit less within-category variation than systems with fewer contrasts” (Lindblom, 1986). ◮ Hypothesis: Variation predicted by number of phonemes in an inventory. ◮ But phonological contrasts are not unidimensional in phonetic space 2

Introduction → Question: How do different systems of phonological contrast affect patterns of phonetic variation? ◮ Hypothesis: “Systems with more phonological contrasts should exhibit less within-category variation than systems with fewer contrasts” (Lindblom, 1986). ◮ Hypothesis: Variation predicted by number of phonemes in an inventory. ◮ But phonological contrasts are not unidimensional in phonetic space ◮ Issues with quantifying within-category variation: What are the relevant phonetic dimensions? What counts as a system/inventory? 2

Introduction Proposal: We expect to see less variation in languages that realize more phonological contrasts only along the particular phonetic dimensions which realize additional contrasts. 3

Test case: Hindi and English stops Hindi has four contrasting stops at each place of articulation; English has two (Kagaya and Hirose, 1975; Ohala, 1994; Quirk et al., 1972). Labial Coronal Retroflex Velar / p / p h / / b / / b h / / t / / t h / / d / / d h / / ú / / ú h / / ã / / ã h / / k / / k h / / g / / g h / Hindi English / p / / b / / t / / d / / k / / g / 4

Test case: Hindi and English stops ◮ If variation is predicted by number of phonemes in an inventory : 5

Test case: Hindi and English stops ◮ If variation is predicted by number of phonemes in an inventory : We might expect Hindi speakers to constrain variation on all dimensions, including lag time, prevoicing, f0, etc. 5

Test case: Hindi and English stops ◮ If variation is predicted by number of phonemes in an inventory : We might expect Hindi speakers to constrain variation on all dimensions, including lag time, prevoicing, f0, etc. → Hindi / k h / should vary less than English / k h / in voiceless lag time. 5

Test case: Hindi and English stops ◮ If variation is predicted by number of phonemes in an inventory : We might expect Hindi speakers to constrain variation on all dimensions, including lag time, prevoicing, f0, etc. → Hindi / k h / should vary less than English / k h / in voiceless lag time. ◮ If variation is predicted by more contrasts along a single dimension : 5

Test case: Hindi and English stops ◮ If variation is predicted by number of phonemes in an inventory : We might expect Hindi speakers to constrain variation on all dimensions, including lag time, prevoicing, f0, etc. → Hindi / k h / should vary less than English / k h / in voiceless lag time. ◮ If variation is predicted by more contrasts along a single dimension : Hindi speakers will only exhibit less variation along phonetic dimensions which distinguish additional contrasts relative to English. 5

Test case: Hindi and English stops ◮ If variation is predicted by number of phonemes in an inventory : We might expect Hindi speakers to constrain variation on all dimensions, including lag time, prevoicing, f0, etc. → Hindi / k h / should vary less than English / k h / in voiceless lag time. ◮ If variation is predicted by more contrasts along a single dimension : Hindi speakers will only exhibit less variation along phonetic dimensions which distinguish additional contrasts relative to English. → Hindi / k / and / g / should vary less than English / g / in voicing. 5

Phonetic dimensions in Hindi and English stops Hindi lag time / k h / / k / closure voicing / g h / / g / English lag time / g / / k / closure voicing 6

Phonetic dimensions in Hindi and English stops Hindi lag time / k h / / k / closure voicing / g h / / g / English lag time / g / / k / closure voicing ◮ No difference expected in voiceless lag time (positive VOT). ◮ Difference expected in prevoicing because Hindi has additional voicing contrasts relative to English. 6

The experiment ◮ Stimuli: C 1 VC 2 words in carrier phrases. ◮ C 1 was a stop and V was one of [i a u]. ◮ Carrier phrases: Say X again; Dobara X doharao. 7

The experiment ◮ Stimuli: C 1 VC 2 words in carrier phrases. ◮ C 1 was a stop and V was one of [i a u]. ◮ Carrier phrases: Say X again; Dobara X doharao. ◮ 4 repetitions of each distinct stimulus per speaker. 7

Analysis: Lag time All data were forced aligned with the Montreal Forced Aligner (McAuliffe et al., 2017) and analyzed in Praat (Boersma and Weenink, 2001) 8

Analysis: Lag time All data were forced aligned with the Montreal Forced Aligner (McAuliffe et al., 2017) and analyzed in Praat (Boersma and Weenink, 2001) Voiceless stops: voiceless lag time measured from the burst to the onset of voicing with AutoVOT (Keshet et al., 2014) and manual correction. 8

Analysis: Lag time All data were forced aligned with the Montreal Forced Aligner (McAuliffe et al., 2017) and analyzed in Praat (Boersma and Weenink, 2001) Voiceless stops: voiceless lag time measured from the burst to the onset of voicing with AutoVOT (Keshet et al., 2014) and manual correction. 8

Expected results: Lag time Expected results: inventory prediction 0.04 0.03 density density 0.02 0.01 0.00 − 50 − 25 0 25 50 centered lag time (ms) Language English Hindi All graphing done in R (R Core Team, 2013; Wickham, 2009). 9

Results: Lag time Voiceless aspirated coronal VOT values 0.03 Language 0.02 density English 0.01 Hindi 0.00 − 50 − 25 0 25 50 centered VOT Voiceless aspirated velar VOT values 0.03 Language density 0.02 English 0.01 Hindi 0.00 − 50 − 25 0 25 50 centered VOT 10

Discussion: Lag time Why no difference in variation between languages? (Levene’s Test for homogeneity of variance not significant.) 11

Discussion: Lag time Why no difference in variation between languages? (Levene’s Test for homogeneity of variance not significant.) ◮ Voiceless lag time realizes one contrast in both languages, no difference expected. 11

Discussion: Lag time Why no difference in variation between languages? (Levene’s Test for homogeneity of variance not significant.) ◮ Voiceless lag time realizes one contrast in both languages, no difference expected. ◮ Additional evidence for understanding prevoicing and lag as separate dimensions (Mikuteit & Reetz, 2007) 11

Discussion: Lag time voice onset time / g h / / k h / / g / / k / 12

Discussion: Lag time voice onset time / g h / / k h / / g / / k / Hindi lag time / k h / / k / closure voicing / g h / / g / 12

Analysis: Voicing Closure duration was measured from the end of the vowel to the stop burst. 13

Analysis English voiced stop - prevoiced postvocalically 14

Analysis English voiced stop - voiceless postvocalically 15

Results: Prevoicing 16

Results: Prevoicing Prevoicing categories (Beckman et al., 2013) none = less than 25% voicing during closure part = 25-90% percent voiced full = 90%+ voiced 17

Results: Prevoicing All Hindi voiced stops All English voiced stops 100 100 percent of all voiced stops percent of all voiced stops 75 75 50 50 25 25 0 0 none part full none part full degree of prevoicing degree of prevoicing 18

Examining the English variation There is more variation in English prevoicing, in accordance with the revised hypothesis. 19