Influence of level tones on vowel duration in Yoruba

class: center, middle, inverse, title-slide

# Influence of level tones on vowel duration in Yoruba
### Dine Mamadou
### Rutgers University | ACAL 49 | MSU 3/24/18

---

# Claim
## The negative correlation between f0 and vowel duration claimed to be a universal property of tone languages is a parameter instead.  
--

# Asumptions

## a. There's a correlation between tone and vowel duration

## b. Yoruba High, Mid and Low tones are all level tones, for fairness of comparison

---
# Cross-linguistic short H and long L
### In a variety of studies in world languages, H and L level tones have been found to have shortening and lengthening effects on vowel duration, respectively:
--

###- Acoustic studies: in Standard Thai (Abramson 1962, Gandour 1977), in Mandarin (Dreher & Lee 1966, Chuang 1972), in Chatino (Upson 1968)

###- Auditory studies: in Cantonese (Kong 1987), in Tenango Otomi (Blight & Pike 1974), in Kutchin (Scollon 1975), in Yuman langugages (Langdon 1976)

###- Articulatory studies: Erikson (1976) found that pitch drops from M to L in Thai was done actively, which takes more time than a simple muscle relaxation (unactive) process
---

#Gandour's generalizations

###The three types of evidence presented above and the diachronic evidence of long vowels developing from low tones (in Thai) seem to beg for a generalization:

###(a)	'Vowels on low tones are longer than those on high tones,
--

###(b)	Vowels on rising tones are longer than those on falling tones and,
###(c)	Vowel duration is inversely related to the approximate average fundamental frequency' [Gandour (1977:60)]
---
#...and then Yoruba happened

###While Gandour's generalizations seem to be strongly well-motivated, Yoruba seems to point at the opposite correlation:

###Nagano-Madsen (1992:132), in a prose reading task experiment, found vowels in CV words to be 143ms for H, 137ms for M and 139ms for L

###Note that although the duration differences reported in that study are small (and potentially non-significant), they challenge Gandour's generalizations (a) and (c)
--

###Another acoustic experiment in Yoruba to  test for significance with a slight modification compared to what Nagano-Madasen (1992) did
---

# The Experiment

###- Two male participants (both L1 speakers of Yoruba - born in Abeokuta, Nigeria and raised in Lagos, Nigeria)
--

###- 55 CV words (25 target and 30 fillers)

###- 2 * vowel (i, u, e, ẹ, e, ọ, a) + tone (H, M, L) + consonant ([t], [k])

###- In the carrier phrase "sọ kpé [...] kí ɱ gbọ́",  ("say [...] for me to hear")

###- Yielding a total of 212 randomized tokens
---

# Data processing

## The collected data was analysed in Praat in the following way:
--

###- Vowel duration was measured from the onset of F1 to its offset

###- The mean f0 of the middle 1/3 of each vowel was measured

###- Influences from the preceding H tone (of the carrier phrase) were minimized by reporting f0 of target words as ratios of the target tone over that of the preceding H.

###- In other words f0 could only have a value between 0 and 1

---
# Results/Descriptive Stats

###- Vowel duration:  127ms for H, 123ms for M and 97ms for L
###- The H tone has the largest SD which could be indicative of outliers and/or more variability in their duration

---
# The Scatterplot

.pull-left[
<img src="acal_presentation_files/figure-html/unnamed-chunk-22-1.png" width="720" />
]

.pull-right[
###* The scatterplot represents the mean vowel durations in _ms_ (accross speaker, repetitions and tone) by the mean pitch
###* The L-toned vowels are in the lower left quadrant of the plot while the H-toned ones cluster in the upper right quadrant
]

---
# The Model
##A general linear model was used to analyze vowel duration as a function of pitch:
##- Main effects and the pitch by vowel quality and consonant type interaction were assessed using the nested model comparison with the alpha set at 0.05
##- There was a main effect of pitch (F(1) = 15.93, p < 0.05) [p<0.00134 to be exact]

---
#The Model
##- The model containing pitch provided the best fit of the data (R^2 = 0.37)
##- Vowel duration increases as a function of pitch
##- That is, duration showed an increase of approximately 127 _ms_ +/- 34.79 se (t = 3.65, p < 0.05) per 0.1 _Hz_(ratio)
##So the difference in vowel duration as a function of pitch was not due to chance, in fact the significance of the difference  is important given R^2

---
#In conclusion
##- The data discussed confirmed that there is in fact a correlation between tone and vowel duration
##- But counter Gandour's generalization, the Yoruba data provided evidence for a positively correlated f0 and vowel duration
##- SOOOOO at best, the correlation between vowel duration and f0 is a UG property but the direction of this correlation must be parameter setting   
  
---
#An open can of worms?

##What does it mean for Yorùbá to have long vowels with the H tone since vowel length is not phonemically contrastive?
##Can vowel length alone be used as perceptual cue for different levels of pitch?

---
#Thanks!!