Vowel harmony and coarticulation in Khalkha Mongolian

This is a collaborative project with Prof. Indranil Dutta at the Speech Dynamics lab, Jadavpur University, Kolkata.

While speaking, sounds are not produced discretely, one after the other. Instead, while producing a sound our articulators are still affected by the previous vowel/consonant, and already starting to plan for the next, resulting in overlapping articulatory gestures. This kind of coarticulation causes acoustic variation, which means that the same phoneme might have different acoustic properties (and thus potentially sound different) when it occurs in different phonetic environments. Vowel-to-vowel (V-to-V) coarticulation is where the quality of a vowel is affected by a preceding or following vowel. Coarticulation can be both mechanico-inertial (the gestures of an earlier vowel perturb a later vowel) and anticipatory, (planning of the following vowel perturbs the former).

Khalkha Mongolian has a robust and regular vowel harmony system: vowels in polysyllabic words must share certain phonological features. Specifically, in most words the features [ATR] and [round] of non-initial vowels must match those of the initial vowel. This is a grammatical constraint of the language. Some vowels block vowel harmony, which gives non-harmonic vowel sequences (words in which the vowels have different values for [ATR] and/or [round]).

In this project, we want to understand how the low-level phonetic process of coarticulation interacts with this kind of grammatical knowledge. To do this, we are comparing coarticulation in harmonic (words where vowel harmony is seen) vs. non-harmonic (words where vowel harmony is blocked) sequences in the speech of Khalkha Mongolian speakers, to see if/how these differ.

As part of this project, we trained an acoustic model on our own corpus of Khalka Mongolian speech and used this to force-align and annotate the data. The acoustic model can be downloaded here.

The data for this study can be found here. These are the Praat and R scripts used for data processing, annotation, and acoustic analyses.

Preliminary analyses suggest that the direction of coarticulation in non-harmonic sequences is primarily anticipatory (right-to-left), opposite to the direction of vowel harmony. This is unlike harmonic sequences, where coarticulation is primarity mechanico-inertial (left-to-right), the same as that of vowel harmony.