Cue-based features

Phonological features are great for analyzing patterns and seeing broader connections cross-linguistically. Phonetic cues are great for understanding individual implementations of features in production and perception. Why not put them together?

The results of experimental work (both mine & from many many other linguists!) provide evidence for the importance of acoustic cues in the realization of phonological contrast. A single binary feature, such as [±voice] or [±spread glottis], is unable to fully represent complex voicing contrasts in obstruents. In response, I developed a framework of cue-based features. The goal of cue-based features is to take what is already known about cues and incorporate it directly into featural representations in formal phonology. This two-part model represents the interface of phonetics and phonology.

How it works

In cue-based features, each feature is made up of a bundle of acoustic cues. These cues have individual weights that can differ to model speaker variability, but there are cumulative thresholds that must be hit across the cues in order to activate [+feature] or [-feature]. Cue weight totals that fail to reach the threshold do not activate a feature, leaving the segment unspecified for that feature. This model allows for variation in how a feature is phonetically implemented in a language, while still capturing larger cross-linguistic patterns.

Cue-based features map acoustic cues to a single feature, but the cues themselves are not bound to a single segment. This allows for more flexibility in phonological contrast: a difference in phonological voicing can be signaled by the preceding vowel duration or the fundamental frequency of the following vowel. This flexibility also allows for the modeling of changes in cue weights over time, and how these changes can lead to phonological changes such as incomplete neutralization and tonogenesis. Finally, cue-based features allow contrast to have levels: two segments can be more or less contrastive, depending on what the cumulative cue weights are and their relation to a particular featural threshold.

I created this model to account for complex voicing contrasts, but it has grown and evolved! I’m currently working on modeling phonological processes such as palatalization, fortition, lenition, and nasalization. There is also a Python implementation in the works thanks to Jessica Gaines (PhD Candidate at UC Berkeley - UCSF Graduate Program in Bioengineering). You can read more about this model in this manuscript.