Sounds, Statistics, and Software: Santiago Barreda

Linguistics is a broad, methodologically diverse discipline. A language field worker doing descriptive work on an undocumented language may spend most of the day transcribing that language in phonetic symbols. Meanwhile, a psycholinguist may be gathering subjects for an experiment involving subjects reading sentences while their brains are scanned. Down the hall, a syntactician may be drawing complex trees and developing a rigorous formalism to account for some complex word movement operations. 

Santiago Barreda is primarily, but not solely, a phonetician. Having received his PhD from the University of Alberta in 2013, Barreda went on to work as a postdoctoral researcher in the University of Arizona’s Department of Physiology. In 2014, he joined the faculty at UC Davis as an assistant professor of linguistics. 

So far, Barreda has not only conducted phonetics research using a wide range of experimental techniques—he has contributed to the development of statistical methodologies and software tools, too. Vaulting metaphorical departmental walls, Barreda's research incorporates cutting-edge developments from a broad range of disciplines.

From phonology to phonetics

Barreda began as a phonologist. Phonologists study abstract categories of sound in speakers' minds. For example, while English spelling only employs five vowel letters, English speakers distinguish a great many vowel sounds.

Originally drawn to phonology’s puzzle-like task of identifying patterns of sounds in languages, Barreda eventually became more interested in the phonetic features used to differentiate phonemes in phonology. While he remains fascinated by the same sorts of things that originally led him to become a researcher in linguistics, Barreda’s focus on the actual sounds produced by speakers—as well as his commitment to statistical analysis and other quantitative techniques—means that he is now best categorized as a phonetician.

Empirical advances

Phonetics has been influential in pushing linguistics research to be more empirical. In many ways, phonetic data lends itself well to being studied as numeric data. Firstly, phonetic data is relatively plentiful. Secondly, phoneticians often study the least abstract structures in linguistics, many of which can be described using numeric parameters.

As a phonetician, Barreda is interested in 1) how we know what we are hearing, 2) how we know who is saying it, and 3) how these two things relate. When we hear a person speak, that speaker provides some context for interpreting the actual physical sounds that we hear. (This is not the case with, say, color: a given color may seem different in different contexts, as “the dress” infamously proved.) But how might we interpret those sounds when deprived of that context?

Synthesizing stimuli

In "Investigating the Use of Formant Frequencies in Listener Judgments of Speaker Size" (2015), Barreda conducted an experiment using synthesized vowel sounds to study how a number of acoustic parameters are used by listeners to make judgments about speakers' height. To synthesize vowels, Barreda used a software synthesizer to manipulate specific acoustic parameters and create experimental stimuli. His analysis pointed to no single acoustic parameter as a listener's main criterion for assessing speaker size. Rather, it suggested that listeners rely on token word-specific information to make such judgments.

In 2015, Barreda received an ISS Individual Research Grant for a project entitled Learning Social Cues from Language: The Development of Gender Perception from Speech in Hearing and Deaf Children. “In the past, I have worked on training methods to teach listeners with normal hearing to better use the acoustic information in speech sounds,” he said. “I was motivated to find out if any of these training methods could be used to improve the perceptual abilities of listeners with cochlear implants. Assuming the training proves to be effective in general and appropriate for children, we plan to use the training method with children and adults with such implants.”

Software and statistics

Barreda has also worked on statistical methodologies useful for data interpretation in any field. In "Bayes Covariant Multi-Class Classification" (2016), he approached the general problem of multi-class classification (sorting data into multiple categories, as opposed to two just categories in binary-class classification) using an application of Bayesian statistics.

As an open source software developer, Barreda has also contributed to the language research community, developing software for phonetics research. For example, he created phonTools (2015) to facilitate the organization, display, and analysis of data frequently encountered in phonetics research. He also contributed a package to R, aiming to simplify many repetitive tasks in his own and others’ research. Linguistics students at UC Davis (the author of this article included) count themselves as grateful benefactors of these tools, as well as of his broader contributions to quantitative research. 

Learn more about Santiago Barreda.