Research Statement

The stunning ability to communicate abstract messages is a quintessential human trait that uniquely defines us in the animal kingdom. At the same time, human language is a complex behavior that presumably draws in large parts on evolutionarily younger neural/cognitive systems. Research in the Human Language Processing (HLP) Lab seeks to understand the computational cognitive systems that allow the human brain to communicate information at a rate and complexity that far exceeds that of other animals.
Read more »

Fieldwork

Yucatec fieldwork
T. Florian Jaeger and Katrina Housel Furth during fieldwork in the Yucatan in 2009

Typological analysis

family distribution

How can one estimate the Type I and II error rates of typological analysis? How can one model genealogical dependencies between languages without loss of power? How can one estimate the effects of language contact? In some of our work, we have discussed the used of multi-level models for typological research. Power analyses can be conducted by repeatedly sampling from plausible pseudo-world (see Figure which shows one such pseudo-world). We have introduced and tested simple algorithms that capture most variance in typological properties due to language contact. Interested? Read more:

Robust language understanding in a variable world

noisy channel + adaptation

To understand each other, we must infer messages from the linguistic signal, which is often perturbed by noise. Our brain seems to overcome this problem by drawing on implicit knowledge about the distribution of signals and linguistic categories. Going further, some research from our lab suggests that listeners and readers can adapt their expectations about these distributions based on recent input. This is a critical ability, allowing efficient processing even in the face of inter-talker variability in pronunciation, as well as lexical and syntactic preferences. Read more about this with regard to

Speech Perception

vowel normalization

Example of cross-talker vowel variability before (left) and after (right) normalizing for differences in vocal-tract length based on F3. Top Panel: the average vowel space for adult male and adult female talkers in the vowel corpus collected by Hillenbrand et al. (1995). Talkers in this corpus are from the Northern dialect region of American English. Bottom Panel: the degree of overlap among adult male productions of /ʊ/ and adult female productions of /u/. Plots show individual vowel tokens (small dots), category means (large dots), and 95% confidence ellipses. Note that the unnormalized male and female vowel spaces (top left panel) have approximately the same geometric configuration, but the male vowels are characterized by comparatively lower absolute F1 and F2 values, which reflects the fact that longer vocal tracts resonate at lower frequencies than shorter vocal tracts. As a result, the distribution of adult male tokens of /ʊ/ is highly overlapping with the distribution of adult female tokens of /u/ in F1xF2 space (bottom left panel). That is, the same acoustic information maps onto different phonological categories with different probabilities depending on the talker’s sex. Hence neither F1 nor F2 provides reliable information for discriminating these vowel categories across talker sex. Normalizing F1 and F2 based on F3 (which is correlated with vocal tract length) considerably reduces the difference between the average male and female vowel spaces (top right panel), while preserving the overall shape of the space (i.e., the relative position of vowels). As a result of F3-normalization, tokens of /ʊ/ and /u/ are discriminable within and across talker sex.

Alignment

When we interact with one another, we tend to align our behaviors, including the way we talk. Linguistic alignment is due, at least in part, to automatic and subconscious aspects of language production, but is also influenced by social factors like one's attitude toward their interlocutor. Using a novel web-based paradigm, we investigated how speakers adapt (align or potentially anti-align) their sentence productions in response to their environment. The figure shows mixed logit model predictions for social factors that affect syntactic alignment (i.e., the tendency of participants to use the same sentence structures that they've heard other talkers produce). PO = the prepositional object structure (i.e., give [the object] [to the recipient]). DO = the double object structure (i.e., give [the recipient] [the object]). Hearing another talker produce either DO or PO structures increased the likelihood that participants would subsequently use the same sentence structure (i.e., overall alignment). However, the degree of alignment was influenced by a range of social factors, including perceived interpersonal similarity. Note that alignment rates were never reliably below baseline (i.e., no anti-alignment). Together, these results suggest that alignment is largely automatic but socially mediated.

RSS