An artificial language learning study of the relationship between stress position and quantity sensitivity in stress patterning

Stress and accent are important prosodic characteristics of the sound patterns of many human languages, and although their acquisition is crucial for the mastery of a second language, non-native prosodic patterns can be difficult to learn. Decades of research in linguistics has produced a typology of existing stress patterns and a basis for estimating the prevalence of these types in human languages.

A key goal of linguistics is to understand natural patterns in language, and frequency of occurrence has been taken as one indicator of naturalness. The premise behind this research program is that another way of identifying natural patterns is to study ease of learning: more natural patterns should be more easily acquired than less natural patterns.

However, while psycholinguists have studied listeners’ ability to perceive stress, little research has investigated the ease with which humans learn characteristics of stress patterns and the latent biases that influence learning. This project investigates how adult speakers of three natural human languages learn non-native stress patterns in a series of artificial language learning experiments.

The cross-linguistic nature of the research program is critical, given that an understanding of human learning requires studying speakers with diverse cultural and linguistic experiences. The research program complements received knowledge about stress typology, contributes new information to cross-disciplinary studies of human behavior and cognition, and supports education.

A series of artificial language learning studies investigates relationships between edge alignment (whether stress is positioned at the beginning or end of a word) and quantity sensitivity (whether syllables with more content are preferred locations for word stress) using a poverty of the stimulus design. In this approach, participants are exposed to linguistic forms that are sufficient for them to learn components of a complex stress pattern in a constructed mini-language.

However, forms necessary to acquire a complete generalization are withheld during an initial training phase. The missing forms are then introduced in a subsequent test, along with new forms that either are or are not consistent with the learned aspects of the pattern. Participants decide which new forms presented in the test are compatible with the mini-language.

The participants’ inferences, when presented with new data, provides information about untaught preferences that are interpreted as evidence for cognitive biases that influence learning. Methodologically, controls are used to tease apart “built-in” preferences from biases which might be related to patterns in the native language. The project includes plans for four interrelated studies with native speakers of three distinct languages in the three project years, for a total of twelve experiments.

Participants will be tested at the University of Texas at Austin and at off-campus sites. This research contributes new information about ease of learning in regard to stress patterning and about humans’ use of learned information to complete generalizations about linguistic patterns when they are initially exposed to an impoverished data set.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.