TAU Phonological Computation Lab

Our goal is to uncover the cognitive architecture of phonology, the component of the human mind that puts together linguistic sound representations. We integrate methods from theoretical linguistics and computer science by reverse-engineering the sound systems of individual natural languages, searching for abstract universal generalizations that hold across languages, and constructing machine-learning algorithms that simulate human phonological acquisition.
The lab is affiliated with the Linguistics Department and the Sagol School of Neuroscience at Tel Aviv University.

Research


The phonology of endangered Judeo-Arabic languages

There are at least eight understudied Jewish varieties of Arabic spoken in Israel by Jewish immigrants from Arab countries (Algeria, Egypt, Iraq, Libya, Morocco, Syria, Tunisia, and Yemen). Those varieties are endangered and are unfortunately expected to disappear in the next decades, as they are not passed on to children as native languages. Our goal is to record, document, and analyze the sound systems of those varieties, investigate their consequences for human cognition, and make the data and results available for the general public. This project thus serves the dual goals of contributing to the preservation of Judeo-Arabic varieties for future generations, while also contributing to our understanding of the human language capacity.

As part of this project, we have published the Tel-Aviv University audio database of Judeo-Baghdadi Arabic.


Phonological universals

At first sight, the phonological systems of the world’s languages seem remarkably diverse: Hebrew, Korean, Akan, and Ukrainian sound nothing alike. But when we carefully investigate languages at a sufficient level of abstraction, great similarities unravel. Are there universal phonological rules or principles that hold in every language? Do they reflect innate restrictions on the kinds of phonological systems that humans can acquire? What can they teach us about the computation of phonology in the mind? We investigate these questions by constructing candidate universals, testing them using cross-linguistic studies of phonological systems, and examining their consequences for theories of phonology.


Opacity

Opaque generalizations - generalizations that lose support on the surface - have played an important role in the development of phonological theory since the 1950's and remain at the center of debate until this day. Can humans acquire generalizations that are not observed on the surface? If so, how are such generalizations discovered? Which opaque patterns are attested in natural languages and which are unattested? What can the typology of opacity teach us about the architecture of phonology? We address these questions by discovering and analyzing attested cases of opacity in the world’s languages, by comparing theories of phonology based on their ability to account for opacity, and by constructing unsupervised learning algorithms that acquire opacity.


Computational models of phonological learning

How do children acquire the phonology of their language? What can theories of learning teach us about the division of labor between nature and nurture? In collaboration with the TAU Computational Linguistics Lab, we are exploring a theory of learning based on the principle of Minimum Description Length – a mathematical approximation of Occam’s Razor – according to which children look for the hypothesis that provides the simplest description of their input. We have been implementing computational learning algorithms for phonological acquisition based on this principle under competing phonological theories, with the goal of comparing theories of phonology based on their predictions regarding learning, and advancing our understanding of human phonological acquisition.

People


Researchers:

Ezer Rasin

Principal Investigator

rasin@tauex.tau.ac.il

Si Berrebi

Research Coordinator

cberebi@gmail.com

Eyal Marco

PhD student

eyalmrc@gmail.com

Moty Bistry

MA student

mobistry@gmail.com

Ben Flickstein

PhD student

Shvutya Bruck

MA student

shvutyabruck@mail.tau.ac.il

Saar Yahalom

Lab manager

Saaryahalom@mail.tau.ac.il


Research assistants:

Areej Khourieh

khourieh@mail.tau.ac.il

Imry Ziv

imryziv@mail.tau.ac.il

Samar Natour

samarnatour@mail.tau.ac.il


TAU collaborators:

Alma Frischoff

MA student

Yanir Mor

MA student

Vera Rusyanov

verarusyanov@mail.tau.ac.il

Publications


Courses


Opacity and phonological architecture
Fall semester

Opaque generalizations - generalizations that lose support on the surface - have played an important role in the development of phonological theory since the 1950's and remain at the center of debate until this day. In this course, we will examine the types of opacity attested in the world's languages and discuss their consequences for phonological architecture. In particular, we will compare serial theories (like rule-based phonology), parallel theories (like Optimality Theory), and their combinations (like Stratal OT); we will examine the evidence for abstract levels of representation and empty categories (such as traces) in phonology; and we will discuss the relationship between opacity and iterative or cyclic rule application. In addition, we will critically evaluate the empirical evidence suggesting that humans can acquire different kinds of opacity, taking into consideration results from experiments and the evolution of languages over time.


The phonology-morphology interface
Fall semester

This seminar will cover linguistic phenomena that have implications for the theory of the interaction between phonology and morphology. These phenomena include cyclic stress effects, the relationship between affix order and phonology, phonologically-conditioned suppletive allomorphy, and blocking in nonderived environments, among others. Examples of theoretical questions we will discuss include whether morphology precedes phonology (as in Distributed Morphology) or whether the two modules are interleaved (as in Lexical Phonology and Morphology), whether phonology is sensitive to syntactic phases, whether allomorph selection takes place in the phonology, and whether phonological computation is cyclic or transderivational.


Computational phonology
Spring semester

This seminar covers computational tools that can inform the study of phonology. In the first part of the semester, we will explore recent advances in formal language theory. Focusing on the sub-regular region of the Chomsky Hierarchy of formal languages, we will discuss the computational complexity of phonological patterns attested in natural language, as well as the complexity of unattested hypothetical patterns. We will also get familiar with computational implementations of rule-based phonology and Optimality Theory. In this part of the seminar we will ask questions like: are stress and tone more complex than segmental processes? How can an optimal surface form be selected from among an infinite candidate set using graph theory? What is the power of Optimality Theory's optimization?
In the second part of the semester, we will review a variety of competing machine-learning algorithms from the phonological literature, including supervised and unsupervised algorithms and probabilistic and non-probabilistic algorithms, and discuss their advantages and disadvantages as candidates for a theory of human phonological acquisition. In this part we will ask questions like: What is the role of simplicity in the acquisition of rules and constraints? How can token frequency affect phonological acquisition? Can opaque generalizations be acquired without morphological or semantic knowledge?


Phonological and morphological theory through the lens of colloquial Arabic
Fall semester

In this seminar we will examine phonological and morphological phenomena in colloquial dialects of Arabic and their connection to linguistic theory. The focus will be on local varieties of Arabic: varieties of Palestinian Arabic, as well as varieties of Judeo-Arabic spoken by Jewish immigrants from Arab countries (Baghdadi, Tripolitanian, Moroccan, and others). Those varieties are understudied from a theoretical perspective and some of them are endangered and are expected to disappear in the next decades.
This seminar has two goals:
1) To understand what colloquial Arabic has taught us and can further teach us about the human language capacity. We will examine phenomena in colloquial Arabic and the role they have played in the development of phonological and morphological theory since the 1970's, such as cyclic stress assignment, syllable structure and directional vowel epenthesis, long-distance emphasis spread, non-concatenative morphology (root and pattern, broken plural, etc.), discontinuous agreement, and more. Then, we will discuss old and new papers on the local dialects based on the students' interests.
2) To enrich the knowledge and data about understudied colloquial varieties of Arabic, and in the case of the Jewish dialects, contribute to their preservation. To this end, we will discuss different aspects of field research, and each student will pick a local variety of Arabic to investigate during the semester.

Contact


You are welcome to contact us at phoncomplab@gmail.com.
Our address is: Department of Linguistics, Webb Building, Room 408, Tel-Aviv University, Ramat Aviv, Tel Aviv 69978.