Jarosz, Gaja. To appear / 2019. Computational Modeling of Phonological Learning. To appear in Annual Review of Linguistics (5).

direct link: http://websites.umass.edu/jarosz/2018/04/30/computational-modeling-of-phonological-learning/

Abstract

Recent advances in computational modeling have led to significant discoveries about the representation and acquisition of phonological knowledge and the limits on language learning and variation. These discoveries are the result of applying computational learning models to increasingly rich and complex natural language data while making increasingly realistic assumptions about the learning task. This article reviews the recent developments in computational modeling that have made the connections between fully explicit theories of learning, naturally occurring corpus data, and the richness of psycholinguistic and typological data possible. These advances fall into two broad research threads 1) the development of models capable of learning the quantitative, noisy, and inconsistent patterns that are characteristic of naturalistic data, and 2) the development of models with the capacity to learn hidden phonological structure from unlabeled data. After reviewing these advances, the article summarizes some of the most significant discoveries they have led to.

Download Manuscript

A User’s Guide to the Tolerance Principle
Charles Yang
direct link: http://ling.auf.net/lingbuzz/004146
August 2018
A discussion of conceptual and methodological issues in the application of the Tolerance Principle.

Constraints on QUD-addressing appositives: pragmatics or prosody?
Maria Esipova
direct link: http://ling.auf.net/lingbuzz/004144
August 2018
Can appositives address questions under discussion (QUDs), and what does it depend on? Koev (2012; 2013) claims that there is a contrast between clause-
final appositives, which can address QUDs relatively easily, and clause-medial appositives, which are at best marginal at QUD addressing. In this squib I look at sentences like the preceding one, which contain responses to multiple QUDs, and explore constraints on their con
figurations. I conclude that whether an appositive can address a QUD is determined by both pragmatic and prosodic factors that go beyond its linear position with respect to clause boundaries. This 5-page squib is a follow-up to this snippet: http://www.ledonline.it/snippets/allegati/snippets33004.pdf

Exploring sound symbolic knowledge of English speakers using Pokemon character names
Shigeto Kawahara, Jeff Moore
direct link: http://ling.auf.net/lingbuzz/004143
July 2018
This paper is a contribution to the studies of sound symbolism, systematic relationships between sounds and meanings. Two experiments were conducted in which native speakers of English were provided with pairs of pre-evolution and post-evolution Pokemon characters, the latter of which were always larger. The participants were given two name choices whose members are systematically different in some phonological properties. The results show the following sound symbolic patterns to be productive: the participants tend to associate post-evolution characters with (1) names containing voiced obstruents ([b, d, g, z]), (2) names with more segments, (3) names containing [a], (4) names containing [u], and (5) names containing coronal consonants. Overall, the current results suggest that phonological properties of names non-trivially affect the naming patterns of new creatures, implying that the relationships between sounds and meanings are not as arbitrary as modern linguistic theories generally assume.

Fusion is great, and interpretable fusion could be exciting for theory generation
Lisa Pearl
direct link: http://ling.auf.net/lingbuzz/004142
July 2018
Response to “Generative linguistics and neural networks at 60: foundation, friction, and fusion” by Joe Pater. From my perspective, Pater’s (2018) target article does a great service to both researchers who work in generative linguistics and researchers who utilize neural networks – and especially to researchers who might find themselves wanting to do both by harnessing the insights of each tradition. The article does three very useful things. First, it provides primers with historical overviews of each tradition. Second, it highlights what’s been achieved by the fusion of the generative linguistics theoretical framework and the neural networks modeling technique. Third, it notes the increasing interpretability of neural network models, which I think suggests a very exciting path forward for generating linguistic theories of representation.

On the role of culture in language and cognition
Daniel Everett
direct link: http://ling.auf.net/lingbuzz/004132
July 2018
This paper provides an overview of some recent research on how culture is causally implicated in the understanding of human cognition. In particular I review studies on the influence of culture on short-term memory, grammar, numerical cognition, and language evolution. I also provide a list of desiderata for research methodologies on the connections between culture and cognition and a direction for future research.

Against stored abstractions: A radical exemplar model of language acquisition
Ben Ambridge
direct link: http://ling.auf.net/lingbuzz/004131
July 2018
The goal of this paper is to make the case for a radical exemplar account of child language acquisition, under which unwitnessed forms are produced and comprehended by on-the-fly analogy across multiple stored exemplars, weighted by their degree of similarity to the target with regard to the task in hand. Across the domains of (1) word meanings, (2) morphologically inflected words, (3) n-grams, (4) sentence-level constructions and (5) phonetics and phonology, accounts based on independently-represented abstractions (whether formal rules or prototype categories) fail for two reasons. First, it is not possible to posit abstractions that delineate possible and impossible forms; e.g., that (a) rule in pool tables and data tables, but rule out chairs, (b) rule in “mist” and rule out “missid” as the phonological form of the past tense of miss, (c) rule in both John feared Bill and John frightened Bill but rule out *John laughed Bill. Second, for each domain, empirical data provide evidence of exemplar storage that cannot be captured by putative abstractions: e.g. speakers prefer and/or show an advantage for (1) exemplar variation even within word-meaning “category boundaries”, (2) novel inflected forms that are similar to existing exemplars, (3) n-grams that have occurred frequently in the input, (4) SVO sentences with he as SUBJECT and it as OBJECT and (5) repeated productions of “the same” word that are phonologically similar or, better still, identical. An exemplar account avoids an intractable lumping-or-splitting dilemma facing abstraction-based accounts and provides a unitary explanation of language acquisition across all domains; one that is consistent with models and empirical findings from the neuroimaging literature.

Antepenultimate stress in Spanish: In defense of syllable weight and grammatically-informed analogy
Martín Fuchs
direct link: http://ling.auf.net/lingbuzz/004120
July 2018
Spanish has a contrastive stress system with three major possibilities: antepenultimate, penultimate, and final stress. While penultimate and final stress are to some extent predictable, a major point of contention in the literature is whether antepenultimate stress assignment is rule-governed (Harris 1983; Roca 1991; i.a.). By examining different analogical and grammatically-informed models and their predictive power in capturing experimental data, I show that a Maximum Entropy model (Hayes & Wilson 2008) that includes syllable weight in its lexical representations is the best predictor of antepenultimate stress assignment. In doing so, I also dispute the claim that the trill in Spanish is a geminate tap (Harris 1983), and provide support for its status as a singleton consonant.