Harmonic Grammar

This paper is a short version of the talk presented at Going Romance XXXII (December, 2018).

In this paper we analyze, from an experimental and formal perspective, the interaction and the implicational relationships between vowel reduction and word-final nasal deletion in Catalan loanwords. We present the results of both a production and a perception test carried on 31 speakers from the Barcelona area. Loanwords susceptible to undergo both nasal deletion and vowel reduction display different patterns. The most common solution is underapplication of both processes, followed closely by underapplication of nasal deletion alone and at a large distance by the application of both processes. Finally, underapplication of vowel reduction and application of nasal deletion is unattested, that is, it is an impossible nativization pattern. The typology of possible and impossible nativizations and the implicational relationships between the processes under scrutiny are analyzed in the framework of Harmonic Grammar under Weighted Scalar Constraints, following recent proposals by Hsu & Jesney (2017, 2018).

Keywords: loanwords, (im)possible nativizations, implicational relationships, Harmonic Grammar, Weighted Scalar Constraints, Catalan

Jeong, Cheonkam and Sung-Hoon Hong. 2017. Noisy Harmonic Grammar modeling of flapping in American English based on a statistical analysis. Studies in Phonetics, Phonology and Morphology 23.1. 117-143. This paper investigates the gradient aspect of flapping in American English, considering both the language internal factors stress and morphological complexity and the language external factor lexical frequency. To reflect the gradient aspect of flapping, flapping rates were regarded as dependent variables, and a statistical analysis was conducted with both language-internal and language-external factors. Due to the range of dependent variables [0, 1], a zero/one inflated beta regression was conducted. The results verified that the more frequent a carrier word containing a word-medial /t/ is, the more likely it is for the word-medial /t/ to be realized as a flap, and that the word-medial /t/ in a morphologically simple word is more likely to be realized as a flap than one in a morphologically complex word. Thereafter, Noisy Harmonic Grammar analyses were performed, and the extended version showed an improvement over the original model of 83.228%. (Hankuk University of Foreign Studies)

This paper proposes an account of harmony systems that have asymmetries in directional spreading, directional blocking, directional bounds, and trigger strength. I show that adopting a derivational version of Optimality Theory with weighted constraints – Serial Harmonic Grammar – as the analytic framework can successfully generate these systems in a general way without relying on problematic Local Constraint Conjunction. Additionally, building on earlier work, the serial nature of the grammar is used in order to prevent predictions of implausible typological pathologies found in existing parallel Optimality Theory and Harmonic Grammar analyses. Small typologies generated by this grammar are presented, and a complex test case from Chilcotin is examined.

This chapter reviews major results as well as areas of ongoing research on computational models of learning with violable constraints, in the sense of Optimality Theory. The review encompasses learning in classic OT as well as learning in related frameworks that formalize constraint interaction as (probabilistic) weighting or ranking. The learning problem is decomposed into several subproblems, which are considered in turn. The discussion emphasizes the challenges posed by these various subproblems and the insights and differences of the proposed solutions. The chapter first discusses the narrow grammar- learning subproblem, the problem of learning rankings or weightings given data along with their full structural descriptions. Next, the discussion turns to several distinct approaches to the broader grammar-learning subproblem, which makes the more realistic assumption that hidden structure, such as prosodic structure or underlying representations, are not available to the learner and must themselves be learned. The chapter then discusses the subset problem and various approaches that have been proposed for learning restrictive constraint grammars. Finally, the chapter briefly reviews work on acquisition modeling, work that connects the predictions of computational models and the process by which children acquire language.

This article reports on an experiment with miniature artificial languages that provides support for a synthesis of ideas from Usage-based Phonology (Bybee 1985, 2001, Nesset 2008) and Harmonic Grammar (Legendre et al. 1990, Smolensky & Legendre 2006). All miniature artificial languages presented to subjects feature velar palatalization (k → tſ) before a plural suffix, -i. I show that (i) examples of -i simply attaching to a [tſ]-final stem help palatalization (supporting t → tſi over t → ti and p → tſi over p → pi), a finding that provides specific support for product-oriented schemas like 'plurals should end in [tſi]'; (ii) learners tend to perseverate on the form they know, leveling stem changes, which provides support for paradigm-uniformity constraints in favor of retaining gestures composing the known form, for example, 'keep labiality'; and (iii) the same plural schema helps untrained singular-plural mappings more than it helps trained ones. This result is accounted for by proposing that schemas and paradigm-uniformity constraints clamor for candidate plural forms that obey them. Given that competition is between candidate outputs, the same schema provides more help to candidates that violate strong paradigm-uniformity constraints and are therefore weak relative to competitor candidates. A computational model of schema extraction is proposed.

This paper explores the relative merits of constraint ranking versus weighting in the context of a major outstanding learnability problem in phonology: learning in the face of hidden structure. Specifically, the paper examines a well-known approach to the structural ambiguity problem, Robust Interpretive Parsing (RIP; Tesar and Smolensky 1998), focusing on its stochastic extension as first described by Boersma (2003). Two related problems with the stochastic formulation of RIP are revealed, rooted in a failure to take full advantage of probabilistic information available in the learner’s grammar. To address these problems, two novel parsing strategies are introduced and applied to learning algorithms for both probabilistic ranking and weighting. The novel parsing strategies yield significant improvements in performance, asymmetrically improving performance of OT learners. Once RIP is replaced with the proposed modifications, the apparent advantage of HG over OT learners reported in previous work disappears (Boersma and Pater to appear).

This paper discusses the literally one-off case of nasal dissimilation in Bangla to show that the emergence of this hardly visible phonological phenomenon in an obscure corner of the lexicon makes a strong argument for the universality of phonological constraints. The argument being that if all phonological constraints are drawn from the human sensory-motor or cognitive systems, they must be universal and part of the innate biases in all human languages. Many of these might not be visible in the surface representation of these languages as their individual weights (Pater 2009) might be less than the weights of other markedness or faithfulness constraints that might be violated by competing candidates. Nevertheless, it is possible that the cumulative weight of a number of such insignificant constraints could add up to overtake the weight of some heavier, more apparent constraint, resulting in neutralization processes to emerge in very limited parts of the lexicon. In this paper, I show that the emergence of OCP-Nasal in Bangla is a case of The Emergence of the Unmarked (McCarthy 1994), which can be analyzed using standard OT by using constraint conjunction. However, using weighted constraints instead of ranked constraints explains away the nuances of the conditions that allow this well-formedness condition to emerge in Bangla.

We show that a class of cases that has been previously studied in terms of learning of abstract phonological underlying representations (URs) can be handled by a learner that chooses URs from a contextually conditioned distribution over observed surface representations. We implement such a learner in a Maximum Entropy version of Optimality Theory, in which UR learning is an instance of semi-supervised learning. Our objective function incorporates a term aimed to ensure generalization, independently required for phonotactic learning in Optimality Theory, and does not have a bias for single URs for morphemes.
This learner is successful on a test language provided by Tesar (2006) as a challenge for UR learning. We also provide successful results on learning of a toy case modeled on French vowel alternations, which have also been previously analyzed in terms of abstract URs. This case includes lexically conditioned variation, an aspect of the data that cannot be handled by abstract URs, showing that in this respect our approach is more general.

The user guide for the new release of OT-Help version 2.0. (http://web.linguist.umass.edu/~OTHelp/) OT-Help is software for calculating typologies for parallel ‘classic’ Optimality Theory (OT), parallel Harmonic Grammar (HG), Harmonic Serialism (i.e., serial OT), and serial HG as well as a solver for parallel OT and parallel HG. For serial typologies, OT-Help allows user-defined constraints and operations for candidate generation.

In this paper we explore phonological nativization patterns in Catalan loanwords, and we show, on the basis of a production and a judgment test, that the three processes under scrutiny (word-final –n deletion [ND], vowel reduction of unstressed mid-vowels [VR], and vowel laxing of stressed mid-vowels [VL]) interact in an asymmetrical way. We argue that these asymmetrical interactions can be straightforwardly formalized resorting to Harmonic Grammar with Scalar Weighted Constraints (Hsu & Jesney 2017, 2018), in which faithfulness constraints acquire an increasing relevance from the core to the peripheral strata and in which the scaling factor intervals for each stratum abstractly reproduce the most frequent and the least frequent patterns in loanword adaptation.

There has been significant progress in recent years on the learnability of hidden structure in phonology, but a complete solution to this problem remains one of the most significant obstacles to realistic modeling of phonological learning. Many criteria are relevant when evaluating learning models, including their success rates, their efficiency, and their expressiveness. The present study investigates the efficiency of probabilistic learning models for Optimality Theory and Harmonic Grammar equipped with Robust Interpretive Parsing (RIP; Tesar & Smolensky 1998) and the two parsing strategies proposed by Jarosz (2013) – Resampling RIP (RRIP) and Expected Interpretive Parsing (EIP) – in the domain of metrical stress. The results indicate that RRIP and EIP improve not only end- state success rates, but also the speed of learning: learners equipped with the new parsing strategies converge on correct target grammars after fewer iterations of learning. The results suggest the new parsing strategies allow learners to extract information from the learning data more efficiently, improving the potential for learning to scale to the full problem of phonological learning. The discussion emphasizes the need for further work exploring a range of computational criteria in evaluating models of learning.

In this paper we explore phonological nativization patterns in Catalan loanwords, and we show, on the basis of a production and a judgment test, that the three processes under scrutiny (word-final –n deletion [ND], vowel reduction of unstressed mid-vowels [VR], and vowel laxing of stressed mid-vowels [VL]) interact in an asymmetrical way. We argue that these asymmetrical interactions can be straightforwardly formalized resorting to Harmonic Grammar with Scalar Weighted Constraints (Hsu & Jesney 2017, 2018), in which faithfulness constraints acquire an increasing relevance from the core to the peripheral strata and in which the scaling factor intervals for each stratum abstractly reproduce the most frequent and the least frequent patterns in loanword adaptation.