Weight effects on stress: lexicon and grammar (PhD Thesis)
Guilherme Duarte Garcia
November 2017

This thesis examines weight effects on stress and proposes a probabilistic approach based on the notion that weight is gradient, not categorical. Arguments for this proposal are divided into three main chapters, which examine and statistically model weight in the lexicon (Chapter 1), weight in the grammar (Chapter 2), and the interaction of weight and footing (Chapter 3). The statistical analyses in Chapters 2 and 3 also discuss how our linguistic expectations regarding weight effects can be incorporated in statistical models through the use of mildly informative priors, and to what extent the fit of such models compare with that of models based on non-informative priors. In Chapter 1, I examine weight effects in the Portuguese lexicon, and show that they are considerably more intricate than what is assumed in the literature. Previous weight-based studies consider that weight only affects stress in word-final syllables, and that weight is categorical (e.g., Bisol 1992, Lee 2007). In other words, syllables in Portuguese are traditionally classified as heavy or light. I show that weight should not be seen as categorical. By exploring a comprehensive lexicon (Houaiss et al. 2001), I demonstrate that heavy syllables have a gradient effect on stress. This effect is modulated by the position of a given heavy syllable in the stress domain as well as its segmental count. This entails that weight effects are not restricted to word-final syllables. Rather, all syllables in the stress domain present some weight effect on stress. One such effect is in fact puzzling: antepenultimate light syllables are more stress-attracting than antepenultimate heavy syllables. This contradicts the typology of weight and stress, since heavy syllables are not expected to repel stress in weight-sensitive languages (Gordon 2006). Given the non-categorical patterns observed in the lexicon, I propose a probabilistic approach to stress in the language. To demonstrate the empirical advantage of such an approach, I show that the accuracy of probabilistic predictions is substantially higher than that of categorical predictions. In Chapter 2, I examine to what extent these lexical patterns in Portuguese are captured by speakers' grammars. First, I show that speakers do generalise the weight gradience in the lexicon to novel words. The effects monotonically weaken as we move away from the right edge of the word, which mirrors what is found in the lexicon (Chapter 1). Second, I show that speakers do not generalise the typologically contradictory pattern found in antepenultimate syllables in the lexicon. Instead, speakers assign positive weight effects to all syllables in the stress domain; i.e., they repair the negative weight effect in question. Previous findings in the literature on phonological (under)learning have shown that unnatural (or contradictory) patterns are harder to learn, and are often ignored by speakers (e.g., Hayes and Londe 2006, Hayes et al. 2009, Becker et al. 2011, Becker et al. 2012). Chapter 2 shows that speakers can go beyond ignoring such patterns: they can in fact repair them. The probabilistic approach presented in Chapters 1 and 2 raises the question of how footing impacts stress in a language such as Portuguese, where weight effects are gradient. Indeed, a non-categorical weight-based approach poses important challenges to footing. In Chapter 3, I argue that Portuguese does not offer compelling evidence for the foot. First, the gradient weight effects found in the lexicon and in speakers' behaviour cannot be captured with any foot type, given that even antepenultimate stress is directly affected by weight. Second, no phonological process (e.g., truncation, reduplication, hypocorisation) makes reference to the foot. Third, different foot types have been proposed across the literature because of contradictory patterns of stress location---patterns which are mirrored in truncation, reduplication, and hypocorisation in the language. Fourth, subminimal words are not only common in the Portuguese lexicon, but are also productive in the language. As discussed in Chapter 3, the evidence against footing in Portuguese is compelling, and I therefore conclude that the language does not have feet. To further strengthen this argument, I turn to English, where the evidence for footing is robust. Even though English and Portuguese present similar stress patterns on the surface, I show that these two languages are fundamentally different. Unlike in Portuguese, stress patterns in the English lexicon and in speakers' grammars exhibit weight effects that are predicted if one assumes moraic trochees and extrametricality in the language (e.g., Hayes 1982). The probabilistic weight-based approach to stress adopted in this thesis thus concludes that feet are parametric (following, e.g., Özçelik (2013, 2014)), and are therefore present in English, but absent in Portuguese.
Format: [ pdf ]
Reference: lingbuzz/003615
(please use that when you cite this article)
Published in: McGill University
keywords: phonology, stress, weight, probabilistic grammar, portuguese, english, phonology
previous versions: v2 [November 2017]
v1 [June 2017]
Downloaded:189 times


[ edit this article | back to article list ]