Second language (L2) learners often struggle with achieving native-like production and perception of L2 sounds. While there is ample research on the non-native perception of English vowels by native speakers of languages with smaller vowel inventories, scant research exists up-to-date on the opposite scenario. The aim of the present study is to investigate the initial state for Californian English (CE) listeners' perception of two different smaller vowel inventories, namely Spanish and Portuguese vowels, before any perceptual learning has taken place. The objective is to determine how CE listeners' native vowel inventory of (at least) ten phonemes affects their non-native perception of the five Spanish and seven Portuguese vowel phonemes.

Monolingual speakers of English from Southern California (N = 18) completed a perceptual assimilation task, in which they labeled naturally produced tokens of Spanish and Portuguese vowels in terms of ten native English categories, as well as a categorial discrimination task with four Spanish and six Portuguese vowel contrasts. In the perceptual assimilation task, CE listeners mapped almost every Spanish and Portuguese vowel contrast to more than two native categories, resulting in many instances of Multiple Category Assimilation (MCA, Escudero and Boersma, 2002). Furthermore, some phonologically common vowels between Spanish and Portuguese, namely /e/, /o/, and /a/, were mapped differently to English categories, suggesting that cross-linguistic differences in the acoustic distributions of these sounds result in differences in their categorization by non-native listeners.

In the discrimination task, CE listeners were highly accurate at distinguishing all Spanish vowel contrasts but were less accurate on three Portuguese vowel contrasts, /i-e/, /o-u/, and /a-O/. Crucially, MCA was found to be responsible for listeners' lower discrimination accuracy only when it contributed to two vowels being perceptually assimilated to an overlapping set of native categories, i.e., when two different vowels had a high perceptual assimilation overlap score. The results are examined vis-à-vis two main models of second language speech perception, the Perceptual Assimilation Model (PAM, Best, 1995) and the Second Language Linguistic Perception Model (L2LP, Escudero, 2005). Implications for L2 acquisition of Spanish and Portuguese vowels by English-speaking learners are discussed.

This dissertation examines onset and coda /s/ lenition in the Spanish of El Salvador, a dialect in which this phenomenon is particularly advanced. That is, Salvadoran /s/ weakening is not only pervasive in both syllabic positions but manifests as allophones beyond the traditional tripartite conception. In addition to providing the first sociophonetic account of /s/ in El Salvador, this dissertation aims to address another gap in the literature: while /s/ weakening is the most-studied phonological variable in the history of the field, many accounts fail to make connections between observed patterns and important aspects of phonological theory. In my analysis I show that these connections, which are highly reliant on phonetic principles, are crucial to a more complete understanding of the phenomenon at hand.

Speakers for this study are 72 Salvadorans who participated in sociolinguistic interviews in El Salvador in 2015. I segmented and acoustically analyzed 200 occurrences of phonological /s/ per participant (n = 14,400 tokens) in Praat (Boersma & Weenink, 2016), and each token was subsequently coded for linguistic and extralinguistic variables including allophone, which was categorized as follows: [s]: a voiceless strident fricative; [z]: a voiced strident fricative; [sθ]: a voiceless approximant resulting from gestural undershoot; [h]: a voiceless glottal fricative; [ɦ] a voiced glottal fricative; and [∅]: deletion of the segment in its entirety.

Using these 14,400 tokens, this dissertation develops a phonetically-based phonological analysis of /s/ lenition within Optimality Theory (Prince & Smolensky, 1993/2004) in which the need to reduce articulatory effort cost (Markedness) while preserving important perceptual distinctions (Faithfulness) drives variation. I model the Salvadoran data via a maximum entropy algorithm, which assigns weights to OT-style constraints instead of strictly ranking them. Maximum entropy not only allows for variation but maximizes it, assigning probabilities to all possible output candidates according to shares of total harmony.

Markedness constraints in the grammar are styled following Kirchner (1998, 2004) and incentivize the weakening of difficult articulatory gestures in phonological environments that exacerbate their biomechanical effort cost. I find, for example, that /s/ is most likely to lenite when its flanking segments are more open (such as two vowels that are [-high]) or disagree with it in coronality, voicing, or both. I find that all five non-deleted variants can be modeled according to the interaction between their composite articulatory gestures and the phonological contexts in which they occur. Faithfulness constraints in the grammar, on the other hand, drive the preservation of perceptual cues in more salient prosodic positions. I find that the impetus to preserve both features [+strident] and [-voice] is highest in the strongest prosodic positions (phrase-initially > word-initially > syllable-initially) and decreases gradually in turn with position strength. Furthermore, I find that these differences are modulated by syllable stress, with tonic syllables blocking lenition at higher rates than atonic syllables.

With respect to demographic factors, this dissertation shows that while language-internal factors establish basic constraints in the grammar, the relative importance of preserving perceptual cues varies for different social groups; this is implemented in the grammar by scaling Faithfulness constraint weights up or down. I find that speakers from San Miguel, rural speakers, older Salvadorans, women, and those with lower levels of education prioritize Faithfulness (i.e., the preservation of important perceptual cues) less than their respective counterparts, resulting in higher rates of effort-based lenition. Furthermore, I find that speakers from these groups not only lenite /s/ at higher rates but are also more likely to both produce particularly nonstandard variants such as [sθ] and lenite /s/ more often and more extremely in more salient prosodic positions.

In sum, this dissertation contributes valuable data about a pervasive phenomenon in an understudied dialect of Spanish, including an in-depth exploration of the social factors that condition it. More broadly, this study presents a nuanced approach to Spanish /s/ lenition that is able to account for onset and coda /s/ weakening within a single analysis by situating the phenomenon within well-established theories of phonetically-based phonology.