|
Article Excerpt
ABSTRACT. The present study used novel words to investigate adult Spanish learners' sensitivity to subtle, yet statistically predictable subpatterns in stress and diphthongization. It builds on recent findings showing that native speakers have remarkably fine-grained knowledge of the statistical patterns of their language (e.g. Bod et al. 2003, Bybee & Hopper 2001). Specifically, this experiment examined stress assignment in /n/-final nouns and the diphthong/ mid-vowel alternation in derived forms, following Aske's (1990) and Eddington's (1998) work with native speakers. Results showed that early in acquisition adults are sensitive to the fact that Spanish nouns ending in /en/ receive final or penultimate stress equally, whereas other /n/-final nouns almost exclusively receive final stress. Learner performance on diphthongization in derived words, a more complex statistical pattern, did not match the pattern in the Spanish lexicon, but was dependent on particular suffixes. The results are discussed in light of usage-based theories of language acquisition and grammar. *
1. INTRODUCTION. This study sought to explore the acquisition of two subtle phonological subpatterns in adult English speaking learners of Spanish as a second language (L2), stress assignment in /n/-final nouns and the diphthong/mid-vowel alternation in derived words. These subpatterns represent apparent exceptions to the rules of stress and diphthongization in Spanish, but corpus studies and native speaker behavior have shown that they are in fact productive. In particular, it has been shown that while most /n/-final nouns in Spanish pattern with other consonant-final nouns in receiving final stress, /en/-final nouns are equally likely to receive either final or penultimate stress. Furthermore, native speakers assign stress to nonce (invented) words according to this pattern (Aske 1990). Likewise, alternating diphthongs generally appear as diphthongs when stressed, and appear as monophthongs when unstressed. Eddington (1998) showed that each of a series of derivational suffixes was associated with a different probability that an alternating diphthong would surface as a diphthong in unstressed position, in apparent violation of the rule. Native speakers behaved in accordance with this pattern when given novel derivations based on alternating stems. It is striking that native speaker behavior follows so closely the statistical distribution of these small-scale and otherwise seemingly idiosyncratic subpatterns. The present study builds on these findings by asking whether adults learning L2 Spanish over the course of the first five semesters of university level Spanish also respond to this fine level of statistical detail in the lexicon, and it compares their behavior with the patterns in the native speaker lexicon. The approach taken is situated within a usage-based framework that provides a mechanism by which repeated exposure to lexical items and phonological patterns yields a dynamic and probabilistic grammar in which even very small-scale patterns may be productive. The following section presents a discussion of usage-based theory and its possible application to adult second language acquisition (SLA). This is followed by a review of the relevant research on stress assignment and diphthongization in Spanish. The paper then proceeds to describe the experimental methods and results.
2. USAGE-BASED THEORY AND SLA. A robust body of research in the first language (L1) domain increasingly reveals the highly nuanced and fine-grained knowledge that native speakers have of their L1 grammar (e.g. Barlow & Kemmer 2000, Bod et al. 2003, Bybee & Hopper 2001). This research extends to all domains of language use, including phonotactics (Frisch et al. 2000; Luce & Large 2001; Vitevitch & Luce 1998, 1999), phonology (Pierrehumbert 200la, Pierrehumbert 2001b, Pierrehumbert 2003), morphophonology (Baayen 2003, Bybee 2001), and syntax (Goldberg 1995, Manning 2003). It encompasses explanatory approaches to grammar as well as explorations of the effects of lexical probability on language processing (Frisch et al. 2000, Jurafsky 2003).
Common throughout this body of research is the notion that grammatical knowledge is based fundamentally on patterns of language in use. As speakers use and experience language, lexical networks of similarity, or SCHEMAS (Bybee 2001, Langacker 2000), emerge based on statistical patterns of phonetic, semantic, and other types of similarity that then can be used in the creation of novel words and utterances. Crucially, the link between frequent items (e.g. words, morphemes, phrases, and constructions) and the schemas that make up grammar is not severed, making grammatical knowledge dynamic and EMERGENT. That is, schemas evolve in response to changing patterns of language use over an individual's lifespan. There have been a number of attempts at modeling the dynamic nature of grammar using variations on Optimality Theory (Boersma & Hayes 2001), STOCHASTIC, that is, statistical models (Coleman & Pierrehumbert 1997; Pierrehumbert 2001b, 2003), usage-based models (Bybee 2001, Langacker 2000), in which grammar emerges across networks of similarity among lexical items, and ANALOGICAL models (Eddington 2000, 2002; Skousen 1989 1992), where grammatical patterns are determined by analogy to particular lexical items. All of these models depend on the accumulation of language experience through which statistically robust patterns emerge.
The usage-based approach taken in this paper is predicated on the notion that speakers have rich representations of items in the mental lexicon (Bybee 2001), including fine phonetic contrasts, morphologically complex forms, formulaic sequences, and constructions. In speech, speakers do not necessarily access underlying representations of morphemes, concatenate them to obtain inflected and derived forms, and pass these through the phonological grammar to obtain surface forms. Instead, provided that the speaker indeed knows the form s/he seeks, more frequent words are simply retrieved as a whole complex from the lexicon, complete with all their phonetic detail. On the other hand, when the speaker wishes to create a new word or utterance or to use infrequent items that do not have strong memory traces, then schemas, which emerge based on similarities between detailed representations of known forms, are used to determine or reconstruct the new or infrequent item (Bybee 2001). These schemas may differ widely in scale, from large-scale 'regular' rules to subtle 'irregularities' affecting only a few forms. A major implication for SLA, then, is that, if L2 learners' grammatical knowledge is likewise grounded in usage, they should develop similarly detailed knowledge of fine-grained patterns in their L2 based on their actual experience of the language. This study addresses the question directly by comparing how adult learners retrieve phonological detail in the lexicon for two different types of phonological rules.
A second set of implications from the usage-based approach to SLA involves the dynamic nature of grammatical knowledge in individuals. The notion that even L1 grammar continues to develop throughout life suggests continued operation of the mechanisms by which grammar is acquired, making these also available during SLA. While it has been clearly demonstrated that there is a loss of cognitive plasticity with increasing age, evidence that grammatical knowledge is dynamic indicates that a distinct, binary contrast between the child and adult states of mind is over-simplified. If adult L1 grammar continues to develop based on experiencing additional tokens of use, then L2 learners should also be able to exploit the patterns inherent in their growing second language experience in order to construct nuanced knowledge of subtle grammatical patterns, although other factors such as a learners' knowledge of his or her LI may also play a role. This hypothesis receives support from studies where both infants and adults have been shown to be able to segment strings of nonsense syllables in which the only cue available was the transitional probability between syllables (Aslin et al. 1998; Saffran et al. 1996a, 1996b). For this reason, examining SLA from a usage-based perspective may lead to a more nuanced understanding of age differences in language acquisition.
As an individual's experience of language in use increases, the statistical patterns inherent to that language become increasingly stable because each new token is added to larger and larger networks, an effect known as the POWER LAW OF LEARNING (Ellis 2002). This makes SLA an ideal domain for the exploration of adults' sensitivity to lexical statistics because early L2 lexical networks are small, thus making the statistical impact of growth (both in size and emergent structure) greater and potentially enabling researchers to observe stages in the development of statistical structures among adults. Given these possibilities for usage-based L2 research, surprisingly little has been done in this domain (but see Achard and Niemeier 2004, Ellis 2002). The present study seeks to address this issue by building on well established findings for native speakers' knowledge of small-scale subpatterns in Spanish stress assignment and diphthongization (Aske 1990; Eddington 1996, 1998; Face 2004), and by exploring the development of adult learners' knowledge of these two structures.
3. STATISTICAL PATTERNS IN SPANISH STRESS AND DIPHTHONGIZATION. Before proceeding with the present investigation, a description of the Spanish language structures to be explored, including their statistically regular subpatterns, is relevant. Spanish stress and diphthongization provide a useful starting point because, while they present striking regularities, for decades they also have supplied linguistic researchers from a variety of traditions with significant descriptive problems, some of which have been addressed using a statistical approach. In addition to a considerable series of generative and related accounts (e.g. Harris 1969, 1995; Hooper & Terrell 1976; Roca 1988), and an analysis of Spanish stress based on cognitive grammar (Farrell 1990), a number of researchers have experimentally explored the possibilities of usage-based accounts, which take into consideration the effects of frequency and analogy on variability in these structures (Aske 1990; Eddington 1996, 1998; Face 2004). This has resulted in findings that provide a firm reference point for work in the L2 domain. The particular subpatterns of stress and diphthongization to be examined here contrast in morphophonological complexity, as will be seen below. This contrast is well suited to investigating the timetable of development of statistical sensitivity; hypothetically, it should take longer for learner behavior to conform to a more complex pattern such as diphthongization in derived words than to a simpler pattern such as stress placement in /n/-final nouns.
3.1 SPANISH STRESS IN /N/-FINAL NOUNS. Stress in Spanish is a complex area of inquiry which has inspired a lengthy literature attempting to account for its striking regularity and puzzling irregularities (Eddington 2000; Harris 1983; Roca 1988, 1991). For an exploration of adult learners' fine-grained sensitivity to subtle statistical subpatterns in the lexicon it will be useful to focus on one specific subpattern among a well-circumscribed group of lexical items which contrast with the dominant trends in Spanish stress overall. The group of /n/-final nouns provides such a suitable subpattern. Singular nouns in Spanish most often receive penultimate stress when ending with a vowel, and final stress when ending with a consonant. Based on the data in an inverse dictionary (Faitleson-Weiser 1987), Aske (1990) however, calculated that nonverbs ending in /en/ (the vast majority of which are nouns) are fairly evenly split between those receiving final and penultimate stress, whereas nearly 100% of other /n/-final nonverbs (again, mostly nouns) have final stress, following the more general pattern for all consonant-final nouns. This contrasts with the case of /n/-final verbs, where stress is overwhelmingly penultimate. (1) These facts set up two potentially contrasting predictions. If the learners are paying attention to lexical category in subpatterns of stress assignment, as did the native speakers in Aske's experiment (1990), then they will be equally likely to...
|
