Two ways to listen: Do L2-dominant bilinguals perceive stop voicing according to language mode?

Abstract

How listeners categorize two phones predicts the success with which they will discriminate the given phonetic distinction. In the case of bilinguals, such perceptual patterns could reveal whether the listener's two phonological systems are integrated or separate. This is of particular interest when a given contrast is realized differently in each language, as is the case with Greek and English stop-voicing distinctions. We had Greek–English early sequential bilinguals and Greek and English monolinguals (baselines) categorize, rate, and discriminate stop-voicing contrasts in each language. All communication with each group of bilinguals occurred solely in one language mode, Greek or English. The monolingual groups showed the expected native-language constraints, each perceiving their native contrast more accurately than the opposing nonnative contrast. Bilinguals' category-goodness ratings for the same physical stimuli differed, consistent with their language mode, yet their discrimination performance was unaffected by language mode and biased toward their dominant language (English). We conclude that bilinguals integrate both languages in a common phonetic space that is swayed by their long-term dominant language environment for discrimination, but that they selectively attend to language-specific phonetic information for phonologically motivated judgments (category-goodness ratings).

Introduction

When a perceiver is bilingual, performance on cross-language speech perception tasks may indicate whether the first language (L1) and second language (L2) are integrated or kept separate. Performance equal to that of native, monolingual listeners in each language has been interpreted as evidence that bilinguals have developed separate L1 and L2 phonetic categories, whereas intermediate discrimination ability has been taken to indicate that the bilinguals rely on a set of merged categories for both languages (Flege, 2002).

Bilinguals may differ in their perception from monolinguals because of their L1–L2 sensitivity and the interaction that occurs between their languages. Even fluent, early bilinguals who use their L2 daily may show some L1 influence on their perception of the L2. For example, Spanish–Catalan¹ bilinguals retain perceptual dominance from their L1 (Spanish) when categorizing and discriminating Catalan contrasts that Spanish lacks, despite years of daily experience with both languages in a bilingual environment (Pallier et al., 2001, Sebastián-Gallés and Soto-Faraco, 1999). Conversely, however, there is also much evidence that learning an L2 may change the way the listener perceives the L1 (Flege, 1993, Flege et al., 2003, Flege et al., 1999).

Several theoretical models have been proposed to account for L2 users' perception of speech segments. The Speech Learning Model (Flege, 1995, Flege, 2002) was designed to account for L2 speech production, but assumes that it is strongly influenced by perceptual biases and therefore seeks to explain the perceptual reorganization that occurs due to L2 acquisition. The Native Language Magnet model (Kuhl, 1992, 1993) seeks to explain how early attunement to the native language (L1) warps the perceptual space, such that learning after early childhood is influenced by that L1. The Perceptual Assimilation Model (PAM; Best, 1994, Best, 1995) accounts for naive listeners' discrimination of unfamiliar nonnative contrasts, and its extension to L2 perceptual learning, PAM-L2 (Best & Tyler, 2007) predicts the success with which learners should discriminate L2 contrasts. According to PAM-L2, L2 phones are first assimilated into already existing L1 categories or dissimilated from existing categories and established as new categories firstly on a phonetic level. As the learner's L2 vocabulary expands they attune to the higher-order organization of the L2 phonology, and phones come to be discriminated on the basis of meaningful categorial differences that are lexically relevant in the L2.

However, none of the aforementioned models were designed to specifically address the performance of early fluent bilinguals. Here, our purpose is not to contrast the various models of nonnative speech perception, but rather to focus on testing PAM-L2 predictions, which appear to be best suited to the bilingual case as only that model addresses and distinguishes between two levels in cross-language perception: the phonetic level of sensitivity to gradient, physical details and the abstract phonological level of perceiving meaningful distinctions.

The original PAM framework addresses nonnative speech perception by naive adults. But whereas monolingual listeners assimilate unfamiliar nonnative phones to L1 categories, bilingual listeners face a different challenge, as their perceptual system must accommodate to two known languages (Grosjean, 2008, Romaine, 1989). When perceiving speech in a foreign language, the naive monolingual listener often (but not always) perceptually assimilates nonnative phones to existing native language categories. Based on the possible assimilation patterns to the L1, PAM predicts the success with which naive listeners will discriminate nonnative contrasts. According to PAM predictions, two nonnative phones perceived as exemplars of two different native categories will result in two-category (TC) assimilation, and discrimination will be excellent. When one nonnative phone is perceived as speech, but not as reliably belonging to any single native category, and another is assimilated to a native category, that nonnative contrast will form an uncategorized–categorized pair (UC). Two nonnative phones perceived as equally good or poor exemplars of a native category will result in single category (SC) assimilation, and discrimination will be poor. Category goodness (CG) assimilation will result when two nonnative phones are assimilated to the same native category but one is perceived as a better exemplar of that category than the other. The size of this category-goodness difference will result in discrimination ranging from good to very good (for descriptions of all possible assimilation types see Best, 1995).

PAM-L2 extends the original PAM to explain effects of L2 perceptual learning and predict the success with which learners will discern L2 contrasts depending on how they are assimilated to the L1 on the basis of both phonetic and phonological similarities, and predicts on this basis which new L2 phonological categories will be established. Phonetic categories describe clusters of individual phones, which differ from each other in a gradient fashion. Phonological categories are comprised of one or more phonetic category and specify the linguistic function of phonetic segments within the lexicon of a language, that is, which segments are distinctive and which are not. Bilinguals may thus perceive phonetic differences between L1 and L2 segments (e.g., Greek versus English realizations of /p/) while simultaneously recognizing properties that indicate a shared (or ‘common’) phonological contrast (e.g., /b/ vs. /p/). That is, a common L1/L2 phonological category can include two distinct regions of phonetic space for each language, and those language-specific phonetic categories are free to evolve without necessarily influencing one another.

From a PAM-L2 perspective, language users never stop learning, and yet, there is a qualitative difference between individuals who are actively acquiring an L2 versus stable bilinguals (early state vs. end state). PAM-L2 focuses on learners who are actively acquiring their L2, and identifies factors that are important in the formation of L2 speech categories (Best & Tyler, 2007). For instance, the contribution of age of L2 learning is modulated by the relative quantity and quality of input from native L2 speakers (Flege and Liu, 2001, Jia and Aaronson, 2003, Jia et al., 2006), length of residence in an L2-speaking environment, and relative L1:L2 usage (Flege, 1999, Flege, 2002, Flege and MacKay, 2004). Based on these assumptions, it follows that early bilinguals who have used their L2 so extensively that they have become L2-dominant should be the most likely to have developed L2 categories because they have strong L2 biases on all of these factors. Specifically, the type of L2-dominant bilingual we focus on here acquired both their L1 and L2 from native speakers, they were born and have lived the entirety of their lives in the country in which their L2 is spoken, and their L1:L2 usage ratio is heavily skewed toward the L2 as it is used in a much wider variety of social situations. Thus, this type of bilingualism could be highly informative to our understanding of L2 speech category formation. However, little to no research has focused on this type of bilingual. Most bilingual and L2-learner speech perception research has instead focused on other questions suited to the populations that have much more often been studied: L1-dominant bilinguals, and bilinguals living in more balanced bilingual environments where both languages are spoken widely and on a daily basis.

While researchers have interpreted bilinguals' performance in speech perception tasks as evidence for their stable phonological organization for the two languages, bilingual performance on such tasks may instead depend on what language-specific information a bilingual attends to. This may be especially important to consider in light of possible differences between L2-dominant bilinguals and L1-dominant or bilingual-environment bilinguals. The Language Mode framework (Grosjean, 1998, Grosjean, 2001, Grosjean, 2008) posits that, depending on the communicative situation, just one of a bilingual's languages may be activated (called monolingual mode, although the other language is never completely deactivated), or both languages may be activated (bilingual mode), thereby directing attention to language-appropriate information. As we will be comparing (L2-dominant) bilinguals with monolinguals of each language, we instead refer to monolingual mode as unilingual mode to avoid confusion. These language modes are thought to influence all aspects of language processing (Grosjean, 1998); however, research has barely touched on the perceptual effects of language mode in bilinguals. Although clear language mode effects have been observed for speech production (Antoniou et al., 2010, Antoniou et al., 2011, Hazan and Boulakia, 1993, Magloire and Green, 1999), language mode effects on speech perception are less clear.

Despite the lack of a theory of segmental speech perception that explicitly considers the contribution of the situational language context in bilinguals, the effect of language mode on perception has been investigated in a number of studies. Language mode has not always resulted in shifts in perception. In one study, Canadian-French–English bilinguals identified synthetic /p, t, k/ in separate French and English unilingual language modes (Caramazza, Yeni-Komshian, Zurif, & Carbone 1973). The bilinguals were unaffected by language mode and showed intermediate identification patterns relative to the monolingual French and English listeners, in both English and French language modes. Similarly, language mode failed to influence the identification of Spanish–English bilinguals' discrimination of /b/–/p/ (Williams, 1977). Their category boundaries were at an intermediate point relative to those of monolingual listeners of Spanish and English. These combined findings suggested that bilinguals differ from monolinguals in their perceptual sensitivity.

Yet, some other studies have demonstrated language mode effects on bilinguals' perceptual sensitivity, as evident in phoneme category shifts for both language modes. For example, Dutch–English bilinguals shifted their phonemic boundaries (by ∼5 ms) when identifying tokens from a 16-step voice onset time (VOT) /da/–/ta/ continuum in Dutch versus in English (Flege & Eefting, 1987). Similarly, English–Spanish early bilinguals identified a series of naturally produced syllables which varied in VOT from /ba/ to /pa/ (Elman, Diehl, & Buchwald, 1977). The bilinguals perceived a reliably greater number of the test items as /ba/ in the English than in the Spanish unilingual mode. Importantly for the present study, the magnitude of this perceptual switching effect was more pronounced for bilinguals who were more proficient in the L2. Consistent with these observations, Hazan and Boulakia (1993) reported that ‘stronger’ French–English bilinguals showed phoneme boundary shifts due to language mode (∼5 ms). Therefore, despite their L2 proficiency later in life, bilinguals showed identification patterns that reflected their dominant language, as defined by those researchers. However, we must note that in that study ‘language dominance’ was equated with ‘strength of bilingualism’, and was determined from each bilingual's language of exposure during their early years. Thus, it did not necessarily represent their dominant language at the time of testing.

In a more recent study, Spanish–English early bilinguals (who started learning their L2 between 3 and 12 years of age) categorized stimuli from a /ɡa/–/ka/ continuum in separate language modes (Garcia-Sierra, Diehl, & Champlin, 2009). The bilinguals' category boundaries shifted toward the voicing lead (negative VOT) end of the continuum in Spanish mode and toward lag VOT values of English in English mode. Despite the small shift in boundary with the two language modes, the bilinguals nonetheless showed an English-like phonemic boundary overall, and this was observed even though 12 of the 15 bilinguals had acquired Spanish as their L1. All had reported that they were more proficient at speaking, reading, writing and understanding in English (6.5 out of 7) than in Spanish (5.6). All were students at the University of Texas and reported that they used English (60%) more than Spanish (40%) on a daily basis, with their teachers, classmates and friends. Therefore, it is likely that the bilinguals were L2 dominant,² and this was reflected in their perceptual performance.

This review of the bilingual perception literature demonstrates that language mode influences bilinguals' perception, and language dominance appears to be an important factor that mediates whether language mode effects will emerge. However, none of these perceptual studies explicitly controlled for the environment in which the bilinguals acquired their two languages, or their language dominance at the time of testing. Therefore, in order to clearly understand the influence of language dominance and language mode on bilingual perception, it is necessary to use strict selection criteria and test the perception abilities of bilinguals who are clearly L2 dominant, and to do so in L2 versus L1 language modes. If both language mode and language dominance affect bilinguals' speech perception, theories of L2 segmental perception (e.g., PAM-L2) will need to be extended to account for these factors.