The Influence of Language Experience on Speech Perception: Heritage Spanish Speaker Perception of Contrastive and Allophonic Consonants

Abstract

It is well known that a listener’s native phonological background has an impact on how speech sounds are perceived. Native speakers can distinguish sounds that serve a contrastive function in their language better than sounds that are not contrastive. However, the role of allophony in speech perception is understudied, especially among heritage speakers. This paper highlights a study that directly tests the influence of the allophonic/phonemic distinction on perception by Spanish heritage speakers, comparing their results to those of late bilingual and monolingual speakers of Spanish and English in the US. Building on an earlier study, the unique contribution of this paper is a study of the perceptual pattern shown by heritage speakers of Spanish and a comparison of bilingual and monolingual speakers of English and Spanish. The participants completed a similarity rating task with stimuli containing VCV sequences with the intervocalic consonants [d], [ð], and [ɾ]. The heritage speakers, who are early sequential bilinguals of Spanish and English, showed a perceptual pattern that is more like monolingual Spanish listeners than monolingual English listeners, but still intermediate between the two monolingual groups. Specifically, they perceived [d]/[ɾ] like the L1 Spanish participants, treating them as very different sounds. They perceived the pair [d]/[ð], which is contrastive in English but allophonic in Spanish, like the L1 Spanish participants, as fairly similar sounds. Finally, heritage speakers perceived [ɾ]/[ð], contrastive in both languages, as very different sounds, identical to all other participant groups. The results underscore both the importance of surface oppositions, suggesting the need to reconsider the traditional definition of contrast, as well as the importance of considering level and age of exposure to the second language when studying the perception of sounds by bilingual speakers.

1. Introduction

Speech perception is a fundamental aspect of language processing, influencing communication and linguistic development. Understanding how heritage, bilingual, and monolingual speakers of Spanish perceive consonant sounds can provide insights into the interplay between language experience and phonological processing. This study hopes to add to the body of literature on the perception of contrastive and allophonic sounds, with a focus on the bilingual listener’s linguistic experience and age of acquisition of a second language.

Listeners of one language do not always process sounds in the same way that listeners of another language might. Speech perception is dependent on the language(s) spoken by the listener and the degree and age of their exposure to the language. The phonological inventory of the language spoken affects the degree of perception, with contrast playing a significant role. Sounds that are not contrastive in the listener’s native language (L1) will be more difficult, if not impossible, to perceive compared to contrastive sounds in the listener’s L1. For example, Japanese listeners have difficulty distinguishing the English liquids [l]/[ɹ] because these sounds are not contrastive in Japanese (MacKain et al., 1981; Flege et al., 1996; Bradlow et al., 1999). In addition, monolingual Spanish speakers perceive English vowel pairs that are non-contrastive in Spanish, such as [i]/[ɪ], to be more similar than do monolingual English speakers (Boomershine, 2013). Additionally, studies in second language learning have found that listeners are more adept at perceiving sounds of their native language than those of a second language acquired later in life, e.g., (Polka & Werker, 1994; Strange, 1995; Dupoux et al., 1997; Best et al., 1988; Francis & Nusbaum, 2002). The study that we report here expands on this research in L2 speech perception in a study of the perceptual consequences of allophony for bilingual and heritage language speakers.

2. Background

2.1. Speech Perception

When a person starts to learn a new language (their L2 if they currently speak just one other language—the L1) there are a few different phonetic scenarios that they may face. It has been found that the learning outcome (when it comes to developing a native-like ability to perceive the words of the L2) is determined to a substantial degree by the type of challenge that the phonetic scenario presents (Flege, 1987; Best, 1994; Kuhl, 1991; Kuhl et al., 2008; Escudero, 2005). The easiest case is when a sound in the L2 matches almost exactly a sound that has been learned in the L1. Interestingly, this sort of close match may lead to the longest lasting speech production and perception accentedness—deviation of the L2 speaker from the L1 norms for the language (Best, 1994) but generally the deviation from nativeness is relatively small.

Another scenario is when a sound in the L2 can be associated with a somewhat similar sound in the learner’s L1, but only somewhat. The phonetic deviation between the languages can be quite detectable to the phonetician but escape the notice of the learner; for example when a speaker of English is learning Spanish and must produce the voicing contrasts on stops as [±voice] during the stop closure, versus the English use of [±aspiration] after the stop closure. In this case, the identity of the corresponding L1 and L2 sounds in the mind of the listener (see Kuhl et al., 2008) leads to what Flege (1987, 1995) called and “Equivalence Classification” which in turn is associated with quite substantial deviation from nativeness in the L2 speaker’s pronunciation and perception.

A third type of scenario that the L2 learner may face is one in which there is no identifiable “similar” sound in the L1. This occurs, for example when an English speaker learns a click language like Zulu (Best, 1994). The clicks are not very similar to any English sound, and thus are unlikely to be confused with any existing L1 phoneme for the English speaking learner. In this non-identity scenario it is often the case that the learner can develop quite good perceptual ability in identifying the new sound (Best, 1994).

A fourth scenario that has been a focus of research in the L2 acquisition literature is the case where the languages have mismatched phonological categories (Escudero, 2005). This scenario is similar to the non-identity scenario; however the new sound is similar enough to an existing sound in the learner’s L1 that they may perceptually merge two contrasting L2 sounds. For example, the English contrast between /l/ and /r/ is difficult for Japanese speakers to master. The situation is that Japanese has a sound that is similar to both /l/ and /r/ and so initially at least the learner does not notice that in English these two are different (Lively et al., 1994). It is like an equivalence classification times two—two sounds in L2 are both being equated with a single sound in L1. Chang (2014) describes this scenario as L1 and L2 sounds not being in a ‘one-to-one relationship’ so that the listener must ‘distinguish more categories in the L2 than exist in the L1’. Escudero (2005) also treats this category mismatch scenario as a phonological structure mismatch—where the phonology of L2 divides up a phonetic region into more categories than is done in the phonology of L1. In her theory, the solution to this problem requires a lexically-driven tuning of one’s phonetic expectations—with the key element being that lexical contrast drives the development of a new phoneme. Another example of an L2 acquisition scenario that requires phonological recategorization occurs when a Spanish speaker learns English vowels. The region of vowel space that is occupied by the phoneme [i] in Spanish, must be divided between [i] and [ɪ] in English. A new phonological contrast has to be learned—phonetic space has to be more finely divided.

A fifth scenario (not so thoroughly considered in the literature) which Boomershine et al. (2008) studied in monolingual Spanish and English listeners, and to which we return now with Spanish/English bilinguals, is different from all of these. In this scenario of cross-linguistic differences in speech perception, the sounds that are presented to listeners are present in both languages, but they have different allophonic and contrastive relationships to each other in the languages’ phonologies. For example, in one language, phones [a] and [b] are allophones of phoneme /A/ but in the other language, [a] is a realization of phoneme /A/ while [b] is a realization of phoneme /B/. So, in this scenario all of the listeners are familiar with the phones [a] and [b], but for one group of listeners the phones are contrastive, while for the other group of listeners they are non-contrastive allophones of the same phoneme. Boomershine et al. (2008) found a perceptual effect of language-specific allophony that differed by the language of the listener such that allophones of a single phoneme are more perceptually similar to each other than the same sounds are when they are constrastive phones. We turn to this fifth scenario on the role of allophony in the next section.

2.2. Perceptual Consequences of Allophony

Research has shown that in addition to the presence or absence of contrast between two sounds in a listener’s phonology (for example, that Spanish has the high front vowel /i/ but not the lax vowel /ɪ/) the allophonic relationship between sounds (for example, that Spanish /r/ is realized as either a trill or a tap depending on context) also affects how those sounds are perceived (Babel & Johnson, 2010; Boomershine et al., 2008; Chappell, 2017; Harnsberger, 2001; Huang & Johnson, 2010; Johnson, 2004; Johnson & Babel, 2010). Huang and Johnson (2010) studied the perception of Mandarin tones by native and nonnative speakers. In their study, native Mandarin listeners demonstrated sensitivity to tone contours while the English listeners attended to pitch levels during the task. The Mandarin speakers rated tones that had been contrastively neutralized due to tone sandhi rules in Mandarin as being more similar than did the English listeners—meaning the Mandarin speakers tapped into the allophonic relationships found in their tone system when completing the perceptual rating task. Harnsberger (2001) studied the perception of contrastive and allophonic nasals by Malayalam listeners, finding a near merger of allophonically-related dental and alveolar nasal consonants. These coronal nasals are in complementary distribution in the language, with the dental occurring morpheme-initially and the alveolar occurring both morpheme-finally and intervocalically. Contrastive nasals such as bilabial [m] versus velar [ŋ], on the other hand, showed greater perceptual separation in Harnsberger’s study.

Boomershine et al. (2008) found that participants in their speeded discrimination and similarity rating tasks tapped into their L1 phonologies when rating the contrastive and allophonic pairs of [d]/[ɾ], [ɾ]/[ð], and [d]/[ð]. Spanish and English listeners were asked to rate these pairs using a five-point similarity scale. In Spanish, [d] and [ð] are allophones and [ɾ] is its own phoneme while in English, [d] and [ɾ] are allophones and [ð] is its own phoneme. The Spanish listeners rated [d] and [ð] as being similar while the most similar pair for the English listeners was [d] and [ɾ]. In other words, the listeners rated the pairs that are allophones in their L1 as being more similar than the other pairs. The same results were found in the speeded discrimination task with the same groups of participants, with those pairs that were allophones in Spanish ([d]/[ð]) being responded to more slowly by Spanish participants when compared to their reaction times to pairs containing an allophone and a phoneme ([d]/[ɾ], [ð]/[ɾ]). The pairs that contained allophones in English ([d]/[ɾ]) were responded to more slowly by English listeners when compared to their reaction times to pairs containing one allophone and one phoneme ([d]/[ð], [ɾ]/[ð]).

Another recent study that has investigated the role of allophony on speech perception is Babel and Johnson’s (2010) study of fricative perception by Dutch and English listeners. Their study included voiceless fricatives which were either contrastive in both languages, contrastive in only one of the languages, or allophones in one of the languages. For now, we will focus on the latter category and review how their participants perceived [s] and [ʃ]. In English, these phones are contrastive, given minimal pairs such as ‘sip’ and ‘ship’, respectively. In Dutch, on the other hand, these sounds are not used contrastively. In the perceptual similarity task, Dutch listeners rated [s] and [ʃ] as being significantly more similar compared to the ratings of the English listeners.

In her 2017 study, Chappell investigated how well monolingual speakers of Costa Rican Spanish are able to perceive differences in allophonic and contrastive pairs in Spanish. The participants were asked to complete similarity rating and AX discrimination tasks, evaluating word pairs that were identical or differed only in one phoneme or allophone. She found that the listeners perceived phonemic contrasts best, followed by allophonic pairs and then identical sounds. That said, not all allophonic pairs were perceived equally, with the [s]~[z] pair being the most difficult to distinguish compared to other allophonic pairs. Chappell contends that those allophonic pairs that encode linguistic meaning are more salient than those that do not.

2.3. Bilingualism and Speech Perception

Several studies have investigated the perception of sounds by early bilinguals, including the study by Bosch et al. (2000). They found that early bilinguals of Catalan and Spanish exhibited distinct vowel perception patterns for each language, suggesting that bilingual experience influences vowel categorization. They assert that balanced bilingual Catalan and Spanish speakers maintain separate vowel systems, which in turn affects their perceptual processing of vowels.

Other studies have investigated the perception of vowels by Catalan and Spanish bilinguals. In their 2005 study, Navarra et al. (2005) used an implicit method for measuring the L1 effects on the perception of L2 sounds. They asked Catalan-Spanish simultaneous bilinguals who either grew up in Spanish-speaking homes or Catalan-speaking homes to categorize the first syllable of bisyllabic stimuli, with the only difference in the stimulus items being the vowel in the second syllable—it could contain a Catalan contrastive variation (/ε/–/e/) or no variation. Catalan dominants responded more slowly in lists where the 2nd syllable could vary from trial to trial, suggesting an indirect effect of the /ε/–/e/ discrimination. Spanish dominants, however, did not suffer this interference, performing indistinguishably from Spanish monolinguals. These findings seem to suggest that even simultaneous bilinguals’ perceptual systems are affected by their L1 phonology.

A similar study was conducted by Mora and Nadeu (2012) to study the effects of an L2 (Spanish) on the perception of L1 (Catalan) contrastive sounds. In their study, they asked L1-Catalan, L2-Spanish speakers who lived in a Catalan-dominant community and L1-Catalan, L2-Spanish speakers living in a Spanish-dominant community to complete an identification task and an AXB discrimination task. They found that both groups performed at near ceiling levels in the tasks, but the reaction time for those participants who have more exposure to Spanish than Catalan was slower than for the Catalan-dominant speakers. The findings suggest that extensive L2 use/exposure to Spanish and Spanish-accented Catalan in a bilingual language contact setting may modify Catalan natives’ phonetic category /ε/.

In their study, Ning et al. (2022) studied the effect of age of L2 acquisition on the perceptual processing of lexical tones in Cantonese and Urdu speakers. Three groups of subjects, Cantonese monolinguals, Cantonese/Urdu simultaneous bilinguals, and late L1 Urdu/L2 Cantonese bilinguals, living in Hong Kong participated in a four-condition ABX task. When there were no conflicts, the simultaneous bilinguals were able to process Cantonese tones like Cantonese L1 speakers. However, when conflicts were introduced, the simultaneous bilinguals had significantly slower reaction times and produced more errors when compared to native Cantonese speakers on the same tasks. Even with years of schooling and exposure to Cantonese in and outside of the home, simultaneous bilinguals were still unable to process Cantonese tones like native Cantonese speakers.

While there has been some research on how bilinguals who live in bilingual settings where both languages are used extensively in the speech community, there has been little perception research on heritage speakers of Spanish living in the United States. Heritage speakers of Spanish differ significantly from early Catalan/Spanish bilinguals in Spain or simultaneous Cantonese/Urdu bilinguals in Hong Kong, for example, in that heritage speakers of Spanish grow up hearing and speaking Spanish in the home from birth, and then, in the case of the US, learn English upon entering school. In the US, they generally live in English-dominant communities with little to no access to schooling or literacy in Spanish. The domains available for Spanish exposure are limited to the home and culturally specific sites in the community, including church, social outings/festivities, and cultural celebrations.

2.4. Heritage Speakers

The current study repeats the Boomershine et al. (2008) rating task with three new groups of listeners—bilinguals dominant in English or Spanish, and heritage speakers of Spanish in the US who grew up hearing speaking Spanish in the home, and then learned English upon entering school. The bilingual speakers in Boomershine et al. (2008) began learning English as their L2 as teenagers (Table 2). In this study we were interested in learning whether heritage speakers would pattern more like native speakers of English, given that they learned English early, had no formal education in Spanish, and that the experiment was conducted in English at an institution where English is the main mode of communication. To preview the results, bilingual speakers patterned with monolingual speakers of their L1, and heritage speakers of Spanish patterned between monolingual speakers of English and Spanish to some extent, but their pattern of responses was much more like that of Spanish monolinguals.

Given the research that has found a language effect for the perception of contrastive and allophonic sounds, with babies as young as 6 months able to perceive sounds in their L1 (Best, 1994; Kuhl et al., 1992), the early linguistic experiences of heritage speakers of Spanish are unique and given their early and extensive experience with a second language, it is not at all certain that they would pattern with other bilingual speakers who maintain active use of both languages. The limited number of studies on the speech perception of heritage speakers of Spanish (HSS) has found that HSS perceive certain sounds like Spanish speakers and other sounds like English speakers (Kim, 2011; Boomershine, 2013). Studies have shown that heritage speakers often exhibit differences in consonant perception compared to monolingual speakers. For example, Kim (2011) conducted a production and perception study of English and Spanish stop consonants with heritage Spanish speakers who were English dominant. In the production study, she found that the heritage Spanish speakers had VOTs that were nearly identical to native English speakers and significantly different from those of native Spanish speakers. In her perception study, Kim had participants listen to carrier phrases containing natural and cross-spliced stimuli and then identify the word they heard. For the natural stimuli, the HSS perceived the English stops like English speakers and the Spanish stops like Spanish speakers—in other words, like two monolinguals. However, when presented with the cross-spliced stimuli, the heritage Spanish speakers patterned like native Spanish speakers in that they judged the cross-spliced stimuli produced with pre-voicing to be voiced, whereas the native English speakers relied more on cues from the vowel portion of the stimuli when identifying the stimulus item as voiced or voiceless.

Research with HSS perception of English vowels, however, found differing results. Boomershine (2013) asked heritage, monolingual, and (late) bilingual speakers of Spanish and English to listen to pairs of English front vowels and rate them on perceptual similarity. In this study, the HSS reported being exposed to Spanish at birth and to English on average at age 2. The bilingual participants reported being exposed to their L2 on average at age 11. She found several significant differences between the heritage and monolingual groups, but no significant differences between the HSS and the L1 Spanish bilinguals. In addition, the HSS patterned similarly to the L1 English bilinguals for all vowel pairs except ‘bait’/’bit’, where the HSS found this pair to be more similar than did the L1 English bilinguals.

Other studies suggest that heritage speakers can maintain robust perception skills of consonants as well, finding that early exposure to the heritage language has long-lasting effects on heritage speakers’ ability to distinguish heritage language phonological contrasts. When compared to Hispanic immigrants who had lived in the US for a long period of time and to Spanish native controls who had recently moved to the US, Spanish heritage speakers performed similarly to the control group when distinguishing the Spanish stop voicing contrast (Mazzaro et al., 2016). Another study comparing the perception of consonant perception by heritage Spanish speakers and Spanish native controls found that when listening to stimuli with contrasting acoustic information in the consonant portion and the vowel portion (/b/ from /be/ + /e/ from /pe/), both the heritage speakers and the native controls attended to the consonant portion more than the vowel portion (Kim, 2011).

Research on consonant perception among heritage, bilingual, and monolingual speakers of Spanish highlights the complex interplay between language experience and phonological processing. Given the limited perception research on HSS, the current paper has as its goal to better understand the perceptual processing system of heritage speakers, especially with respect to the perception of contrastive and allophonic sounds, as compared to late bilinguals and monolingual speakers of Spanish and English. As native or near-native speakers of both Spanish and English, HSS are in the unique position to shed light on how the bilingual brain processes sounds that have different phonological relationships in the two languages spoken.

Given what is known about how heritage speakers differ from late bilingual speakers in speech perception, the researchers pose the following two hypotheses:

H1: 

Heritage speakers of Spanish will perceive contrastive sounds in their L1, Spanish, with a similar pattern of their late L1 Spanish bilinguals due to their early exposure to Spanish.

H2: 

Heritage speakers of Spanish will exhibit some influence from their L2, English, when perceiving pairs that are allophonic in their L2, as they have more continuous and consistent exposure to English as college students living in a majority-English community.

To test these hypotheses, the current study will partially replicate the Boomershine et al. (2008) research on the perception of the phones [d], [ð] and [ɾ], this time testing the perception of heritage speakers of Spanish as compared to their monolingual and bilingual speakers of English and Spanish participants in a similarity rating task.

3. Materials and Methods

The three consonants [d], [ð], and [ɾ] exist in both the Spanish and English phonetic inventories, but have different roles in their phonological systems. In English, [d] and [ɾ] are allophones of the same phoneme, while [ð] and [d] are contrastive. On the other hand, [d] and [ð] are allophones of the same phoneme in Spanish, and [d] and [ɾ] are phonemically contrastive. On the surface, however, [d] and [ɾ] generally do not appear in the same phonetic contexts and thus are not lexically contrastive in most varieties of Spanish, including Mexican Spanish, the variety spoken by the participants in the current study (Hualde, 2013). There is, though, a surface contrast between [ð] and [ɾ] for all varieties of Spanish ([ka.ða] ‘each’ vs. [ka.ɾa] ‘face’). These relationships are illustrated in

Table 1. Relationships and examples of [d], [ð] and [ɾ] in Spanish and English.

Sound Pairs	d/ɾ		d/ð		ɾ/ð
Language	Spanish	English	Spanish	English	Spanish	English
Underlying contrast	(yes)	no	no	yes	(no)	no
Surface contrast	no	no	no	yes	yes	yes
Allophony	no	yes	yes	no	no	no
Examples	N/A	Fred Freddy	en diez ‘in ten’ a diez ‘to ten’	dough though	cara ‘face’ cada ‘each’	ladder lather

.

In order to explore the role that age and type of exposure play on the perception of contrastive and allophonic sounds in Spanish and English, the perceptual similarity rating task used by Boomershine et al. (2008) was utilized in the current study, with a new panel of listeners, including monolingual speakers of English and Spanish, bilingual speakers of English and Spanish whose L1 was either Spanish or English, and heritage speakers of Spanish. The same audio recordings that were used in Boomershine et al.’s (2008) study were used in the current study and consisted of two tokens of each of the following VCV sequences: [ada], [aða], [aɾa], [idi], [iði], [iɾi], [udu], [uðu], and [uɾu]. Boomershine et al. (2008) recorded multiple tokens of these which were produced by two native speakers of Greek, one male and one female, using a head-mounted microphone in a soundproof booth. Greek speakers were chosen because all three of the test phones, [d], [ð], and [ɾ], are contrastive in Greek and are produced naturally in intervocalic position. The speakers attempted to produce equal stress on the first and second syllables. In order to control the amplitude across tokens and speakers, the peak amplitude was equated for each of the tokens. The two best recordings for each VCV sequence were used as stimuli in their study and in this one.

Table 2. Demographic information for participants. * Participants self-rated their L1 and L2 proficiency using a scale from 1–7.

Native Language	L2 Language	M	F	Total	Avg Age	Average Age of L2 Exposure	* L1 Prof. Self-Rating	* L2 Prof. Self-Rating
Native English	Little to no Spanish	1	11	12	20.08	19	7	2.9
	Advanced Spanish	6	2	8	26.5	13.2	7	6.19
Native Spanish	Little to no English	6	4	10	24.4	21	7	2.1
	Advanced English	2	5	7	32	12.5	7	5.28
Heritage Spanish	Native/Fluent	4	9	13	19.3	4.3	6	6.97

depicts the demographic information for the 50 participants in the current study. It should be noted that the heritage speakers of Spanish are considered native speakers of Spanish but were given their own category in order to distinguish their age and amount of exposure to English. The heritage speakers of Spanish were exposed to Spanish at birth and reported being exposed to English, on average, at the age of 4.3 years, usually when they started attending pre-kindergarten. This is considerably earlier than that of the late bilinguals, with an average age of exposure to English for the native Spanish, advanced English participants of 12.5 years and an average age of exposure to Spanish for the native English, advanced Spanish participants of 13.2 years. The participants labeled in this study as monolingual speakers of either Spanish or English reported having no or very little exposure to the other language, with an average self-rating of English proficiency for the monolingual Spanish speakers of 2.1 and an average self-rating of Spanish proficiency for the monolingual English speakers of 2.9 using a scale of 1–7, with 7 being ‘native proficiency’. The monolingual speakers who were exposed to an L2 reported receiving this exposure as young adults in academic settings.

The participants in this study were affiliated with the University of North Carolina Wilmington as students, staff, or acquaintances of students or staff. Thirteen heritage speakers of Spanish participated in the current study. They were all students at the University of North Carolina Wilmington and were enrolled in the advanced Spanish for Bilingual Speakers course. They received course credit for participating in this study. The participants completed a post-experiment demographic questionnaire, and only those participants that were exposed to Spanish in the home since birth, had parents from Mexico (and thus spoke Mexican Spanish), and were exposed to English by age 5 were included in the study. The L1 Spanish bilingual speakers were all native speakers of Mexican Spanish that reported that their primary exposure to Spanish was with other speakers of Mexican Spanish. Speakers of other varieties of Spanish were excluded from this study in order to control for variation in the realization of intervocalic stops in some varieties of Spanish. All L1 English participants in this study were speakers of American English and all L2 speakers of English reported having lived in the US as their only English-speaking country of residence. None of the participants reported any hearing, speech, or language disorders.

Participants were seated in front of a PC with an attached five-button response box and were presented with the stimuli through headphones. Participants heard pairs of physically different stimuli separated by one second of silence, such as [ada] < 1 s silence> [aða], and responded with a rating score from 1 (very similar) to 5 (very different). The pairs were presented in a different random order for each participant, using E-Prime software (v. 1.1; Psychological Software Tools, Pittsburgh, PA, USA). The talker and vowel context were the same for every pair so that the only difference in each pair was the consonant. The stimuli presented in each pair were always physically different tokens, even when they were both examples of a single sound (e.g., [ada]…[ada]). The participants were given four practice trials, and then the opportunity to ask questions before proceeding to the four test blocks (360 test trials total). Participants were able to take a short break between test blocks if needed, and the average participation duration was less than one hour. They received no feedback in this experiment.

Boomershine et al. (2008) also presented pairs for speeded discrimination (‘same’ vs. ‘different’ responses) and measured perceived similarity as a function of the time it takes to press the ‘different’ button when the pair is physically different. We chose in this study to only use the rating task because it is statistically less noisy and thus more reliable, and because it taps a level of perception that is more sensitive to linguistic knowledge, as opposed to being based on purely auditory discrimination.

In order to account for some participants not using the response option endpoints, the rating scores for each speaker were normalized to compensate for differences in use of the 5-point scale. The scores were normalized using a standard z-score transformation, such that each participant’s scores were centered around 0, with scores above zero indicating “more different” and scores below zero indicating “more similar”. This normalization allows the participants’ similarity ratings to be compared with each other and with those of the previous study (Boomershine et al., 2008).

4. Results

This study compared the normalized similarity ratings of three consonant pairs across five groups of participants with varying levels of experience and age of exposure to Spanish and English. The results of this similarity task were analyzed across sound pairs. We will first look at the results by the first language of each participant, followed by the results based on second language exposure of each participant group.

4.1. L1 Effects on Perception

To better understand the role of L1 on the perceptual rating of contrastive and allophonic sounds, the participants were first divided into two groups based on their first language (the language they first acquired). In this study, 30 of the 50 total participants acquired Spanish as their first language (the monolingual Spanish, L1 Spanish bilingual, and heritage Spanish speakers), with the remaining 20 participants being L1 speakers of English (monolingual and L1 English bilingual speakers). The normalized results by the first language of the participants (leaving out the heritage speakers) are shown in Figure 1.

A repeated measures ANOVA revealed a significant main effect for consonant pair, F(1,65) = 4.835, p < 0.05, η² = 0.092, as well as a significant interaction of first language on the similarity rating of consonant pairs, F(1,65) = 17.854, p < 0.001, η² = 0.271.

Overall, the perception results patterned around the participants’ first language, with contrastive sounds in each L1 being perceived as more different for that language group. To determine if these similarity ratings were significantly different based on the first language of the participants, independent samples t-tests were run. In both Spanish and English, [ɾ]/[ð] has a surface contrast (as in ‘cara’/‘cada’ and ‘ladder’/‘lather’, respectively), leading both participant groups to rate those sounds as being different (p = 0.761). L1 (English vs. Spanish) did not produce a significant difference in similarity rating as both L1 English and L1 Spanish participants rated [ɾ]/[ð] as being different. The pair [d]/[ð], with a surface contrast in English (as in ‘dough’/‘though’) and an allophonic relationship in Spanish (as in ‘en dos’/‘a dos’), was found to be significantly more different by the L1 English participants than by the L1 Spanish participants (p < 0.001). While the pair [d]/[ɾ] does not have a surface contrast in Spanish or English, it does have an underlying contrast in Spanish and is allophonic in English. Accordingly, the L1 Spanish participants found that pair to be significantly more different compared to the L1 English participants (p < 0.001).

4.2. L2 Exposure Effects on Perception

Knowing that a person’s first language affects their rating of these consonant pairs, the participants were subdivided into five groups based on the level of experience in their L2 and their age of exposure to that L2. The five participant groups are monolingual English, monolingual Spanish, L1 Spanish bilingual (late sequential bilingual), L1 English bilingual (late sequential bilingual), and heritage Spanish (early sequential bilingual) speakers. The normalized similarity ratings of the three consonant pairs across five groups of participants with varying levels of experience and age of exposure to Spanish and English are shown in Figure 2. To determine the effects of L2 exposure on the perception of these consonant pairs, a repeated measures ANOVA was conducted to compare the means of the five groups, revealing a significant effect of participant group on the similarity rating of consonant pairs, F(1,62) = 6.005, p < 0.05, η² = 0.109.

In order to determine if the participants’ language experience had a significant effect on their perception of these pairs of sounds, a Tukey’s HSD post hoc test was conducted. For the pair [ɾ]/[ð], the results showed that there was no significant difference by participant language group, with all participant groups rating that pair as being different (p = 0.901). This is expected since this pair is contrastive in both English and Spanish.

For the pair [d]/[ð], the Tukey’s HSD post hoc test showed a significant difference between monolingual English participants and heritage Spanish participants (p < 0.05), monolingual English participants and L1 Spanish bilingual participants (p < 0.05), and monolingual English participants and monolingual Spanish speakers (p < 0.01). There was also a significant difference between the L1 English bilinguals and the monolingual Spanish participants (p < 0.05). This pair is contrastive in English but allophonic in Spanish.

Tukey’s HSD post hoc test results for the pair [d]/[ɾ] revealed a significant difference between the monolingual English speakers and L1 Spanish bilinguals (p < 0.05) and the monolingual English speakers and the heritage Spanish speakers (p < 0.05). There was not a significant difference among the participants for whom Spanish is their first language, regardless of their English level, nor among the participants for whom English is their first language, also regardless of their Spanish level. This pair is allophonic in English and has an underlying contrast in Spanish, but no surface contrast in either language.

5. Discussion

Previous research has found that across languages, speakers of a language in which a particular pair of sounds is contrastive at a phonemic level perceive that pair as being more perceptually distinct when compared to speakers of a language in which the pair is not phonemically contrastive (Boomershine et al., 2008). The present study aimed to examine how monolingual and early and late bilingual speakers of Spanish and English perceive sounds whose phonological structures differ across the two languages, and specifically whether heritage speakers of Spanish in the US perceived sounds using their first language (Spanish) or their dominant (English) sound system.

5.1. L1 Effects on Perception

The first language of participants in this similarity rating task was found to significantly impact their perception of contrastive and allophonic sounds, with those pairs that are allophonic in the L1 being rated as more similar than sounds that are contrastive in that language. The results of this study align with those of the Boomershine et al. (2008) study in which the L1 speakers of English found the contrastive pair [d]/[ð] to be different while the L1 speakers of Spanish, where this pair is allophonic, found it to be similar, even for L1 speakers of Spanish who had very early exposure to English and considered themselves to be English-dominant. The opposite is true for the pair [d]/[ɾ], which is allophonic in English, resulting in L1 speakers of English finding it to be similar when compared to the L1 speakers of Spanish.

The findings of these studies add to evidence that the presence or absence of a sound in a language’s inventory is not the only source of information when a speaker perceives that and other sounds. The phonological relationship of sounds within a language is an integral piece to the perception puzzle. Sounds that are contrastive, either at the surface or underlying level, are perceived as more different by native speakers of that language than are sounds that are allophonic in that language (Kuhl, 2000; Pisoni, 1997; Johnson, 1997).

5.2. L2 Exposure Effects on Perception

When the participants are grouped not only by their first language but are further grouped by their age and type of exposure to their L2, we find that both early and late bilinguals tend to perceive sounds in a way that is similar to their L1 monolingual counterparts. The age and type of exposure to the L2 was clearly a factor in the similarity rating of the pair [d]/[ð], which is allophonic in Spanish but contrastive in English. The monolingual English speakers rated that pair as the most different, followed by the L1 English (late) bilinguals, with both groups rating it significantly more different than monolingual Spanish speakers. The monolingual Spanish speakers found the pair to be the most similar, followed closely by the L1 Spanish (late) bilinguals. The heritage speakers of Spanish, who were exposed to Spanish since birth but acquired English at an average age of 4.3 and were then surrounded by English throughout their lives including in school, in the community, etc, found this pair to be significantly more similar when compared to monolingual English speakers, as did the L1 Spanish (late) bilinguals and Spanish monolinguals. The fact that the heritage Spanish speakers patterned like monolingual and late bilingual speakers of Spanish, rather than monolingual English speakers provides evidence that even early sequential bilinguals tend to perceive sounds with their L1 phonology. However, it is notable that the heritage speaker ratings were also not reliably different from those of the L1 English bilinguals.

Turning to the pair [d]/[ɾ], which is allophonic in English but has an underlying contrast in Spanish, the monolingual English speakers found that pair to be more similar than did the Spanish L1 groups. Interestingly, the heritage speakers of Spanish again had a significantly different similarity rating for this pair as they did for [d]/[ð] when compared to monolingual English speakers, rating it as a more different sound than the English speakers did. This finding follows previous studies on L1 and L2 speech perception that found that even early bilinguals perceived L2 sounds by reference to perceptual expectations that are based on their L1 phonological categories (Archila-Suerte et al., 2012; Baigorri et al., 2019; Flege, 1991; Flege et al., 1999; Flege & MacKay, 2004; Levy, 2009; Levy & Strange, 2008; Morrison, 2002; Pallier et al., 1997). So, for example, Flege et al. (1997) found that L1 Korean listeners’ perception of the English contrast between the vowels of “beat” and “bit” (which is signaled by both vowel duration and vowel formant frequency differences) used vowel duration to distinguish the vowels and did not use the vowel formant differences. It should be noted that the Korean listeners in the study were not heritage speakers of Korean, but instead were L1 speakers who were exposed to English upon moving to the US as adults. Flege et al. attributed this pattern of results to the fact that Korean has long and short i: versus i, but no distinction between tense [i] and lax [I]. Returning to heritage speakers of Spanish for a moment, it is interesting to find that these participants patterned nearly identically to L1 Spanish (late) bilinguals, even though the heritage speakers in this study were exposed to English at a much younger age (4.3 years of age) than the L1 Spanish (late) bilinguals (12.5 years of age). In addition, the heritage speakers of Spanish’s exposure to both Spanish and English was different from that of their L1 Spanish late bilingual counterparts, as the heritage speakers were born and raised in a country where English is the majority language (the United States), whereas the L1 Spanish (late) bilinguals were born and raised in a Spanish-speaking country. The amount of exposure to schooling in English and Spanish differs across the two groups, with heritage speakers receiving schooling exclusively in English up until high school, where they may have taken a Spanish elective course, and the L1 Spanish (late) bilinguals attending school in Spanish, with some exposure to elective English classes at times. As seen in Table 2, the heritage speakers of Spanish self-rated as being dominant English speakers (6.97 out of 7) as compared to a score of 6 out of 7 for Spanish, whereas the L1 Spanish (late) bilinguals rated themselves as dominant Spanish speakers (7 out of 7) and less proficient in English (5.28 out of 7). Montrul’s (2008) study on the linguistic abilities of heritage speakers of Spanish suggests that heritage speakers may face difficulties in distinguishing phonemic contrasts in their L1 that native speakers can easily perceive. Similarly, Rao and Myers’ (2014) found that heritage Spanish speakers perceive vowel duration and prosodic contours differently from native speakers, often leading to non-native-like speech patterns. The results of the current study differ from those of Montrul and Rao and Myers, finding that heritage Spanish speakers’ perception was in fact very similar to that of their L1 Spanish (late) bilingual counterparts. More research needs to be done to determine how heritage speakers of Spanish perceive other allophonic vs. contrastive sounds in the languages they speak to determine if the pattern found in this study is due to L1 sound categories or if there are other factors at play.

The results reported here suggest a nuanced view of the critical period hypothesis for second language bilingualism. Two groups of L1 Spanish, L2 English bilinguals were observed to have responded largely in line with Spanish monolinguals: early sequential bilinguals (heritage speakers), and later L2 English learners. Though the heritage speakers showed greater similarity to the English monolinguals’ pattern of responses in the allophone similarity task, as might be expected by their early acquisition of English, it is noteworthy that their pattern of responses was strikingly similar to that of Spanish monolinguals. Heritage speakers learned English well within the critical time period for language acquisition as most authors would delimit it, yet their perception responses were very similar to native Spanish listeners: both monolinguals and bilingual listeners who acquired English later in life. These findings are similar to those reported by Kim (2016) in her study comparing the perception and production of Spanish lexical stress by heritage and L2 Spanish speakers, finding that heritage speakers did not differ statistically from native speakers in the perception of Spanish lexical stress, unlike their L2 Spanish counterparts who performed at chance. This, coupled with our current findings, suggests that very early linguistic experience has a lasting effect on speech perception.

5.3. Theoretical Implications

Of the theoretical approaches to L2 speech perception that we presented in Section 2.1, Kuhl et al.’s (2008) perceptual magnet theory is the most evocative of the allophonic similarity phenomenon, in that a linguistic category serves as a metaphorical “magnet” in that members of the category are more similar to each other than would be expected on purely psychophysical grounds. However, like the recategorization phenomenon that inspired Escudero’s (2005) phonological theory of L2 perception and production, allophonic similarity is a phonological effect.

What has been demonstrated in this paper (replicating and extending the findings of Boomershine et al., 2008) is that allophonic effects in perception are language-specific and may be tied to the amount and type of the listener’s linguistic experience. So, what kind of theory could account for experience-based language-specific allophonic perceptual similarity? Although there may be several alternatives, we will just mention one here: the exemplar resonance theory of Johnson (2006). This model envisions an activation loop between abstract and phonetic levels of representation, where the phonetic level is populated by phonetically detailed exemplar traces in memory. The resonance loop feeds activation from abstract categories back into the exemplar cloud which can result in co-activation of exemplars of different allophones of the same phoneme. This co-activation of the exemplars of allophones results then in increased similarity between the perceptual activation patterns generated by the allophones of a phoneme -even when they are not terribly phonetically similar—like [d] and [ð]. And this increased similarity in activation patterns will result, so the model predicts, in decreased perceptual separation between them. This model predicts the pattern of results found in this study, that allophones will be more perceptually similar to each other than would be predicted by their psychophysical representation alone, and that this pattern of perceptual similarity will be language-specific (driven by the language’s phonology) and will be sensitive to the listener’s specific linguistic experience (their clouds of exemplars).

However, the data here are challenging for the exemplar resonance theory because the ’abstraction’ layer in that model is lexical/semantic (Johnson, 2004), so in an experiment such as the one reported here we have to imagine that the non-word [aCa] stimuli activate specific lexical items. Future research developing exemplar resonance theory will need to develop a specific account for how non-words can activate lexical items and how activation in a bilingual linguistic system may be regulated. Particularly, it seems important to note that the L1 pattern of allophony-driven speech perception is evidently learned at an early stage of L1 acquisition, and persists despite the early and extensive exposure to L2 that is experienced by heritage speakers. This is consistent with the suggestion that “the speech perceptual system does not seem to be prone to modify initial phonemic categories.” (Sebastián-Gallés & Soto-Faraco, 1999, p. 112; see also Mack, 1989; Pallier et al., 1997).

5.4. Limitations and Future Directions

While this study begins to address the question of age of L2 exposure as a factor in the perception of contrastive and allophonic sounds, there are some limitations. Ideally, more participants from each of the L1/L2 groups would be included in a future study. Also, future studies could include simultaneous bilinguals, or speakers of Spanish and English that acquired both languages from birth, rather than only early and late sequential bilinguals as were included in this study. Additionally, it would be useful to see how heritage speakers of Spanish who had received academic exposure to Spanish and English from an early age, such as via a Dual Language/Immersion program, would perform in this task, especially based on the research already available concerning not just the age of acquisition, but also the availability of input (Montrul, 2016).

Additionally, the realization of the consonants in question varies by dialect, in both English and Spanish. In this study, only speakers of Mexican Spanish and American English were included to minimize dialectal differences in production and perception. However, it would be of interest to have speakers of British English, where intervocalic alveolar stops are not produced as taps, participate in this study, as it would for speakers of some Central American and highland varieties of Spanish where voiced stops often do not undergo lenition in expected environments, instead being realized as voiced stops (Carrasco et al., 2012).

Turning to the discussion on the role of the phonological relationship of L1 and L2 sounds in speech perception, more studies are needed in this area. For instance, a study that compares the perception and production of the sounds in the current study would be helpful in determining the heritage and late bilinguals’ proficiencies with the two languages generally, and the sounds at hand, in particular. It would also be helpful to have studies similar to this one but that look at the perception of other allophonic and contrastive pairs to determine if they pattern the same way as these sounds do. For example, how do heritage speakers of Spanish perceive nasals as compared to bilingual and monolingual speakers? This and other studies looking at the role of contrast and allophony in the perception of L1 and L2 sounds by heritage and late bilingual speakers will continue to shed light on how the bilingual brain processes sounds.