Literature Review

Most research on heritage language has focused on morphological and syntactic similarities and differences between bilingual and monolingual speakers (^{Rao & Ronquest, 2015}). Until recently, it was assumed that HSS' phonology matched that of monolinguals. A growing body of research, however, shows quantifiable differences between bilingual and monolingual speakers in various aspects of pronunciation (^{Amengual, 2012}; ^{Colantoni, Cuza, & Mazzaro, 2016}; ^{Knightly, Jun, & Au 2003}; ^{Rao, 2015}; ^{Robles-Puente, 2014}; among others). These differences seem to be largely the result of transfer from the dominant second language onto the heritage language. For example, ^{Ronquest (2013)} find that HSS Spanish in the US produce vowels in unstressed syllables that are shorter in duration than do monolingual speakers of Spanish. This is likely due to the influence of English phonology, since durational differences between stressed and unstressed syllables are greater in English than in Spanish (^{Delattre, 1969}). English and Spanish have differences in their consonant systems as well. English has longer voice onset timing (VOT) for voiceless stops (/p/, /t/, /k/) than Spanish (^{Lisker & Abramson, 1967}). ^{Au, Knightly, Jun, and Oh (2002)} find that HSS have monolingual-like VOT. However, in a study that controlled for cognates vs. non-cognates, ^{Amengual (2012)} finds that HSS produce longer VOT in Spanish words with corresponding English cognates in Spanish words without English cognates. His results show that there is some English influence on heritage speaker pronunciation of words that are shared between the two languages. English voiced stops (/b/, /d/, /g/) between vowels have more closure than do the corresponding sounds in Spanish. The results of studies conducted on heritage production of these sounds are mixed. ^{Au et al. (2002)} finds HSS to be like monolinguals in their pronunciation of all three stops. ^{Knightley et al. (2003)} find the same for /b/ and /d/, but /g/ displayed more closure. Rao (2015) finds no significant differences between monolingual and bilingual speakers for /d/ and /g/, but /b/ was more variable.

Despite these differences between monolingual and heritage pronunciation, segmental knowledge serves HSS well in the classroom. In a corpus study of HSS' spelling mistakes, ^{Beaudrie (2012)} finds that heritage writers do well with grapheme use; their main difficulty with spelling, accounting for over two thirds of the errors, is with the use of accent marks. While grapheme use depends upon an ability to map individual sounds to letters, being able to correctly place accent marks relies on suprasegmental knowledge. Heritage intuitions of suprasegmental phonological features are just beginning to be explored.

Spanish orthography has a set of rules governing the use of accents; a writer's ability to apply these rules depends upon their metalinguistic knowledge of syllabification and lexical stress. ^{Kim (2015)} performed a study on HSS's abilities to perceive and produce contrastive stress with minimal pairs like paso/pasó. She finds that HSS have monolingual-like perception, but their production of paso is inconsistent - they do not always produce the stressed /a/ with a longer duration than the /o/, unlike monolinguals. Heritage participants produce penultimate stress as ultimate (thus pronouncing paso like pasó), as do the second language learners in her study. Kim's finding shows that HSS do not have trouble with auditory comprehension - HSS can hear whether a speaker means 'I pass' (paso) or 'he/she passed' (pasó). However, HSS do not reliably produce this difference in their own speech. This difference in production is in line with ^{Beaudrie's (2017)} assertion that the difficulty of correctly placing accent marks has to do with HSS's trouble with producing lexical stress.

Another important aspect of suprasegmental knowledge that has implications for the use of accent marks is syllabification. The test items in ^{Kim's (2015)} study have a single vowel in the nucleus of each syllable. The syllabification of these words, therefore, follows the same patterns that would be expected in English. That is, paso, if it were an English word, would be syllabified as /pa.so/, just as in Spanish. Thus, syllabification intuitions are highly likely to be like those of monolinguals and not interfere with the study of lexical stress. Little is known, however, about how HSS perceive syllables in contexts where English and Spanish phonologies diverge. As mentioned in the introduction, English and Spanish differ in their treatment of two-vowel sequences. English tends to separate these sequences into two separate syllables, especially in the case of raising sonority (^{Davidson & Erker, 2014}; ^{McCarthy, 1993}). Spanish, conversely, has a preference to combine the vowels into a single syllable, with exceptions for sequences of two non-high vowels and for sequences with a stressed high vowel (^{Chitoran & Hualde, 2007}; Hualde & Prieto, 2002).

^{Zárate-Sández (2011)} looked at syllabification judgements for nine two-vowel sequences (<ie>, <ia>, <ai>, <io>, <ua>, <au>, <ue>, <eu>, <ei>) in undergraduate second language learners of Spanish ranging from beginning to advanced levels of study. All of the vowel sequences that Zárate-Sández includes in his study are diphthongs (according to standard descriptions of Spanish - see ^{Hualde, 2005}). His control group of monolingual speakers reports diphthongs 90% of the time, whereas all three groups of second language learners indicate diphthongs for only 56-59% of the stimuli. Second language learners are more accurate with two-vowel sequences that have analogous diphthongs in English, which are generally those that have falling sonority. For example, second language learners accurately combine the vowels in words with <ai>, like baile ('dance'); the diphthong in the English word right closely resembles the vowel sequence in Spanish. The accuracy rate for sequences with no corresponding diphthong in English, like <io> (35.5%), is much lower. Zárate-Sández does not explore the effect of stress on dipthongization rates; about half of his stimuli have stress on the diphthongized syllable, but he does not enter this into the analysis.

To date, few studies have examined the syllabification judgements of HSS. ^{Chappell (2018)} performed a perception study in which HSS listened to nonce words ending with familiar (-fono) and unfamiliar (-pine) suffixes. She finds that heritage listeners, unlike monolinguals, are less accurate with words with the unfamiliar suffix (-pine) and concludes that HSS rely more heavily on familiar structures than do monolinguals. Two previous studies have examined HSS' syllabification of two-vowel sequences. ^{Shelton, Counselman, and Gutiérrez Palma (2017)} look at the effects of cognate status, whether the diphthongs exist as well in English, rising vs. falling diphthongs, and the glide type (palatal vs. velar). They find that each of these factors influences the rate at which HSS syllabify the vowel sequences as diphthongs or hiatus. Heritage participants perform more like monolinguals with words that are not cognates in English, with diphthong sounds that are also existent in English (like the sequence /ai/ in baile 'dance' which is similar to the diphthong in the English word right), with falling diphthongs (which are rare in English), and with velar glides (which are more common in English than palatal glides). In a study of the syllabification of -ear and -iar verbs, ^{Beaton (2020)} divides heritage speakers into two groups based on Spanish language exposure and compares their performance to that of monolingual speakers. Overall, neither group of HSS differs significantly from monolinguals. However, the lower-exposure HSS are more affected by the task type than the other groups; they syllabify /ea/ as a diphthong (which is common in colloquial speech) in the reading task significantly more than in the listening task. ^{Chappell (2018)}, ^{Shelton et al. (2017)}, and Beaton (2020) do not consider the roles of written accent marks and lexical stress.

Studying io sequences with differing stress patterns makes it possible to observe whether stress placement affects heritage speakers' ability to syllabify. Strong literacy skills go hand in hand with the ability to analyze language itself. Therefore, the task of counting syllables, although not predictive of a speaker's language production, does tell us about their metalinguistic intuitions. These intuitions, as discussed above, are important for aspects of writing such as correctly placing accent marks. To account for the fact that heritage speakers are often more adept at processing audio information than written information (^{Bolger & Zapata, 2011}; ^{Montrul, 2008}), participants in the present study complete a task first with written words and then a second task with audio stimuli. This will be explained in greater detail in the methodology section. The research questions for this study are formalized below.

Research Questions

Participants

The forty monolingual participants for this study are college students from the Universidad de Querétero. All were born and raised in Mexico. The heritage and second language learners were recruited from universities across the United States. Most of the heritage speakers (49) have Mexican parents; a few heritage participants (5) are of Central American descent. The fifty-one second language learners were at the intermediate level in their third to fifth semester of college Spanish. After completing the syllabification tasks, all participants filled out a demographic questionnaire, which was used to categorize participants as monolingual, heritage, or L2. This questionnaire was based upon ^{Montrul's Bilingual Background Questionnaire (2012)}. Participants indicated the city and country of their birth, where they currently live, where their parents are from, the language(s) spoken at home while they were growing up, a self-rating of speaking and listening abilities in Spanish and English, and how old they were when they started frequently hearing English and Spanish respectively. The demographic questionnaire is included in Appendix A.

Participants who described their childhood home language as 'Spanish only' and who had spent birth to age 18 in Mexico were classified as monolingual speakers. Second language learners grew up speaking 'English only' and were born and raised in the United States. Heritage homes could be 'Spanish only,' 'mostly Spanish,' or 'an equal mix of Spanish and English.' All the heritage participants began to learn Spanish in infancy and are sequential bilinguals, meaning that they learned Spanish before they learned English. They rated themselves highly on both their English and Spanish speaking and listening skills (fours and fives on a scale of 1-5). Some heritage participants were born outside of the United States and came with their parents as small children or in their early teens. Others were born and raised in the US. In still other cases, heritage participants were born in the US but had spent significant portions of their lives in their parents' home countries. ²

Tasks

Participants completed syllabification tasks on an online Qualtrics survey. The survey had two parts - a reading task and a listening task. In the first part, participants saw a word and were asked to indicate the number of syllables in the word and then to show how they would divide the stimulus into syllables. Participants performed this task for 8 target stimuli and 15 distractor items. Figure 1 below shows an example with the distractor item abeja 'bee.'

Figure 1 Reading task. 

After completing the reading task, participants moved on to the listening task. This second half of the survey was identical to the first except the stimuli were presented as sound clips. The stimuli were read by a phonetically-trained native speaker of Spanish from Guadalajara, Mexico. Participants performed this task for 8 target stimuli and 13 distractor items. See Figure 2 below for a screenshot from the listening portion of the task.

Figure 2 Listening task. 

The target stimuli with io sequences in each task are listed in Table 1 below. The stimuli selected were two or three syllable words with open syllable structure (i.e. none of the syllables in these words have codas). Each of the three stress patterns is represented in both the reading and listening tasks. That is, each task contains a set of words with stress on the i of the io syllable, a set with stress on the o of the io syllable, and a set with stress on the syllable preceding the io syllable. A short training segment was presented to participants before both the reading and listening tasks. Within the tasks, the target items were presented in a random order and distractor stimuli with single-vowel nuclei were included in both tasks.

Accent on i

Vacío ('empty') and judío ('Jew') are examples of Spanish words with stress on a high vowel in a two-vowel sequence. Unlike the other two stress patterns in this study (accent on o and atonic), these sequences are syllabified as separate syllables. As stated above, the intercept in the regression model is stress on i, monolingual speakers, reading task. The insignificant p-values for the three groups in the speaker category main effect indicates that the heritage and second language learners do not behave differently from monolinguals in the listening task. The main effect for task type shows that monolingual speakers perform differently in the listening task than the reading task for this stress pattern; they syllabify in hiatus more in the listening task. Hence, they are more accurate in the listening task than in the reading task. The lack of significant effects for the interaction between speaker category and the reading task shows that the behavior of the other speaker groups is as would be expected for their groups in that task. The HSS and the second language learners, therefore, also assign more syllables in the listening task than the reading task, and their responses do not differ from those of monolinguals.

Accent on o

Words like vació ('emptied') and pidió ('asked for') feature stress on the non-high vowel of the two-vowel sequence and thus are prescriptively indicated as diphthongs.

The main effect for the accent on o stress pattern is significant and the estimate is -0.26. This means that monolingual speakers assign fewer syllables to ió words than ío words in the reading context. The significant findings for the interaction effect for both speaker groups and o accent both have positive estimates. This means that HSS and L2s have more similar rates of hiatus across the i accent and o accent categories than do monolingual speakers

The negative estimate for the interaction effect for o accent and the listening task shows that monolingual speakers assign significantly fewer syllables in the listening task than the reading task. The significant effect for the three-way interaction term between the heritage group, o accent, and the reading task means that heritage speakers are particularly sensitive to task type for this stress pattern. This is evident in the descriptive statistics; monolinguals only differ 10% between the two tasks (22% hiatus in the listening task and 32% in the reading task) whereas heritage speakers have more than double the 'incorrect' hiatus response in the reading task. Unlike in the reading task, HSS do not significantly differ from monolinguals in the listening task. Although the L2 group improves their performance in the listening task, the improvement is less dramatic than that of the heritage group.

The t-tests with Tukey's multiple comparisons adjustment below in Tables 4 and 5 calculate the estimates from the regression model and compare them. This clarifies the differences in group behavior when presented with i accent vs. o accent stimuli. In the reading task, only monolinguals tend to follow the prescriptive rule for i accent words to be in hiatus and o accent to feature diphthongs. Thus, only monolinguals display a significant difference between the two accent patterns in the reading task, as shown in Table 4 below In the audio task (Table 5), however, all three groups successfully distinguish the hiatus from the diphthong, assigning significantly more syllables to the i accent pattern.

Atonic io

Stimuli such as novio ('boyfriend') and medio ('middle') feature the io sequence in a post-tonic syllable. Atonic two-vowel sequences are prescriptively syllabified as diphthongs. The main effect for the atonic stress pattern is significant and the estimate is negatively valanced (-0.42). This shows that monolinguals follow the prescriptive norms and thus are less likely to syllabify in hiatus for the atonic pattern than for the i accent pattern. The two-way interactions for the heritage (0.14) and L2 (0.34) groups are significant and positive, which means that they syllabify in hiatus more than monolingual speakers. When compared to their rates of hiatus for i accent words, heritage speakers, like monolinguals, differentiate the two patterns, assigning significantly more syllables to i accent words than atonic words. Second language learners, however, fail to distinguish the two patterns in the reading task. See Table 6 below.

Unlike with the other two syllable types, there is not a significant difference for monolinguals between the reading and listening tasks. The interaction effect for the atonic pattern in the listening context is significant, but the estimate (-0.1858) essentially cancels out the main effect for the listening task (0.1715). This can be seen in the percentages of hiatus for monolinguals in Table 2 above (16% in reading task and 15% in listening task). The heritage and L2 groups also do not differ in their responses between tasks. This is intuitive from the percentages for each group across tasks (for HSS 25% in the reading task and 29% in the listening task; for L2, 59% and 52% respectively). All three groups assign more syllables to the i accent audio stimuli than the atonic audio stimuli.

Since i accent is in the intercept, the other two stress patterns are compared to it and not to each other in the regression model. Post-hoc t-tests with Tukey's adjustment were performed to compare the o accent and atonic patterns to each other. The results are shown in Tables 8 and 9 below. In the listening task (Table 9), each participant group has insignificant differences between the two contexts. That is, they respond to o accent stimuli in the same way that they respond to atonic stimuli. In the reading task (Table 8), monolingual and second language learners also do not respond significantly differently to these two stress patterns. Heritage speakers, however, are less accurate with o accent stimuli than atonic stimuli in the reading task; they assign more hiatus to o accent words than with atonic words.

1). Do HSS perceive syllables with io sequences like monolingual speakers, like second language learners, or does their behavior differ from both groups?

Their performance lies somewhere between monolinguals and second language learners. In the case of the stress on i pattern, which is prescriptively syllabified in hiatus, neither HSS nor L2 participants differ from monolinguals. Heritage speakers reported hiatus more than monolinguals in the two diphthongization stress patterns (accent on o and atonic - see between interactions speaker category and stress pattern in Table 3). However, their intuitions are less divergent from monolinguals than those of second language learners.

2). Does the stress pattern in io sequences make a difference as to how participants syllabify?

This depends upon the speaker group. For the reading task, monolingual and heritage speakers are influenced in different ways by written accents. Second language learners do not seem to be influenced by the presence or absence of written accents; their hiatus rates are similar across the three stress patterns. Monolingual speakers' intuitions align with prescriptive descriptions of the patterns in both tasks; i accent words are in hiatus whereas o accent and atonic words have diphthongs (see main effects for stress pattern in Table 3). It appears that in the reading task heritage speakers interpret written accents as indications of hiatus and therefore do not significantly differ in their treatment of i accent and o accent but do correctly assign more diphthongs to atonic (which has no written accent) than i accent. These effects are most clearly represented by the comparisons in Tables 4 and 6. The comparison of o accent and atonic is consistent with this hypothesis, heritage speakers display significantly more hiatus judgments with o accent than with atonic. See Table 8. The listening task is addressed in question 3 below.

3) . Does reading a word vs. listening to a word influence how syllable structure is perceived?

For i accent and o accent, yes. For atonic, no. All three groups are more accurate in the listening task than the reading task for i accent and o accent. That is, they use more hiatus for i accent in the listening task than the reading task (see main effect for listening task) and they report less hiatus for o accent in the listening task than the reading task (see interaction between o accent and listening task). The interaction effect for the atonic listening task is significant, but this simply indicates that participants behave differently in the atonic context than the i accent context. The estimate for the interaction between atonic and listening (-0.19) cancels out the main effect for the listening task (0.17), resulting in insignificant differences between the reading and listening tasks in the atonic context.