Editor

JLLT edited by Thomas Tinnefeld
Journal of Linguistics and Language Teaching
Volume 10 (2019) Issue 2, pp. 207-229



Perception and Production of French Nasal Vowel Contrasts by English- and Spanish-Speaking L2 Learners

Vincent Chanethom (Fairfax (VA), USA) & Patrick-André Mather (San Juan, Puerto Rico)


Abstract (English)

This study investigates the acquisition of French nasal vowels among 10 English-speaking and 10 Spanish-speaking L2 learners. Specifically, the study examines how L2 learners maintain these categories separate acoustically in their speech production and whether their ability to perceive such contrasts influences their acoustic implementation. Participants were asked to complete a language proficiency test, a production task where participants were asked to read words containing the three French nasal vowels, and a perception task where they were asked to discriminate between pairs of words. Results revealed significant differences between the two language groups, mainly with respect to the /ɑ̃/ - /õ/ contrast. While L2 Spanish speakers showed no particular difficulty with that contrast, their English-speaking counterparts displayed lower accuracy rates in distinguishing /ɑ̃/ from /õ/ and implemented significantly smaller acoustic distances between the two categories. These empirical findings suggest that, for the study of French nasal vowels, foreign language instructors should design separate pronunciation activities for English native-speaking students, with more emphasis on the /ɑ̃/ - /õ/ pair than the other contrasts.
Keywords: French nasal vowels, production-perception, phonetic acquisition


Abstract (Français)

Cet article étudie l’acquisition des voyelles nasales du français par 10 locuteurs natifs de l’anglais et 10 locuteurs natifs de l’espagnol. L’étude examine en particulier la manière dont les apprenants maintiennent ces catégories vocaliques séparées dans leur production orale et, le cas échéant, l’influence de leur capacité perceptive sur l’implémentation acoustique de ces contrastes. Les participants à cette étude ont complété une série de tâches expérimentales, dont un test de niveau en langue française, un test de production orale pour lire des mots illustrant les voyelles nasales du français, et un test de perception orale pour distinguer des paires de mots. Les résultats de l’analyse indiquent des différences significatives entre les deux groupes de locuteurs natifs, surtout par rapport au contraste /ɑ̃/ - /õ/. Le groupe d’apprenants hispanophones n’a pas montré de difficultés particulières dans la production ou perception de ce contraste. En revanche, les apprenants anglophones ont affiché un taux d’exactitude relativement faible pour la distinction entre /ɑ̃/ et /õ/. De plus, les résultats indiquent aussi que les apprenants anglophones appliquaient une plus petite distance acoustique entre les deux catégories que les apprenants hispanophones. Ces données empiriques suggèrent que pour l’étude des voyelles nasales du français, les enseignants de français langue étrangère doivent développer des activités pédagogiques en prononciation tenant compte des spécificités des apprenants anglophones, en insistant sur le contraste /ɑ̃/ - /õ/.
Mots-cs : voyelles nasales du français, production-perception, acquisition phonétique



1   Introduction

Several phonemes of French are difficult to acquire for both English- and Spanish-speaking learners, in particular the nasal vowels /ɑ̃/, /ɛ̃/ and /õ/. One of the reasons is that while both English and Spanish have secondary nasalization of vowels through coarticulation with neighboring nasal consonants, French makes a phonemic distinction between oral and nasal vowels, as evidenced in the minimal pairs las /la/ 'tired' vs. lent /lɑ̃/ 'slow,' paix /pε/ 'peace' vs. pain /pɛ̃/ 'bread,' and pot /po/ 'jar' vs. pont /põ/ 'bridge.' Vowel nasalization in French is not only a phonemic process, but also shows both acoustic and aerodynamic properties that are distinct from those in English or Spanish. In both English and Spanish, vowel nasalization is a result of anticipatory coarticulation (e.g., Sole 1992 1995, Cohn 1990, Chen 1997, Delvaux, Metens & Soquet 2002; Delvaux 2009. Petersen 2014), by which the velum is lowered during the production of the vowel in preparation for the articulation of the following nasal consonant.

In this study, we investigate how second language learners (henceforth, L2 learners) acquire those French nasal vowels, especially given the difference in phonological status that these phonemes have in their native language. In particular, the study examines how L2 learners maintain these categories separate acoustically in their speech production and whether their ability to perceive such contrasts influences their acoustic implementation. By selecting L2 learners from two different native language backgrounds, namely English and Spanish, we also aimed to examine whether proficiency in another Romance language, i.e. Spanish, presented an advantage in the perception or production of French nasal vowels.


2   Acoustic Correlates of French Nasal Vowels

Although nasal vowels are relatively easy to define relative to oral vowels in terms of their articulatory features – namely, the lowering of the velum that allows the airflow to pass through both oral and nasal cavities -, they are more difficult to characterize acoustically, since the interaction of the vocal and nasal tracts modifies both the frequencies and amplitudes of formants, compared to oral vowels which only make use of the vocal tract.

Chen notes that, according to several studies, the reduction in amplitude of the first formant spectral peak (F1) has been observed to be the primary cue of nasalization” (Chen 1997: 2360). She also shows that, in addition to the three main formants found in oral vowels, nasalized vowels display two additional peaks or resonances, one below the F1 formant (with amplitude P0), and another between F1 and F2 (with amplitude P1). Thus, one way of quantifying nasality is to measure the differences in amplitude between F1 and the first peak (A1-P0), and between F1 and the second peak (A1-P1), both measured in decibels (dB). Comparing amplitude differences between French and English nasalized vowels, Chen (1997: 2369) found that both differences are, on average, slightly greater among English nasalized vowels than French nasal vowels, although she analyzed a relatively small sample of words and speakers.

Styler (2015) examined the acoustic properties of nasality in English (coarticulatory nasality) and French (contrastive nasality), as well as the perception of nasality in English. With regards to the articulatory and acoustic differences between French nasal and oral vowels, Styler (2015: 17-23) also mentioned the additional nasal peaks and amplitude differences noted in Chen (1997), adding that there is a third nasal peak, at around 1250 Hz, albeit only mentioned by one author (Schwartz 1968). In addition to nasality, Styler (2015: 17) further notes that there are also important articulatory differences between oral and nasal vowels, based on research by Delvaux, Metens & Soquet (2002: 349) who found that /ɛ̃/ is more open and centralized than /ɛ/, that /ɑ̃/ is more rounded and back than /ɑ/, /õ/ is more rounded than /o/ (as well as more back / closed for female speakers), and that /œ̃/ is more open and back relative to /œ/. These findings were confirmed both based on formant frequency measures (F1 and F2) and in terms of actual tongue position.

Other researchers (e.g. Carignan 2014) argue that a combined examination of both acoustic and articulatory properties of oral versus nasal vowels in French can help distinguish between the acoustic effects of velo-pharyngeal (VP) coupling and articulatory differences (namely, the position of the tongue) in the production of nasal vowels. Carignan (2014: 24), who analyzed the production of twelve female speakers of Northern Metropolitan French, points out that VP coupling, or lowering of the velum, produces an increase in F1 for high vowels, and a decrease in F1 for low vowels. Similarly, in nasal vowels, there is a decrease in F2 for non-back vowels, and a potential increase in back vowels. Concretely, VP-coupling produces a retraction of /ɛ̃/ (F2 lowering), and a raising of /ɑ̃/ (F1 lowering). Finally, /õ/ is more retracted and more rounded than its oral counterpart (Carignan 2014: 25).

In the current study, we also focus on the acoustic and, to some extent, the articulatory characteristics of the French nasal vowels rather than their nasality, especially with respect to the acoustic distance between each vowel based on F1 and F2 formant frequencies. By examining how L2 learners maintain a contrast between the French nasal vowels, instead of measuring the categories themselves, we investigated their acquisition patterns without necessarily comparing the L2 production with that of monolingual speakers. Indeed, comparing bilinguals' speech production to that of monolinguals may be problematic, given that bilingual speakers sometimes produce nonnative sounds with intermediate acoustic properties between those of the monolinguals (e.g., Caramazza et al. 1973, Flege 1987, Flege, Schirru & MacKay 2003). In other words, their production of L2 sounds is different from their L1 counterparts but may still be significantly different from those of monolingual speakers of the target language. Therefore, we also assumed the possibility that the L2 speakers in the current study may produce the French nasal vowels with acoustic characteristics that are not within the range of those of native French speakers. However, the issue of how much contrast or how much acoustic distance is implemented between each nasal vowel in L2 learners' speech production is of greater interest for our study.

For general reference purposes, the nasal vowels produced by Chen's (1997) French adult native speakers are illustrated in the vowel F1-F2 space in Figure 1:



Figure 1: Acoustic Differences between French Nasal Vowels
as Produced by Chen's (1997) French Native-Speaking Subjects 
(Legend: an= /ɑ̃/; in= /ɛ̃/; on= /õ/)

The acoustic distances between each vowel is represented in solid lines. The data in Chen's (1997) study was elicited using methods comparable to those used in the current study and represent the formant frequency values of eight adult speakers (five female and three male subjects).

To summarize, there are a series of acoustic differences between oral and nasal vowels in French, in terms of both formant frequencies and amplitudes, which are part of the acoustic cues used in the perception of nasality by speakers. The current paper does not directly address the oral vs. nasal difference, but rather, the acoustic contrast between each pair of French nasal vowels, namely /ɑ̃, õ/, /ɛ̃, ɑ̃/, and /ɛ̃, õ/, in terms of Euclidean distance within the vowel space based on the first two formants, i.e. F1 and F2. In particular, we examined L2 the learners' ability to distinguish between the contrasts in their speech perception and to maintain acoustic separation between each pair of French nasal vowels in their speech production. In the following section, we further discuss the acquisition patterns of the French nasal vowels by L2 learners.


3   Acquisition of French Nasal Vowels

As mentioned in the Introduction, the acquisition of French nasal vowels is challenging for both English and Spanish speakers, since nasalization is phonemic in French, whereas in English and Spanish it is produced through co-articulation with neighboring nasal consonants, usually through a process of regressive assimilation. Thus, while nasalized vowels exist in all three languages, their distribution and phonological status are very different.

As Petersen points out, where English displays allophonic nasalization of oral vowels, French suppresses nasal coarticulation when the oral vowel contrasts phonemically with a corresponding nasal vowel(Petersen 2015: 1).  The same rule applies to Spanish, which also nasalizes vowels which are followed by a nasal consonant (Sole 1992, 1995). In an experiment on the production of French nasal vowels by English-speaking learners, Petersen (2015) found that learners at all stages can produce categorical nasal vowels (in the absence of neighboring nasal consonants), but that they produce nasal coarticulation in contexts where oral and nasal vowels contrast, whereas native speakers of French have very low co-articulation effects, in order to maximize the phonemic contrast (Petersen 2015: 8-9). She found that these effects are most marked among beginning speakers, who tend to transfer English nasal coarticulation into French: More advanced speakers are better able to suppress nasalization to preserve the contrast between oral vowels and their nasal counterparts (Petersen 2015: 9).

The acquisition of L2 phonological processes (and suppression of L1 phonological processes in L2 acquisition) are particularly challenging for L2 learners, not only in secondary nasalization (Petersen 2015), but also in other processes. For example, Swanson (2006) reports that English-speaking learners of French tend to transfer English palatalization and aspiration of consonants in French and require additional training to suppress these processes. Another phonetic process, described in Paradis & Prunet (2000), is that L2 learners of French tend to pronounce nasal vowels as an oral vowel followed by a nasal consonant, with regressive nasalization of the vowel, instead of treating them as one single segment. They term this strategy unpacking and use as evidence the adaptation of French loanwords in different languages, including English, Moroccan Arabic and Kinyarwanda (Figure 2 for an acoustic illustration of the unpacking process on a spectrogram). Although their data was taken from French loanwords, it may be hypothesized that the unpacking strategy for nasal vowels may also be used by L2 learners of French, as suggested by Marquez who, in addition to unpacking, looks at a second possible strategy, which he defines as nasal-stripping, in which "learners ignore nasality from the vowel representation, hence turning the nasal vowel into an oral vowel" (Marquez 2016: iv). In an experiment involving a discrimination task, Marquez (2016) found that, while beginning learners of French initially analyze nasal vowels as an oral vowel followed by a nasal consonant (i.e. unpacking), intermediate learners apply the nasal stripping strategy. Only advanced learners are able to acquire French nasal vowels as distinct units with the native phonological representation (Marquez 2016: 124-125):



Figure 2: Spectrogram Illustrating the Unpacking (Paradis and Prunet 2000) of Nasal Vowel /ɛ̃/ as a Repair Strategy by L2 Learners of French

With regards to perception, a listener's ability to distinguish between L2 phones is affected by the acoustic (Flege 1995) or articulatory similarities and differences between the phones (Best, McRoberts & Sithole 1988, Best 1995). The Perceptual Assimilation Model (PAM; Best 1995), as well as its extension PAM-L2 (Best & Tyler 2007), stipulates that listeners perceive nonnative contrasts based on the phonological system of their native language. In other words, L2 learners assimilate L2 phones to existing L1 phonemes within their phonological boundaries with varying levels of phonetic goodness of fit, i.e. based on whether the L2 phone constitutes a good or poor exemplar of an L1 phoneme. PAM predicts good discriminability of L2 phones if they are assimilated to categories of different phonological statuses, and poor discriminability if assimilated to the same phoneme in L1.

As detailed in Best, McRoberts & Goodell (2001), the model considers multiple types of assimilation processes to predict good or poor discrimination of nonnative contrasting phones by L2 learners. For instance, L2 learners may assimilate the two L2 phones to two separate phonemes in their native language. This process is labeled Two Category Assimilation (TC) and predicts an excellent discrimination of the two categories. For example, in Best et al. (2001), English-speaking L2 learners were successful at discriminating between Zulu voiced and voiceless lateral fricatives, despite the absence of such sounds in their L1, given that the nonnative contrast is articulatorily comparable to the coronal contrasts /s, z/ or /ʃ, ʒ/ in English.

By contrast, PAM also predicts the opposite scenario where the two L2 phones are assimilated to the same L1 phoneme with equal goodness of fit. This process is labeled Same Category Assimilation (SC) and leads to poor discrimination of the nonnative phones. In the same study, Best et al. (2001: 13) found that English-speaking L2 learners had difficulty contrasting /b/ from the implosive /ɓ/, because they both represent good exemplars of English /b/, despite the absence of the implosive variant in English. However, if the two L2 phones are assimilated to the same L1 phoneme, but one of the two phones is a better fit, the process is then referred to as Category Goodness Difference (CG) and predicts good discriminability.

Another situation that favors discriminability is the Uncategorized-Categorized Process (UC), where one L2 phone is assimilated to an L1 phoneme while the other one falls between two L1 phonemes. As was the case for TC, the contrasting L2 phones in UC are perceived as entities "on opposite sides of the native phonological boundary" (Best et al. 2001: 5), leading to relatively good discriminability of the nonnative phones. As for the last two situations included in PAM, by which both L2 phones are Uncategorized (UU) or Non-Assimilable (NA), both contrasting L2 phones fall between two L1 phonemes or are perceived as non-speech sounds that do not match any L1 phonemes, respectively. In both cases, discriminability is based on the perceived articulatory differences relative to each other and may range from good to excellent for NA, but fair to good for UU (for further details on PAM, Best 1995; Best et al. 2001).  

In Flege's (1995) Speech Learning Model (henceforth SLM), new category formation is also influenced by the amount of phonetic difference involved. However, while PAM is based on listener's perception of articulatory gestures, SLM is based on acoustic and auditory differences. Regardless of these differences between the two models, both agree that the nonnative contrasts involving the largest phonetic differences are predicted to be the easiest ones to acquire. In other words, the greater the acoustic difference between the two categories, the easier it will be for the L2 learner to distinguish the contrast. Therefore, with regards to French nasal vowels and based on both SLM and PAM, we hypothesize Two Category Assimilation for /ɛ̃, ɑ̃/ and /ɛ̃, õ/ because they are the most distant acoustically and involve a front vs. back vowel contrast, comparable to /ɛ, o/ in both English and Spanish. For /ɑ̃, õ/, on the other hand, we hypothesize Same Category Assimilation, despite the existence of /ɑ, o/ contrast in English and /a, o/ in Spanish, because the categories are very close acoustically (back vowels) and both involve lip rounding. The two models also predict that perception accuracy is a prerequisite for accurate production, which is further discussed in the next section.


4   Perception-Production Correlation

The relationship between the perception and production of sounds by learners of a second or foreign language has been elusive at best. According to Flege (1995), auditory discrimination is a necessary pre-condition for producing a distinction successfully. He points to four theoretically possible relations between production and perception of L2 sounds:
1. Learners can perceive the contrast between two nonnative sounds and are able to produce them.
2. Learners can perceive the contrast between two nonnative sounds but cannot produce them accurately.
3. Learners do not perceive the contrast between two nonnative sounds, but yet are able to produce them.
4. Learners can neither perceive the contrast nor produce them accurately.

Intuitively, hypotheses (1) and (2) seem the most likely, but previous studies on perception and production of L2 sounds have produced a mixed picture. In a study on the acquisition of /r/ and /l/ by Japanese learners of English, Hattori & Iverson (2009) found, as in previous studies, that learners tend to identify English /r/ and /l/ with their Japanese apico-alveolar tap /ɾ/ and therefore interpret the English phonemes as poor exemplars of this Japanese phoneme. However, category assimilation was not predictive of /r/ and /l/ production and perception. They also found only a weak correlation between the perception of these two sounds by learners and the quality of their production as judged by native speakers of English.

In an earlier study, Sheldon & Strange (1982) found that, among some Japanese learners of English in the United States, the production of the contrast between /r/ and /l/ was more accurate than their perception of this phonemic distinction, which seems counterintuitive. Their study was based on a small sample, six Japanese learners of English in the United States, but they mention that two of them "could produce the contrast appropriately, but made perceptual errors" (Sheldon & Strange 1982: 251). In addition, "perceptual difficulties were most marked for /r/ and /l/ in prevocalic stop + liquid clusters" (ibid.). They conclude that
These results indicate that, at least for the contrast studied here, perceptual mastery of a foreign contrast does not necessarily precede adult learners' ability to produce acceptable tokens of the contrasting phonemes, and may, in fact, sometimes lag behind production mastery. (Sheldon & Strange 1982: 254).
Strange (1995) comes to a similar conclusion, namely, that even when learners have a native-like production, perceptual difficulties may persist.

Flege & Schmidt (1995) studied the acquisition of English voiceless stops, specifically the voice onset time among Spanish-speaking learners, and found that perception and production do align, but only among advanced learners.

In a large-scale study (72 subjects) of Italian learners of English in Canada, Flege, MacKay & Meador (1999) also noted a strong correlation between the discrimination of English vowels by learners and the intelligibility of their production of these vowels as evaluated by native speakers of English. They claim that
The significant correlation observed to exist between the measures of L2 vowel production and perception is consistent with another hypothesis of the speech learning model, viz. that the accuracy with which L2 vowels are produced is limited by how accurately they are perceived. (Flege, MacKay & Meador 1999: 2973)
In a meta-analytic review of 25 years of studies on the correlation between perception and production of phones among second-language learners, Sakai & Moorman (2018) found that training in the perception of sounds can bring about modest improvements in production of these sounds, although more so in obstruents than in sonorants or vowels (Sakai & Moorman (2018: 187). These results partially confirm Flege’s (2003) SLM, which claims that perception of L2 sounds can be improved through sufficient exposure, and that accurate perception is a necessary precondition for target-like production (Sakai & Moorman 2018: 188). The 18 studies included in the meta-data analysis were all focused on perception training and tested production before and after training to ascertain whether perception training was indeed useful, as predicted by the SLM model. Most (14) of the studies dealt with the acquisition of English by speakers of Spanish, Chinese and Japanese, so the results are not necessarily applicable to the second-language acquisition of French.

Our study was not aimed at measuring the outcome of perception training, but at testing another part of the SLM model’s hypothesis, namely, whether or not there is a positive correlation between the perception of L2 sounds, namely the three French nasal vowels /ɑ̃/, /ɛ̃/ and /õ/,  and the distinction between these vowels in production in terms of acoustic contrast, regardless of whether the production is “target-like” or not.

In this paper, we address several research questions:
(i) Are learners able to successfully separate the three French nasal vowels in their production, as measured by F1 and F2 frequency differences?
(ii) Are learners able to distinguish the French nasal vowels from each other, based on auditory discrimination tests?
(iii) Is there a correlation between the accuracy in perception and production of French nasal vowels by L2 learners?

To answer these three questions, we recruited 20 undergraduate students of French at two public universities, ten of whom were English-speaking and the other ten, Spanish-speaking. Both groups performed three different tasks, as detailed in the next section.


5   Methodology

As explained above, the current study aims to examine the acquisition patterns of the French nasal vowels /ɑ̃/, /ɛ̃/ and /õ/ produced by learners with different native languages and to investigate the relationship between their production and perception. In our study, native speakers of English are compared to native speakers of Spanish.


5.1 Participants

A total of 20 adult subjects participated in this study (10 in each language group). The native speakers of Spanish were recruited at a university in Puerto Rico and the native speakers of English, at an American university in the state of Virginia. Both language groups were comprised of seven female and three male subjects. Most participants (14) began studying French at university, while the others received some instruction prior to entering college. Despite the difference in terms of onset of French instruction among the participants, a language proficiency test, described in further detail in Section 5.2, indicated that the two language groups were, on average, of intermediate proficiency in French at the time of the experiment. A background questionnaire also showed that the participants had no history of speech delay, developmental disability or neurological impairments.


5.2 Materials and Procedures

The stimuli for the current analysis are a subset of a larger study which investigates the acquisition of three distinct nonnative contrasts, which include the phonemes of interest for the current analysis (the French nasal vowels /ɑ̃/, /ɛ̃/ and /õ/) as well as the phonemic oppositions /u/-/y/ and /s/-/z/. However, the current paper only focuses on French nasal vowels.

Participants were asked to complete four tasks:
(i) a demographic questionnaire to determine their language background;
(ii) a language proficiency test based on Gaillard & Tremblay's (2016) elicited imitation task to determine their level of French;
(iii) a production task where participants were asked to read words in isolation and, again, in a carrier phrase; and
(iv) a perception task where participants were asked to discriminate between pairs of words by indicating whether they were the 'same' or 'different.'

For the French language proficiency test, 50 sentences were recorded by a native speaker of French (the same person who recorded the stimuli for the perception task) and subsequently programmed into PsychoPy (Peirce 2007, 2009). The subjects would hear the sentences being read out and were asked to repeat them to the best of their ability. This method for determining participants’ proficiency in French was chosen for several reasons:
● It emphasizes spoken, rather than written proficiency, which is more relevant for the current study.
● It has been shown to provide a reliable measure of proficiency based on ratings from 1 to 6 along five different scales, namely pronunciation, vocabulary, morphology, syntax, and meaning (Gaillard & Tremblay 2016, for the complete list of sentences and further details on the experimental procedures).

For the production experiment, the stimuli were composed of 21 target words illustrating three different phonetic environments for each phoneme. All vowel targets, i.e. the French nasal vowels and the /u/-/y/ contrast, were preceded by a voiced stop consonant which varied in three places of articulation, namely the bilabial /b/, alveolar /d/, and velar /g/ stops. As for the /s/-/z/ contrast, these consonant targets were followed by three different vowels, namely /i, u, a/. For the French nasal vowels, the target words were all monosyllabic words (Table 1). Each target word was elicited in five repetitions and produced both in isolation and in the carrier phrase je dis _____ pour toi ('I say ___ for you') to obtain more spontaneous speech samples:

Experiment
POA*
/ɑ̃/
/ɛ̃/
/õ/
Production
bilabial
banc (bench)
bain (bath)
bon (good)
alveolar
dans (in)
daim (deer)
don (gift)
velar
gant (glove)
gain (earnings)
gond (hinge)
Perception
bilabial
paon (peacock)
pain (bread)
pont (bridge)
alveolar
temps (time)
thym (thyme)
thon (tuna)
velar
quand (when)
requin (shark)
con (idiot [slang])
   * Place of articulation
Table 1: Stimuli for the Production and Perception Experiments

When a subject did not know a target word, we resorted to delayed repetition, by which a model was provided and the same target was then elicited again at a later stage of the experiment. Thus, a total of 105 tokens were obtained from each participant (Table 1). Among those 105 tokens per subject, 45 were for the French nasal vowels. Thus, a grand total of 900 tokens were obtained from all subjects across language groups. However, upon visual inspections of the spectrograms and auditory inspections of the recordings, 40 tokens were removed from the analysis due to the presence of a nasal consonant following the target sound or the obvious substitution with an oral vowel. This was evidence that the L2 learners used repair strategies, such as unpacking and stripping as discussed in section 3, which were expected to affect the vowel nasality and consequently its formant structures.

As for the perception experiment, which was an AB discrimination task programmed in PsychoPy (Peirce 2007, 2009), the stimuli were composed of target words recorded by an adult female French native speaker from the southern Parisian region of France. As was the case for the production experiment, the wordlist illustrated the target phonemes in three different phonetic environments. For the French nasal vowels, however, to prevent potential carry-over effects from the perception to the production experiments or vice versa, the preceding consonants were voiceless stops instead of voiced (bilabial /p/, alveolar /t/, and velar /k/). As illustrated in Table 1, the list also included nonsense words. Each participant was exposed to a total of 153 word pairs (54 pairs for the category of nasal vowels) with an interstimulus interval of 750 ms. These word pairs were all possible AA, AB, and BB combinations from the aforementioned target words, presented three times each to the subjects. After hearing each word pair, participants were asked to indicate whether the words were similar or different by clicking on the computer keyboard ('Y' for similar and 'N' for different).

Participants' speech production  (i.e. the elicited imitation task and the production task) was recorded using a Zoom H4N 4-band digital recorder, with an ATM75 Audio Technica condenser headset microphone. In total, the four tasks took one hour per participant, including a 5-minute break between each task and an optional break in the middle of the elicited imitation task after the 25 first sentences. The order of the tasks was random. Finally, the procedure was identical for both groups, i.e. the Puerto Rico participants and the Virginia participants, in the same conditions (sound-proof booths) and with the same equipment (recording equipment and laptop computers). The recordings were analyzed using Praat acoustic analysis software (Boersma & Weenink 2017), and the statistical analyses were performed using R statistical package (R Core Team 2016).


5.3 Analyses

Three types of analyses were carried out in the current paper, including acoustic implementations of nasal vowel contrasts, response accuracy in discriminating between pairs of nasal vowels, and correlations between those two variables. For the analysis of acoustic contrasts, frequency values were extracted at midpoint along the first three formant trajectories of each nasal vowel. Normalization was achieved using Bark transformation (Syrdal & Gopal 1986). Mean Euclidean distances between each nasal vowel within the F1-F2 vowel space were then calculated and compared across language groups. For the statistical analysis, linear mixed effects models in R (R Core Team 2016) were performed using lme4 (Bates, Maechler & Bolker 2012), with mean Euclidean distances as the dependent variable, L1 and vowel contrast as fixed effects, and subjects and place of articulation (POA) as random effects. The data were log-transformed to adjust for normal distribution, and likelihood ratio tests were performed to obtain p-values. Pairwise comparisons with Holm-Bonferroni corrections were then used as post-hoc analyses.
As for the perception experiment, a score value of 1 was attributed to correct answers (either 'same' or 'similar'), whereas mismatches were given a score of 0. Accuracy rates were calculated as the average score received for each contrast and for each participant. To test for statistical significance, logistic regressions were carried out using the generalized linear models in R, with response accuracy as the dependent variable, L1 and vowel contrast as fixed effects, and subjects and place of articulation as random effects.

Finally, we also examined the correlations between the Euclidean distances extracted from the production task and response accuracy from the perception task using Pearson's correlation coefficient r.


6   Results

6.1 Production

The objective of the formant analysis is to examine whether L2 learners' ability to maintain a separation between the three French nasal vowels differed with respect to both the subject's native language and the type of vowel produced. In Figure 3, plot (a) illustrates the French nasal vowels /ɑ̃, ɛ̃, õ/ (labeled as an, in, on respectively) in the Bark-transformed F1-F2 vowel space as produced by each L1 group, along with the Euclidean distances between each vowel. The English native speakers are represented in solid lines, whereas Spanish native speakers are represented in dotted lines. Plot (b) shows the same vowel differences in terms of Euclidean distances, but in a bar graph format:





Figure 3: Euclidean Distances between the French Nasal Vowels /ɑ̃, õ, ɛ̃/ as Produced by English and Spanish Native-Speaking Subjects (legend: an= /ɑ̃/; in= /ɛ̃/; on= /õ/)

As shown in Figure 3, the Spanish native speakers implemented greater acoustic distances in terms of Bark differences for the /ɑ̃, õ/ and /ɛ̃, õ/ contrasts (M = 2.06, SE = 0.47 and M = 3.58, SE = 0.41 respectively) than the English native speakers (M = 1.15, SE = 0.39 for /ɑ̃, õ/; and M = 3.09, SE = 0.58 for /ɛ̃, õ/). For the /ɑ̃, ɛ̃/ contrast, on the other hand, the Euclidean distance was larger for the English native speakers (M = 2.32, SE = 0.42) than their Spanish-speaking counterparts (M = 2.27, SE = 0.46). A Linear mixed effects model with Euclidean distance as the dependent variable, L1 and contrast type as the fixed effects, and subject and POA as random effects was performed on the log-transformed data to test for statistical significance. While removing L1 effect did not significantly reduce the model's goodness of fit, removing the effect of contrast type did (X2(2) = 38.58; p < .001). However, the model also revealed a marginally significant interaction between the two factors (X2(2) = 5.83; p = 0.054). Simple effects analyses using pairwise comparisons with Holm-Bonferroni corrections showed no significant effect of L1 for the nasal contrast with the exception of /ɑ̃, õ/ (p = 0.004), for which English native speakers showed significantly smaller distances between the two vowels than their Spanish-speaking counterparts.

The pairwise comparisons further indicated that the /ɑ̃, õ/ difference was significantly smaller than either /ɑ̃, ɛ̃/ (p = 0.03) or /ɛ̃, õ/ (p < 0.001) for the English native speakers. By contrast, the /ɑ̃, õ/ vowel difference implemented by the Spanish native speakers was significantly smaller than the /ɛ̃, õ/ contrast (p = 0.004) but not /ɑ̃, ɛ̃/. Instead, their vowel difference for /ɛ̃, õ/ was significantly larger than their /ɑ̃, ɛ̃/ contrast (p = 0.004), which was not the case for the group of English native speakers. Therefore, the results suggest that English native speakers in the current study may have more difficulty implementing acoustic distance between the French nasal vowels /ɑ̃/ and /õ/ than the Spanish native speakers.


6.2 Perception

To assess the ability to distinguish the three French nasal vowels, we carried out an AB discrimination task in which participants were instructed to listen to different combinations of words and indicate whether the words were similar or different by clicking on the computer keyboard ('Y' and 'N', respectively). Participants' response accuracy was recorded and is represented in Figure 4 as a function of L1 and contrast type. English native speakers are represented in solid lines and Spanish native speakers in dotted lines:





Figure 4: Response Accuracy Rates by Native Language and Contrast Type for the Perception of French Nasal Vowels 
(Legend: an= /ɑ̃/; in= /ɛ̃/; on= /õ/)

As illustrated in Figure 4, the two language groups showed different patterns with respect to their perception of the French nasal vowels. The Spanish native speakers in the current experiment showed the lowest response accuracy rate for discriminating /ɑ̃/ from /ɛ̃/ (M = 0.89, SE = 0.02) and the highest rate for the /ɛ̃, õ/ contrast (M = 0.98, SE = 0.01), with that of the /ɑ̃, õ/ contrast slightly below it (M = 0.97, SE = 0.01). Overall, the mean scores for the Spanish-speaking group were all at ceiling. For the English native speakers, on the other hand, while their response accuracy rates were also at ceiling for the /ɑ̃, ɛ̃/ (M = 0.92, SE = 0.02) and /ɛ̃, õ/ (M = 0.95, SE = 0.02) contrasts, the mean score for the /ɑ̃, õ/ contrast fell within the range of the 70th percentile (M = 0.74, SE = 0.03). A logistic regression was performed using the generalized linear model in R with response accuracy as the dependent variable, L1 and contrast type as the fixed effects to test for statistical significance. The likelihood ratio tests revealed that removing either L1 (X2(1) = -20.85; p < 0.001) or contrast type (X2(2) = 26.45; p < 0.001) significantly affected the model. However, the likelihood ratio tests also yielded a significant interaction between L1 and contrast type (X2(2) = 26.63; p < 0.001). To further investigate the interaction effect, post-hoc Tukey-adjusted pairwise comparisons were performed and revealed that the English native speakers' response accuracy rate for the /ɑ̃, õ/ contrast was significantly lower than that of the other two contrast types (p < .0001 for both). The difference between the /ɑ̃, ɛ̃/ and /ɛ̃, õ/ did not reach significance for that language group. For the Spanish-speaking group, their response accuracy rate for the /ɑ̃, ɛ̃/ contrast was significantly lower than that of the /ɛ̃, õ/ contrast (p = 0.003) and marginally significantly lower than that of the /ɑ̃, õ/ contrast (p = 0.06). The difference between the /ɑ̃, õ/ and the /ɛ̃, õ/ contrast did not reach significance. As for the cross-linguistic comparisons, the two language groups differed significantly with respect to their accuracy rate for /ɑ̃, õ/ (p < 0.0001), but not for the other two contrasts. This result suggests that the Spanish speakers were better at distinguishing /ɑ̃/ from /õ/ than the English native speakers, which is consistent with the patterns discussed in the speech production analysis in the previous section.


6.3 Correlations

To assess the link between perception and production in L2 learners with regards to French nasal vowels, we computed Pearson product-moment correlation coefficients to examine the relationship between the subjects' ability to implement formant frequency distances between the categories in their speech production and their response accuracy rate during the auditory discrimination task. Given Flege's (1995) SLM, we hypothesized such correlation to be positive. In other words, the better the subjects are at distinguishing between the French nasal vowels, the more acoustic difference (in terms of Euclidean distance) they are able to implement in their production.

The analysis did yield positive correlations between the two variables for both language groups. However, only the correlation for the English native speakers reached significance (r(84) = 0.42, p = 0.02). The correlation for the Spanish native speakers was marginally significant (r(84) = 0.33, p = 0.08) and weaker than the correlation coefficient for the English native speakers. These results are summarized in the scatterplots provided in Figure 5, which illustrates the formant frequency differences in terms of Euclidean distance as a function of response accuracy rate for all French nasal vowel contrasts, along with their regression lines. The English native speakers are represented in plot (5a) and their Spanish-speaking counterparts in plot (5b). For convenience, the Pearson's correlation coefficients are also reiterated under each scatterplot:


Figure 5: Correlation Scatterplots of Response Accuracy Rates by Euclidean Distances for all French Nasal Vowels Combined for English (a) and Spanish (b) Native Speakers

To examine the relationship between production and perception in further details, the data were arranged by nasal vowel contrast type within each language group. Pearson product-moment correlation coefficients were then computed for each category. As was the case for the data set collapsed over contrast types, there were positive correlations between vowel difference in terms of Euclidean distance and response accuracy rate, with the exception of the /ɛ̃, õ/ contrast category for the Spanish native speakers, which showed a weak negative correlation (r = -0.014). However, only one category showed a significant correlation, namely the /ɛ̃, õ/ contrast category for the English native speakers (r(26) = 0.44, p = 0.02). The results are summarized in scatterplots in Figure 6, in which English native speakers are represented by the plots in the top row and Spanish native speakers in the bottom row. For convenience, the Pearson's correlation coefficients are provided under each plot.

Given the limited amount of data for each nasal vowel category, these results may not be conclusive, and further analysis may be necessary to fully account for the relationship between the production and perception of French nasal vowel contrasts by L2 learners:



Figure 6: Correlation Scatterplots of Response Accuracy Rates by Euclidean Distances for English (a) and Spanish (b) Native Speakers organized by Contrast Type
(Legend: an= /ɑ̃/; in= /ɛ̃/; on= /õ/)


7   Discussion

The objective of the current study was to investigate the perception and production of French nasal vowels by adult L2 learners from two different native language groups. Adult native speakers of English and Spanish were compared with respect to their acoustic implementation of contrast in their production of French nasal vowels and with respect to their response accuracy in the auditory discrimination of these vowels. Acoustic contrast between each French nasal vowel was measured in terms of Euclidean distance within the F1-F2 vowel space. Both variables, i.e. acoustic contrasts and response accuracy rates, were analyzed with respect to the speaker's L1 and the type of nasal vowel contrast.

Similar results were found in the production and perception analyses, which indicated that the two language groups did not show significant differences with the exception of the /ɑ̃, õ/ contrast. The English native-speaking subjects were not only found to implement significantly shorter acoustic distances in their production of that French nasal vowel contrast, but were also less accurate in distinguishing them auditorily than the Spanish native speaking subjects. This finding is consistent with both Best's (1995) PAM and Flege's (1995) SLM, by which L2 learners may only produce a nonnative contrast if they can perceive it accurately. Thus, since the English native speakers have difficulty discriminating /ɑ̃/ from /õ/, they will also have difficulty maintaining them separate in their speech production.

The within-group analysis further corroborates the special status of the /ɑ̃, õ/ contrast for the English native-speaking subjects, which had significantly shorter acoustic distance than the other two contrasts. In the perception task, it also received significantly lower response accuracy rates than the other contrasts, whose scores were at ceiling. For the Spanish native speakers, the response accuracy rates for all three contrasts were at ceiling in the perception task. As for the production task, their acoustic distance implemented on the /ɑ̃, õ/ pair patterned with that of the /ɑ̃, ɛ̃/ contrast. Their production of the /ɛ̃, õ/ contrast, on the other hand, had significantly greater acoustic distance than for the other two contrasts. In fact, the acoustic distance for /ɛ̃, õ/ was the largest of the three contrasts for both language groups. This finding is not surprising, given that the pair involves the largest phonetic differences of the three French nasal contrasts. /ɛ̃/ is front and unrounded, whereas /õ/ is back and rounded. Therefore, it is easier to perceive and to produce as suggested in our current data, which is consistent with Flege's SLM.

In summary, both production and perception analyses in the current study indicate that the /ɑ̃, õ/ pair may be the most difficult contrast to acquire for the English native speakers but not for the Spanish native speakers.

The first question that occurs is why the /ɑ̃, õ/ contrast is more difficult than the others for L2 learners? The second question is why Spanish native speakers have less difficulty with it than the English native speakers? There are several possible explanations for these findings. As discussed in Carignan's (2014) study, which involved twelve native speakers of French, one of the significant effects of velopharyngeal opening on the production of the French nasal vowel /ɑ̃/ is its raising in comparison to the oral vowel /a/. Such an effect would therefore reduce the acoustic distance between /ɑ̃/ and /õ/, rendering it more difficult for L2 learners to distinguish and, consequently, to produce these vowels. Of the three nasal vowel contrasts, /ɑ̃, õ/ involves the smallest phonetic differences, mainly with respect to tongue height. Both vowels are round and back, whereas the other pairs, i.e. /ɑ̃, ɛ̃/ and /ɛ̃, õ/, differ along both phonetic features. With the added effect of VP coupling (Carignan 2014), the small phonetic differences between /ɑ̃/ and /õ/ become even smaller, hence the difficulty for L2 learners to discriminate one from the other.

As for the difference between the two language groups, one possible explanation is that English native speakers are more sensitive to vowel roundedness than Spanish native speakers. Strange, Levy & Law (2009), for example, found that the set of French front rounded vowels /y, ø, œ/ were most likely assimilated with the English back round vowels /u, o, ɔ/ by English-speaking L2 learners, despite their greater similarity to the English front unrounded vowels /i, e, ɛ/ in terms of tongue articulation. With regards to French nasal vowels, Delvaux, Metens & Soquet (2002) found that /ɑ̃/ and /õ/ are even more rounded than their oral counterparts /ɑ/ and /o/, potentially further affecting the perception accuracy by the English native speakers, who may overlook the small tongue articulatory differences between the two vowels to focus on the prominent phonetic feature of roundedness.

An additional likely explanation for the difference between the two language groups is how the contrasting French nasal vowels are assimilated to the L2 learner's native phonological system as discussed in Best's (1995) PAM. As discussed in Section 3, vowel nasalization is present in both English and Spanish as a result of anticipatory coarticulatory process, which is non-phonemic. In Spanish, for example /a/ in año 'year', /o/ in concepción 'conception' and /e/ in bien 'good' are produced as nasalized variants. It is therefore likely that each French nasal vowel /ɑ̃, õ, ɛ̃/ represents a good exemplar of each of the three nasalized vowels in Spanish, as illustrated by the Two Category Assimilation in PAM, hence the high response accuracy rates and, consequently, relative ease in maintaining them separate during production for that language group. On the other hand, in the American English variant spoken by the participants in the current study, while both /o/ and /ɑ/ are nasalized by coarticulatory processes in the words don't and pan respectively, the latter tends to be produced as the nasalized front mid vowel /ɛ/ as in pen. Thus, the French nasal vowel /ɑ̃/, being rounded and back, is potentially assimilated to the nasalized /o/ as in don't, rather than the nasalized /ɑ/ as in pan. This would exemplify PAM's Same Category Assimilation, because both /ɑ̃/ and /õ/ are assimilated to the nasalized /o/ as in don't, hence the poor discriminability by the English native subjects.

Finally, the current study also investigated the relationship between the production of French nasal vowels and their perception. The correlation analysis yielded relatively low correlation coefficients overall, which is consistent with previous studies, as summarized in Flege (1999), in which only very low correlations were found. When collapsed across nasal vowel contrasts, a significant positive correlation was found for the English native speakers and a marginally significant one for the Spanish native speakers. In other words, the more accurate the subjects were in distinguishing one French nasal vowel from another, the more acoustic distance they implemented in their production of these vowels. This finding is consistent with Flege's (1995) SLM, by which the successful perception of a nonnative contrast is a necessary condition for an L2 learner to produce that contrast.

To further investigate this relationship between perception and production, the correlation analysis was organized by contrast type within each language group. However, only one reached significance, namely a positive correlation for the /ɛ̃, õ/ contrast for the English native speakers. The fact that other correlations were not significant is not surprising, especially given the low number of tokens per L1-vowel type combination. Again, these results are consistent with previous studies. For instance, Flege, Mackay & Meador (1999) found significant correlations between perception and production only for advanced learners. Thus, given that the subjects in the current study were on average at intermediate level of French proficiency, the low number of significant correlations reported in the current analysis is also consistent with Flege, Mackay & Meador (1999). Surprisingly, however, no significant correlations were found for the Spanish-speaking learners, who were on average more advanced in their L2 proficiency. Therefore, it is possible that the question of perception vs. production in L2 acquisition depends on more than just the L2 proficiency factor. It is also possible that the effect of perceptual mastery on speech production may not be immediate. In other words, production accuracy of the nonnative contrasts may occur at a later stage of acquisition. For example, Nagle (2017) found significant time-lagged effects of perception mastery on Voice Onset Time during stop consonant production by English native speakers learning Spanish over time. Indeed, the correlation analyses in our study investigated synchronous effects only. It is possible that similar time-lagged effects, as found in Nagel's (2017) study, may be present in the acquisition of French nasal vowels as well. However, longitudinal data are needed to fully investigate this issue.  


8   Conclusion

Speech perception and production experiments were carried out to investigate how native speakers of American English and Spanish perceive French nasal vowels /ɑ̃, ɛ̃, õ/ and maintain acoustic separation between them in their production. The analysis yielded significant differences between the two language groups mainly with respect to the /ɑ̃, õ/ pair. While the Spanish native speakers showed no particular difficulty with that contrast, their English-speaking counterparts displayed low response accuracy rates in distinguishing /ɑ̃/ from /õ/ and implemented significantly smaller acoustic distances between the two categories than for the other nasal vowel contrasts and compared with the Spanish native speakers.

These results suggest that, for the study of French nasal vowels, foreign language instructors should design separate pronunciation activities for English native-speaking students, with more emphasis on the /ɑ̃, õ/ pair than the other contrasts, especially activities focusing on auditory discrimination. It is a well-known fact that learners’ L1 can be a predictor of challenges with specific L2 phonemes, and our study brings further evidence that teaching methods must take into account important empirical findings on the relation between speech perception and production, and on differences among L1 groups in acquiring a given L2.



References

Bates, Douglas, Martin Maechler & Ben Bolker (2012). lme4: Linear mixed-effects models using S4 classes. R package version 0.999999-0.

Best, Catherine (1995). A direct realist view of cross-language speech perception. In: Strange, Winifred (Ed.) (1995). Speech Perception and Linguistic Experience: Issues in Cross-language Research. Timonium, MD: York Press, 171-204.

Best, Catherine & Michael Tyler (2007). Nonnative and second-language speech perception: Commonalities and complementarities. In: Munro, Murray & Ocke-Schwen Bohn (Eds.) (2007). Second Language Speech Learning: The Role of Language Experience in Speech Perception and Production. Amsterdam: John Benjamins, 13-24.

Best, Catherine, Gerald McRoberts & Elizabeth Goodell (2001). Discrimination of non-native consonant contrasts varying in perceptual assimilation to the listener's native phonological system. In: Journal of the Acoustical Society of America 109, 775-794.

Best, Catherine, Gerald McRoberts & Nomathemba Sithole (1988). Examination of perceptual reorganization for nonnative speech contrasts: Zulu click discrimination by English-speaking adults and infants. In: Journal of Experimental Psychology: Human Perception and Performance 14, 345-360.

Boersma, Paul & David Weenink (2017). Praat: doing phonetics by computer [Computer program]. Version 6.0.31. (http://www.praat.org/; 21 August 2017).

Caramazza, Alfonzo, Grace Yeni-Komshian, Edgar Zurif & Ettore Carbone (1973). The acquisition of a new phonological contrast: The case of stop consonants in French-English bilinguals. In: Journal of the Acoustical Society of America 54, 421-428.

Carignan, Christopher (2014). An acoustic and articulatory examination of the ‘oral’ in ‘nasal’: The oral articulations of French nasal vowels are not arbitrary. In: Journal of Phonetics 46, 23-33.

Chen, Marilyn (1997). Acoustic correlates of English and French nasalized vowels. In: The Journal of the Acoustical Society of America 102, 2360-2370.

Cohn, Abigail (1990). Phonetic and phonological rules of nasalization. UCLA: Phonetics Laboratory, Department of Linguistics.

Cohn, Abigail (1993). Nasalisation in English: phonology or phonetics. In: Phonology, 10.1:43–81.

Delvaux, Véronique (2009). Perception du contraste de nasalité vocalique en français. In : Journal of French Language Studies 19(1), 25-59.

Delvaux, Véronique, Thierry Metens & Alain Soquet (2002). Propriétés acoustiques et articulatoires des voyelles nasales du français. In : XXIVèmes Journées d’Étude sur la Parole, Nancy 1, 348-352.

Feng, Gang & Eric Castelli (1996). Some acoustic features of nasal and nasalized vowels: A target for vowel nasalization. In: The Journal of the Acoustical Society of America 99, 3694-3706.

Flege, James (1995). Second language speech learning: Theory, findings, and problems. In: Strange, Winifred (Ed.) (1995). Speech Perception and Linguistic Experience: Issues in Cross-language Research. Timonium, MD: York Press, 233-277.

Flege, James (1999). Age of learning and second language speech. In: David Birdsong (Ed.) (1999). Second language acquisition and the Critical Period Hypothesis. Mahwah, NJ: Lawrence Erlbaum Associates, 101-131.

Flege, James, Ian MacKay & Diane Meador (1999). Native Italian speakers’ perception and production of English vowels. In: The Journal of the Acoustical Society of America 106, 2973-2987.

Flege, James, Carlo Schirru & Ian MacKay (2003). Interaction between the native and second language phonetic subsystems. In: Speech Communication 40, 467-491.

Flege, James & Anna Schmidt (1995). Native speakers of Spanish show rate-dependent processing of English stop consonants. In: Phonetica 52(2), 90-111.

Gaillard, Stéphanie & Annie Tremblay (2016). Linguistic Proficiency Assessment in Second Language Acquisition Research: The Elicited Imitation Task. In: Language Learning 6(2), 419-447.

Hattori, Kota & Paul Iverson (2009). English /r/-/l/ category assimilation by Japanese adults: Individual differences and the link to identification accuracy. In: The Journal of the Acoustical Society of America 125, 469-479.

Labov, William (22006, 1966). Social Stratification of English in New York City. Cambridge: Cambridge University Press.

Marquez Martinez, Miguel (2016). The Acquisition of French Nasal Vowels: From First Language Allophony to Second Language Phonological Contrast. Unpublished PhD Thesis. Indiana University: Department of French and Italian.

Nagle, Charles (2017). Examining the temporal structure of the perception–production link in second language acquisition: A longitudinal study. In: Language Learning 68(1), 234-270.
Paradis, Carole & Jean-François Prunet (2000). Nasal vowels as two segments: Evidence from borrowings. In: Language 76(2), 324-357.

Peirce, Jonathan (2007). PsychoPy – Psychophysics software in Python. In: Journal of Neuroscience Methods 162(1-2), 8-13.

Peirce, Jonathan (2009). Generating stimuli for neuroscience using PsychoPy. In: Frontiers in Neuroinformatics 2, 10.

Petersen, Stacy (2015). Categorical Nasal Vowel Acquisition in L2 French Learners. In: University of Pennsylvania Working Papers in Linguistics, 21(1), Proceedings from PLC, 38, article 25. 

R Core Team (2017). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. (https://www.R-project.org/; 30 November 2017).

Sakai, Mari & Colleen Moorman (2018). Can perception training improve the production of second language phonemes? A meta-analytic review of 25 years of perception training research. In: Applied Psycholinguistics 39, 187-224.

Schwartz, Martin (1968). The acoustics of normal and nasal vowel production. In: Cleft Palate Journal 5, 125-140.

Sheldon, Amy & Winifred Strange (1982). The acquisition of /r/ and /l/ by Japanese learners of English: Evidence that speech production can precede speech perception. In: Applied Psycholinguistics 3(3), 243-261.

Sole, Maria (1992). Phonetic and phonological processes: The case of nasalization. In: Language and Speech 35(1-2), 29-43.

Sole, Maria (1995). Spatio-temporal patterns of velopharyngeal action in phonetic and phonological nasalization. In: Language and Speech 38(1), 1-23.

Strange, Winifred (1995). Speech Perception and Linguistic Experience: Issues in Cross-language Research. Timonium, MD: York Press.

Strange, Winifred, Erika Levy & Franzo Law (2009). Cross-language categorization of French and German vowels by naïve American listeners. In: Journal of the Acoustical Society of America 126(3), 1461-1476.

Styler, Will (2015). On the Acoustical and Perceptual Features of Vowel Nasality. Unpublished PhD Thesis. University of Colorado: Department of Linguistics.

Swanson, Kimberley (2006). Acquisition versus Suppression of Phonological Processes in the Second Language Acquisition of French and English. Unpublished PhD Dissertation. Indiana University: Departments of French and Italian and Linguistics.

Syrdal, Ann & H. S. Gopal (1986). A perceptual model of vowel recognition based on the auditory representation of American English vowels. In: Journal of the Acoustical Society of America 79, 1086-100.



Authors:

Vincent Chanethom, Ph.D.
Assistant Professor of French Linguistics
George Mason University
Department of Modern and Classical Languages
Aquia Building, Room 318
Fairfax, VA 22030
Email: vchaneth@gmu.edu

Patrick-André Mather, Ph.D.
Professor of French and Linguistics
Universidad de Puerto Rico, Río Piedras
Department of Foreign Languages / Graduate Program in Linguistics
13 Ave Universidad, Ste 1301
San Juan, PR 00925
Email: patrick.mather@upr.edu