Abstract
This study examines how L1 English-L2 French learners use L1 articulatory and acoustic categories to produce L2 vowels that are both similar to and different from their L1 vowels. Previous studies examining the relationship between L1 and L2 sound inventories have found that learners reuse L1 phone categories to produce L2 phones that are perceived as similar, but importantly, there is a lack of articulatory data included in these types of studies, which has reinforced the assumption that vowel categories can be solely represented by their acoustic properties. The present study uses ultrasound tongue imaging data and videos of lip rounding in addition to acoustic data to examine how L1 English-L2 French learners produce the French vowels /i y u e ø o/ compared with their English vowels /i u e o/. The results focus on individual paths to category formation to show how learners articulate L2 vowels, and reveal that they tend to reuse L1 tongue body gestures to produce the French vowels /i u e o/, and lip rounding gestures to produce the round vowels /y u o/. This study demonstrates that transfer of articulatory gestures depends on vowel quality and emphasizes the importance of using articulatory data to inform theories of L2 category formation.
1 Introduction
It is well documented that a learner’s first language (L1) will impact their perception and production of a second language (L2) (Darcy et al., 2012; Flege, 1987; Ingram & Park, 1997; Major, 2008). It is claimed that L2 sound production depends on how L2 sounds are perceived in relation to L1 sounds (Best & Tyler, 2007; Flege, 2005; Kamiyama & Vaissière, 2009). The present study is designed to examine how learners use L1 acoustic and articulatory categories to produce L2 phones. Data of this type provide an avenue to explore the mental representation of sounds by investigating how L1 and L2 sound systems interact. The interaction of two sound systems, and the transfer (or lack thereof) between them, can shed further light on the representation of sound units by examining the phonetic nature of the categories speakers use in L2 productions.
The Speech Learning Model (SLM; Flege, 1987) and Speech Learning Model–Revised (SLM-r; Flege & Bohn, 2021) propose that when an L2 phone is perceived as similar to an L1 phone, learners equate the two categories, which leads to learners reusing L1 categories in L2 productions. SLM postulates that the acoustic-phonetic category is the relevant unit of representation of sound.
The Perceptual Assimilation Model-L2 (PAM-L2) is similar to SLM and SLM-r, but importantly differs as to what is the basic phonetic unit of sound. PAM-L2 suggests that it is the articulatory gesture, rather than the acoustic-phonetic category, that is the basic unit of speech perception (Best & Tyler, 2007). L2 perceptual learning, therefore, depends on how L2 gestures are perceived in relation to L1 gestures.
Production data from L1 English-L2 French learners provide an avenue to test the predictions of PAM-L2 and SLM because these learners have to learn to produce L2 vowel sounds that are predicted to be perceived as different from L1 vowels in addition to L2 vowels that are perceived as similar to L1 vowels. French /y/ and /ø/ are front round vowels and therefore phonologically different from L1 English vowels. French /u/ and /o/ are similar to English /u/ and /o/, but French back vowels are typically produced with lower F2 values (Flege, 1987). This study addresses whether learners reuse L1 acoustic and/or articulatory categories in L2 productions, and whether new phoneme categories are more target-like in either acoustic or articulatory productions.
Learners may use different articulatory strategies to reach the same acoustic targets. For example, both tongue body backing and lip rounding have the consequence of lowering F2, and are thus in a trading relationship with one another (Gay et al., 1981; Guenther et al., 1999). Furthermore, acoustics and articulation are in a quantal rather than linear relationship (Stevens & Keyser, 2010). Regions of acoustic stability, or “quantal regions,” arise as the articulatory dimension varies continuously (Stevens, 1989), such as the relatively sharp change in F2 that occurs when the tongue body moves continuously from a fronted position to a backed position (Stevens & Keyser, 2010). It is, therefore, possible for learners to transfer articulatory gestures from their L1 to their L2 without necessarily reusing the same acoustic categories, or for learners to transfer acoustic categories from their L1 to their L2 without reusing the same articulatory gestures.
1.1 L2 category formation
Models of L2 category formation emphasize how the L1 phonological system influences perception and production of L2 phones for adult second-language learners. One model of category formation that has received much attention is SLM. SLM and the recently revised version SLM-r posit that the L1 and L2 phonetic systems exist in the same phonological space, meaning L1 phones will influence L2 phone perception and production (Flege, 2005; Flege & Bohn, 2021). A non-native phone will be perceived as either “similar” or “different” in relation to an L1 phone, and the perception and production of the L2 phone depend on how similar this phone is to any L1 category (Flege, 1987). One hypothesis of SLM and SLM-r is that the greater the perceived dissimilarity between an L1 and L2 phone, the more likely a new category will be formed for the L2 phone. If an L2 phone is perceived as different from any L1 phone, then a listener is likely to create a new category for this phone. A second hypothesis is that an L2 phone will assimilate to an L1 category if it is perceived as similar to an L1 phone (Flege, 2005). Finally, SLM predicts that the degree of perceptual dissimilarity between two phones strongly influences how “accurately” an L2 phone will be produced (see Major, 2008). If an L2 phone is perceptually similar to an L1 phone, then an L2 learner will use the L1 phone for the production of its L2 equivalent. Alternatively, if an L2 phone is perceptually different from any L1 category, the learner will not use L1 categories to produce this phone, which can cause productions to be more target-like. Because SLM assumes acoustic-phonetic cues underlie L2 phonological learning, phonological categories are acoustic in nature. These specific predictions of SLM have been supported with acoustic data from several languages, including French, which are reviewed in Section 1.2.
SLM-r similarly claims that L2 sound categories form in the same phonetic space as L1 sound categories (Flege & Bohn, 2021). One way in which the SLM-r differs from the SLM is the focus on individual speakers rather than between-group differences as concerns the linking of L2 sounds to L1 sounds. The present study, therefore, examines the transfer of L1 categories to produce L2 phones at the individual rather than the group level.
The Perceptual Assimilation Model (PAM) has also been central to discussions of non-native phone perception, specifically for naïve perception among listeners with no target-language experience. In this model, a listener’s ability to discriminate non-native contrasts depends on how non-native phones are perceptually mapped to the phoneme categories in their L1 (Best et al., 2001). Importantly, PAM claims that the articulatory gesture is the phonological primitive that underlies speech sound perception. Non-native phones may be perceived as similar to an L1 sound, different from any L1 sound, similar to an L1 sound but still phonetically distinct, or perceived as a non-speech sound. If two sounds are perceptually mapped to two separate phoneme categories in the L1 (Two-Category assimilation), a listener will be excellent at perceiving the contrast between the non-native phones. If two sounds are mapped to the same category in their L1 (Single-Category assimilation), a listener will display poor discrimination of these two sounds. If two non-native phones are mapped to the same L1 category, but one phone is a better exemplar of this category (Category-Goodness assimilation), listeners will vary in how well they can discriminate between these phones.
PAM-L2 extends the predictions made by PAM to L2 perceptual learning while adhering to some of the central tenets of SLM (Best & Tyler, 2007). Best and Tyler (2007) discuss central postulates of SLM, and where the assumptions of PAM differ. Important to the present discussion, SLM states that language-specific properties of sounds are stored in memory as phonetic categories. PAM and PAM-L2 differ from SLM as concerns which properties of speech sounds are stored as representations. SLM assumes that sound categories are formed through perception of acoustic-phonetic cues, whereas PAM and PAM-L2 take a direct-realist approach to speech perception (Best, 1995; Fowler, 1986; Gibson, 1979), in that listeners are presumed to perceive articulatory gestures used to produce speech sounds. In this view, learning happens when learners perceive invariant information in the speech signal (Best & Tyler, 2007).
Thus, PAM-L2 assumes two levels of representation are involved in L2 perceptual learning; learners may detect differences and similarities between L1 and L2 sounds on the phonetic or phonological level. The phonetic level involves perception of gestural information that is sub-lexical. The phonological level involves perception of gestural information that is relevant to lexical contrasts, akin to the “higher-order invariant information in the speech signal,” discussed by Best and Tyler (2007). If an L1 and L2 phone are perceived as similar on the phonetic level, it can be assumed that learners do not perceive a difference between the articulatory gestures or organs that are used to produce the two phones. However, if an L1 and L2 phone are perceived as similar on the phonological level, it is not necessarily the case that the two phones are perceived as phonetically similar. One example that shows the distinction between similarity on the phonetic and phonological level involves the French and English rhotics. French /ʁ/ is produced as a uvular fricative, whereas the English /ɹ/ is produced as a post-alveolar approximant. Although these two sounds are often not confusable to learners on the phonetic level, learners tend to equate these two phonological categories (Best & Tyler, 2007).
Keeping in mind these differences from SLM, PAM-L2 also makes explicit predictions about L2 perceptual learning. First, if only one L2 phone is perceived as phonetically similar to an L1 category, learners are not predicted to perceptually learn a new contrast. This L2 phone will be mapped to the corresponding L1 phone category. The next possibility is that learners may perceive an L2 phone as phonologically similar, but phonetically deviant, from an L1 phone. This is the case with /ʁ/ for English learners of French. Because French /ʁ/ is phonetically different but phonologically similar to English /ɹ/, learners will build a new phonetic category for French /ʁ/ within the L1 rhotic category (Best & Tyler, 2007). The next possibility is that two L2 phones will be assimilated to the same L1 phonological and phonetic categories, but one phone will be considered “deviant” or less phonetically similar to the L1 category, similar to the Category-Goodness assimilation of PAM. PAM-L2 predicts that new phonological and phonetic categories will be formed for the deviant phone, whereas no new category will be formed for the “better fit” phone. Finally, two L2 phones may be considered equally good or bad instances of an L1 phone, in which case they would be assimilated to the same L1 phonetic and phonological categories, similar to Single-Category assimilation. Whether or not learners can learn to perceive the phonetic difference between L2 phones that are assimilated to the same category depends on whether the phones are good or poor examples of an L1 phone.
PAM-L2 is a model for L2 perceptual category formation, and is not explicitly a model of L2 production. However, because the articulatory gesture is considered the basic phonetic unit, it can be assumed that production will be guided by the L2 perceptual categories that have been formed. If a learner perceives an L2 phone as phonetically and phonologically similar to an L1 phone, then it can be hypothesized that the learner will recruit L1 articulatory gestures to produce the L2 phone. If a learner perceives two L2 phones as phonetically and phonologically similar to an L1 phone but one L2 phone is more “deviant” (a Category-Goodness assimilation pattern), then L1 gestures may only be recruited for the more similar L2 phone.
The present study uses articulatory data from L2 learners’ productions of mid and high round vowels to evaluate the assumptions of PAM-L2 and SLM/SLM-r. L1 and L2 French learners were recorded producing the vowels /i u e o/ in English, and /i y u e ø o/ in French. If vowels have an acoustic-phonetic target (as assumed by SLM and SLM-r), we would expect learners to produce perceptually similar L2 and L1 vowels with similar acoustic-phonetic values, and perceptually dissimilar L2 and L1 vowels to be produced in a different acoustic-phonetic space. Rather, if vowels have an articulatory target, we would expect learners to transfer L1 articulatory gestures to produce similar L2 vowels, and to produce new L2 vowels with different articulatory strategies from L1 vowels. The specific predictions for each of these approaches are discussed for each vowel in Section 2, after summarizing previous work on the L2 acquisition of French vowels that informs the hypotheses of the present study.
1.2 L2 acquisition of French vowels
Perceptually, L1 English speakers tend to assimilate front round vowels to their back round vowel categories. Using an ABX categorization task, Levy and Strange (2008) found that L1 English listeners with and without French language experience have difficulty perceiving the French /y/-/u/ contrast, and claim that learners who have trouble perceiving this contrast assimilate both /y/ and /u/ to English /u/. Similarly, Darcy et al. (2012) report that intermediate and advanced learners have difficulty perceiving the French /y/-/u/ and /ø/-/ɔ/ contrasts in an ABX categorization task (note that French /œ/-/ø/ and /ɔ/-/o/ are only marginally contrastive, and the present study uses the International Phonetic Alphabet symbols for the close-mid vowels /ø/ and /o/ to represent the French mid round vowels). Thus, learners who have difficulty perceiving the contrast between French front round and back round vowels are predicted to map (or equivalence classify) all French round vowels to their L1 round back vowel categories.
Production of the French /y/-/u/ contrast has been shown to be difficult for learners as evidenced by acoustic data. Beginner, intermediate, and advanced learners all tend to produce acoustically target-like /y/ but have difficulty producing French /u/. Flege (1987) argues that learners produce French /u/ using their L1 category. This causes production errors as French /u/ is typically produced with a lower F2 than English /u/. Flege (1987) explains these results by claiming that the front round vowel cannot be classified as an L1 phone because there is no category similar to /y/ in English. Therefore, learners will make an entirely new category for /y/, thus making their productions target-like. These findings indicate that new phones are easier to acquire than similar phones.
Similar results have been found for high and mid vowel productions by L1 Japanese-L2 French learners. Kamiyama and Vaissière (2009) compared Japanese-French learner productions of /u/, /y/, and /ø/. This study found that Japanese-French learners had difficulty producing target-like /u/, similar to English-French learners. Japanese learners produce French /u/ with the same acoustic values as their L1 /u/. Because Japanese /u/ is typically produced with a higher F2 than French /u/, learners were, therefore, non-target like in their French /u/ productions. For the front round vowels, Japanese-French learners produced target-like /y/ and /ø/ (Kamiyama & Vaissière, 2009). /y/ and /ø/ do not have comparable categories in Japanese, making these phonemically and phonetically new phones. Again, Kamiyama and Vaissière (2009) claim that these results indicate that phonetically new but phonemically similar vowels are difficult to produce.
In contrast, Levy and Law (2010) found that L1 English-L2 French learners have difficulty producing French front round vowels in addition to French /u/. In a production task, learners of varying proficiency levels produced /i y u œ a/ that were judged for accuracy by native speakers and via acoustic analyses. Native French speakers had difficulty identifying the learners’ productions of /y/, /œ/, and /u/ (Levy & Law, 2010). In light of Flege’s (1987) claim that learners were more accurate in producing the “new” French front round vowel /y/, Levy and Law (2010) claim that French /y/ is only “new” in certain consonant environments as English /u/ is often produced as [y] in coronal contexts. Although the current study does not focus on how coarticulation with preceding and following consonants affects vowel production, this finding indicates that the phonetic experience L1 English-L2 French learners have with their L1 English vowels impacts how the French front round vowels are mapped to L1 vowel categories.
From an SLM/SLM-r perspective, L1 English-L2 French learners may initially classify French /y/ and /u/ as English /u/, and French /ø/ and /o/ as English /o/, given the results discussed above that English back vowels have fronted variants in coronal contexts. Learners may also classify French /i/ as English /i/, and French /e/ as English /e/. As learners are exposed to more phonetic input in French, they may create a new category for French /y/ and /ø/. The L1-L2 categories for /i/, /u/, /e/, and /o/ would be linked and influence one another (Flege & Bohn, 2021). Phonetically, North American English tense vowels tend to be produced as diphthongs (e.g., Hillenbrand et al., 2001), whereas French tense vowels are more monophthongal, and the different dynamic properties of French and English vowels may influence how learners use their L1 categories to produce these L2 vowels. From a PAM-L2 perspective, L1 English-L2 French learners who have difficulty perceiving the French /y/-/u/ and /ø/-/o/ contrasts perceive these round vowels as phonetically and phonologically similar to their L1 round vowel categories. If learners perceive phonetic (that is articulatory) differences between French /y/ and English /u/, or French /ø/ and English /o/, they should then create new categories for /y/ and /ø/. Similarly, if learners perceive that French /u/ is produced with a more posterior tongue body than English /u/ and that French /o/ is produced with a more posterior tongue body than English /o/, they should create new phonetic categories for French /u/ and /o/. Finally, if learners perceive the monophthongal production of French /i/ and /e/ compared with English /i/ and /e/, they should also create a new phonetic category for French /i/ and /e/. However, if they do not perceive the different tongue gestures for these vowels, they will not create new categories for /i/ and /e/.
Taken together, the results discussed in this section highlight the importance of L1 categories and phonetic experience in shaping the production of L2 phones. However, little work has been done to understand how L1 articulatory strategies affect L2 productions. L1 English-L2 French learners who have difficulty producing the contrast between French front and back round vowels acoustically may still produce the contrast articulatorily because articulatory movements and acoustic outputs are not in a one-to-one relationship. This result would suggest that the contrast is encoded articulatorily. Furthermore, SLM and, indirectly, PAM-L2 predict that learners will produce an L2 phone as an L1 phone if the two phones are similar, but these theories do not concur about the units of analysis or the targets of speech sounds more generally. If vowels have an articulatory target, learners may use L1 articulatory strategies to produce L2 phones, whereas if they have an acoustic target, learners may use L1 acoustic-phonetic categories to produce similar L2 phones. On the contrary, if an L2 phone does not have an L1 counterpart or is perceived as a deviant exemplar of an L1 phone, this vowel will have its own category created in a speaker’s phonological space. For front round vowels, an L1 English learner must use novel combinations of articulatory gestures. A learner must maintain a fronted tongue body with rounded lips normally associated with back vowels. As previous results have shown that learners can produce target-like /y/ and /ø/, they may be using a combination of novel articulatory gestures to reach this acoustic target. This study focuses on individual speaker results to better understand the articulatory and acoustic categories used to produce the French vowels.
This study examines the articulatory patterns of L2 category formation and seeks to answer the following research questions:
The overarching goal of this study is to better understand the targets of vowels in L2 speech production. This is done using acoustic and articulatory data that examine how the contrast between the French round vowels is realized and thus encoded, and what types of targets learners transfer from their L1 to produce L2 vowels.
2 Hypotheses
L1 English-L2 French learners who have difficulty perceiving the contrast between French /y/-/u/ and /ø/-/o/ are hypothesized to be mapping both round vowel categories to L1 English /u/ and /o/ (see Darcy et al., 2012; Levy, 2009; Levy & Strange, 2008 for more discussion on the perceptual discrimination and categorization of French vowels). Due to their relatively similar phonetic characteristics compared with other vowels in each language’s inventory, L1 English learners are predicted to perceptually assimilate (or equivalence classify) French /i/ and /e/ to the phonological categories of English /i/ and /e/, respectively.
There are several possibilities for how learners will transfer L1 phonetic categories to produce the French round vowels depending on how learners are mapping the L2 vowels to their L1 phonological categories. First, learners may produce French /y/ and /ø/ with the same tongue configuration as English /i/ and /e/, respectively, and with lip rounding comparable with English /u/ and /o/. Under such a scenario, learners would have successfully recruited L1 gestures to produce L2 phones, specifically by combining the appropriate fronted tongue position with lip protrusion. Such productions would suggest that vowels have an articulatory rather than acoustic target because learners are using articulatory targets from their L1 to produce a target-like L2 vowel, thus supporting the assumptions of PAM-L2. The second possibility is that learners produce French /y/ and /ø/ with the same tongue position and lip rounding as the English round vowels /u/ and /o/. In this case, learners would be recruiting an entire gesture constellation from their L1 to produce L2 phones rather than recruiting specific gestures. The acoustic values for /y/ and /ø/ would differ from native speakers, suggesting learners do not have an acoustic target. Rather, vowels may have an articulatory target, and learners are transferring an entire constellation from their L1, again supporting the assumptions of PAM-L2. Finally, learners’ articulations of French /y/ and /ø/ may be different from any English articulation. In particular, learners may have target-like acoustic values of the French front round vowels, but tongue and lip gestures may be different from those used to produce their L1 English categories. This would suggest that tongue and lip gestures are used in a trading relation to reach a specific acoustic value, but perhaps the vowels do not have an articulatory goal, as predicted by SLM/SLM-r.
3 Methods
3.1 Ultrasound tongue imaging
Ultrasound tongue imaging is a technique for imaging the tongues surface during speech production. This technique involves placing an ultrasound probe beneath a speaker’s jaw. The probe emits sound waves that travel through the speaker’s tongue and bounce back when they hit the air above the tongue’s surface (Zsiga, 2013). The resulting image of the tongue’s surface is created by calculating the time it takes the wave to travel back to the probe (Zsiga, 2013). Because the ultrasonic waves travel through fluid and bounce back when they hit air, only the tongue surface is able to be imaged. Other structures in the vocal tract, such as the hard palate, soft palate, and alveolar ridge, are not necessarily visible using this technique (however, see Epstein & Stone, 2005 for ways to image the palate using ultrasound).
Because imaging the tongue using ultrasound involves an external probe, it is necessary to control for any movements of the probe’s position. One of the most common methods is the use of a head stabilization headset to which the probe is attached. In such a set-up, the speaker can move their head throughout the course of the study, but the probe will make contact and stay in the same position relative to the speaker’s head (Scobbie et al., 2008). The current study used a stabilization headset manufactured by Articulate Instruments Ltd (“Ultrasound Stabilisation Headset Users Manual: Revision 1.4,” 2008). For a review of the stabilization headset along with a visualization of the equipment, see Pucher et al. (2020).
One of the major advantages of using ultrasound to image articulation is that both the tongue tip and tongue root are visible (Davidson, 2006). Because ultrasound can image the continuous length of the tongue and does not image contact between the tongue and the upper surface of the vocal tract, it is particularly useful for comparing tongue shapes for vowels (Gick et al., 2006; Hudu, 2014; Mielke et al., 2017; Scobbie et al., 2012; Stone et al., 1988; Strycharczuk & Scobbie, 2017; Zharkova, 2013), rhotics (Boyce et al., 2016; Lawson et al., 2011; Mielke, 2015; Preston et al., 2019), laterals (Charles & Lulich, 2019; Nagamine, 2022; Tabain & Beare, 2018), and fricatives (Chiu et al., 2020; Francisco & Wertzner, 2017; Lee-Kim et al., 2014; Zharkova, 2016). Other imaging techniques such as electropalatography (EPG) measure contact between the tongue and the palate, and therefore are more accurate with detecting place of articulation, but do not show the movement of the tongue when it does not make contact with the palate.
Figure 1 shows an example of an ultrasound image of the tongue taken for the present study, with the tongue tip on the left of the figure, body, and root on the right of the figure. The Articulate Assistant Advanced (AAA) system (which was used in the present study for data collection and analysis) can semi-automatically track tongue movements. An example of the tongue spline for this ultrasound image is shown in red in Figure 2.
Ultrasound image from native French speaker NF2’s production of “futur,” with tongue tip pointing to the left.
Ultrasound image of the tongue, with the red spline tracing the lower surface of the tongue image, the green spline indicating the upper limit of the speaker’s oral cavity, and the white spline indicating the lower limit of the tongue movement.
3.2 Participants
Six L1 English-L2 French learners (identified as FL1–FL6) completed a production task in French and in English. All speakers were self-reported native speakers of American English, aged 18–21. FL1, FL2, and FL4–FL6 were female speakers, FL3 was a male speaker. All learners were enrolled in Intermediate French II at the time of the study, which is a fourth semester, college-level French language course taught at a university on the East Coast of the United States by a speaker of European French. Learners had between 1 and 9 years of French classroom instruction and reported no prior knowledge or exposure to any other language with front round vowels. FL3 was the only speaker who reported having any immersion experience with French, having spent time in Switzerland in middle school (ages 11–14). No learners reported any prior pronunciation instruction. Two L1 French speakers (NF1 and NF2, 21 and 28 years of age, respectively) also completed the production task in French. NF1 is female from the Bordeaux region of France. NF2 is male from Haut-de-France, in Northern France. Both participants were living in Washington D.C. at the time of the study, and working in a French-speaking government office. The relatively small number of speakers allows for a detailed investigation into each learner’s articulatory and acoustic strategies of L1 category transfer. Previous ultrasound studies with comparable samples sizes have similarly provided a detailed investigation of individual articulatory patterns for a variety of other speaker groups (Adler-Bock et al., 2007; Allen et al., 2013; Lawson et al., 2013; Lee-Kim, 2014; Lee-Kim et al., 2013; Mielke et al., 2011; Sisinni et al., 2016). Table 1 provides an overview of each learner’s background information including length of French instruction.
Language Background and Basic Demographic Details for the L1 English-L2 French Learners.
3.3 Materials
Two wordlists were presented in the production task. The first was in French, containing each of the French vowels /i y u e ø o/ in the first syllable of six consonant-initial words, controlled for preceding and following consonant place of articulation. Most words were bisyllabic, but some monosyllabic and trisyllabic words were included in an effort to use only words that would be familiar to learners. Each of the 36 words was repeated twice, for a total of 12 tokens of each target French vowel per speaker. Three additional words containing the target vowel /ø/ were also included in the word list because some of these words were less familiar to learners of this level, meaning that some words would have to be removed. Post-experiment, the researcher verbally asked each participant whether there were any words they did not know, and all expressed familiarity with all of the French words on the wordlist, with the exception of queue /kø/ “tail,” which was not known by five of the learners. Thus, ‘queue’ was eliminated from analysis. The second wordlist was in English, with each of the target vowels /i u e o/ occurring in six real words, controlled for the place of articulation of the preceding and following consonant, and word position. Here too, each word was repeated twice, for a total of 12 tokens of each target English vowel per speaker. The two word lists can be found in Appendix A.
3.4 Procedure
Participants individually completed the production task in one recording session. A researcher was present to set up the equipment and monitor each speaker’s progress. As mentioned earlier, participants wore a stabilization headset (“Ultrasound Stabilisation Headset Users Manual: Revision 1.4,” 2008), which was attached to an ultrasound probe and a video camera. The probe was positioned underneath each participant’s jaw to record a mid-sagittal section of tongue position at a rate of 84 frames per second. The video camera was positioned to capture the frontal view of participants’ lips. All articulatory data were recorded in AAA (Articulate Instruments Ltd, 2012).
The word lists were presented orthographically in AAA as single-word utterances on a computer screen. Participants were instructed to read each word after the screen turned green (indicating AAA was recording) as best they could. Acoustic data were recorded using a Marantz PMD661 recorder and a head-mounted microphone.
3.5 Analysis
Acoustic measurements presented in this study consist of F1 and F2 based on previous research that evaluates the contrast between French front round and back round vowels along the F1-F2 dimension (for example, see Levy & Law, 2010; Melnik-Leroy et al., 2022 for analysis of L2 French learners). Articulatory measurements of the tongue body gestures and lip position were made because these are the gestures widely used to describe the contrast between vowels.
The acoustic data were analyzed in PRAAT (Boersma & Weenink, 2017). F1 and F2 were extracted at the midpoint of each vowel. The midpoint was selected for analysis because the acoustic midpoint of a vowel can be considered to correspond to the gestural target, and furthermore, measuring F1 and F2 at the midpoint helps to minimize the acoustic effect of the preceding and following consonant on vowel quality. First, F1 and F2 values from the native speaker productions of French high vowels and French mid vowels were compared using a multivariate analysis of variance (MANOVA) (Hall-Lew, 2010; Nycz & Hall-Lew, 2013), to confirm that they produce an acoustic difference between /i, y, u/, and /e, ø, o/. This provided a baseline of comparison for the French learners. F1 and F2 values of learner productions of French /y/ and /ø/ were subsequently compared with their French /u/ and /o/ using two within-speaker MANOVAs, with F1 and F2 as dependent variables. This analysis revealed which learners produce a contrast between French front and back round vowels. Next, to examine how learners produce L1 and L2 phones that are hypothesized to be similar, learner productions of French and English /i e u o/ were compared with each other using four different MANOVAs for each speaker. Finally, F1 and F2 of the “new” French phones /y/ and /ø/ were compared with the English back vowels using a MANOVA.
To measure lip rounding, still frames were extracted from the video recordings at the maximum point of constriction of each target vowel. The maximum point of constriction was selected for analysis because the target for lip rounding can correspond to the maximum point of rounding or unrounding. However, it must be noted that, in the present study, the acoustic midpoint measurements were not necessarily taken at the same time point, and therefore lip rounding results cannot be directly compared with other measurements. Following Havenhill and Do (2018), lip rounding was measured as the horizontal distance between the inside corners of the mouth, and the vertical distance from the inside of the upper and lower lips. The distance of lip aperture is directly correlated to lip rounding because the distance between the lips necessarily decreases as the lips round (Fromkin, 1964; Krause et al., 2020). Distances were measured using Inkscape’s measurement tool, which calculates the number of pixels in a calibration frame, and then measures the ratio of pixels to millimeters. This ratio was then used to measure the distance of the mouth opening in all video frames for each speaker separately (Havenhill & Do, 2018). Next, the area of the lip opening for each vowel was approximated as an ellipse, the area of which was calculated by multiplying the radius length of the x-axis (i.e., half the distance of the horizontal opening) by the radius of the y-axis (i.e., half the area of the vertical opening) by π. This provided a single measure for the degree of lip opening for each vowel.
For tongue position measurements, the semi-automatic tongue surface tracker in AAA (Version 218.01) was used to trace the tongue image. The spline tracker was monitored during analysis and hand-corrected when the spline tracker misidentified the tongue surface. For each vowel, one ultrasound frame at the acoustic midpoint of each vowel was selected for analysis. Cartesian coordinates of the tongue spline position were then extracted out of AAA. Smoothing spline analyses of variance models (SSANOVAs) were used to compare tongue positions of target vowels in Cartesian coordinates, using R. The x-value range for each participant was limited to the x-coordinate values for where the tongue is visible for all the vowels and the splines do not extend past the visible tongue. Because the SSANOVA is the average of each of the 12 vowel tokens, some SSANOVA splines appear to be unnatural for tongue shapes. For more discussion of SSANOVAs in linguistics research, see Davidson (2006).
First for the high and then the mid vowels, tongue positions were compared using one SSANOVA for NF1 and one for NF2; the purpose of this analysis was to determine how native speakers produce front versus back round vowels as a point of comparison for learners. Next, learner productions of the French high vowels and mid vowels were compared. Finally, each learner’s tongue position for all French vowels was compared with their English vowels of the same height, to investigate whether learners transfer L1 tongue position for L2 vowels. The results will be discussed in Section 4.
Acoustic measurements, lip rounding, and tongue positions are not directly compared across speakers, but rather how each speaker produces the contrasts between vowels. Observing how learners produce the contrast between the French vowels and use L1 acoustic and articulatory categories to produce these vowels provides insight into individual paths to L2 category formation. Data of this type have received relatively less attention than group averages, but provide important insights into the acoustic and articulatory representation of second-language speech sounds.
4 Results
To simplify the discussion, the results for high vowel productions are presented first, followed by mid vowel productions.
4.1 High vowels
4.1.1 Acoustic results
First, the two native speakers’ productions of the French high vowels are shown in Figures 3 and 4. As can be seen, they produced the high round vowels /y/ and /u/ in non-overlapping acoustic regions, as expected.
F1 and F2 values (Hz) of speaker NF1’s French high vowel productions.
F1 and F2 values (Hz) of speaker NF2’s French high vowel productions.
Overall, many of the learners tended to produce French /y/ in the same F1 and F2 region as French /u/. The vowel plot for one representative learner who produced French /y/ in a different region of the F1-F2 space from French /u/ is shown in Figure 5 and the vowel plot for one representative learner who produced French /y/ and /u/ in a similar F1 and F2 region is shown in Figure 6. All other learners’ acoustic vowel plots can be found in Supplement Appendix A.
F1 and F2 values (Hz) of speaker FL3’s high vowel productions.
F1 and F2 values (Hz) of speaker FL2’s high vowel productions.
MANOVA results reveal that the native speakers did produce a significant difference in the F1 and F2 values of /y/ and /u/. The Pillai scores and significance codes showing the amount of difference between F1 and F2 for French /y/ and /u/ for all speakers are provided in Table 2. Pillai scores indicate the degree to which two distributions are distinct in a given space (for example, the F1-F2 space Hall-Lew, 2010; Nycz & Hall-Lew, 2013). A Pillai score closer to 1 indicates no similarity in F1 and F2 for the two vowels, whereas a Pillai score closer to 0 indicates complete similarity of F1 and F2 for the two vowels.
Output of MANOVAs Evaluating the Difference in F1 and F2 Values of French Vowels /y/ and /u/, by Speaker.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Table 2 shows that two of six learners produced a significant difference in the F1-F2 space between French /y/ and /u/ (FL1 and FL3). A post hoc test shows that the two native French speakers produced a significant difference between French /y/ and /u/ in F2 but not F1. Similarly, a post hoc test shows that FL1’s productions of /y/ and /u/ differed significantly in F2 (p = .0064**) but not in F1 (p = .26), and FL3 also produced a significant difference in F2 for /y/ and /u/ (p = .00024***), but not in F1 (p = .74). The other four learners did not produce an acoustic contrast in the F1-F2 space between the high rounded vowels. Because the native speakers did produce a contrast between /y/ and /u/, learners who did not produce this contrast are referred to as non-target-like.
To look at what might be causing the non-target-like productions of the /y/-/u/ contrast as well as to determine whether there is a difference between the target-like and non-target-like learners, two groups will be referenced for the remainder of the results. Group 1 consists of two speakers, FL1 and FL3, and is the “contrast” group (learners who appropriately produced the contrast between the high round vowels). Group 2 consists of FL2, FL4, FL5, and FL6, and is the “no contrast” group (learners who did not produce the /y/-/u/ contrast).
Again using an MANOVA with F1 and F2 as dependent variables, learner productions of French /u/ and English /u/ are compared and French /y/ to English /u/. Pillai scores and associated p values for Group 1 speakers are shown in Table 3.
Output of MANOVAs Evaluating the Between-Language Difference in F1 and F2 High Round Vowels, by Speaker in Group 1.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Group 1 speakers FL1 and FL3 both produced a significant difference in the F1 and F2 space between French and English /u/. Both of these speakers produced an acoustic contrast between French /y/ and /u/ and were not using their English /u/ category to produce French /u/. However, these two speakers had slightly different production patterns. FL1 produced French /y/ and English /u/ in similar regions of the F1-F2 space. FL1 did not produce French /u/ and English /u/ in similar regions of the F1-F2 space, and thus created a “new” round back vowel category for French /u/. FL3, on the contrary, did not produce French /y/ overlapped with English /u/, or French /u/ with English /u/, thus showing three acoustically distinct high round vowel categories.
Group 2 results comparing acoustic measurements of French /u/ and /y/ and English /u/ are presented in Table 4. Most Group 2 speakers (FL2, FL4, and FL5) did not produce a difference in F1 and F2 for French and English /u/. Among these learners, FL2 and FL5 did not produce French /y/ and English /u/ in similar regions of the F1-F2 space. FL4 was the only learner in Group 2 who did not produce a difference in F1 and F2 for French /y/ and English /u/, as well as French /u/ and English /u/, meaning FL4 only had one acoustic high round vowel category. Only one learner in Group 2 (FL6) did produce a significant acoustic difference between French and English /u/. This learner also did not produce a significant acoustic difference between French /y/ and English /u/, showing a similar pattern to Group 1 learner FL1.
Output of MANOVAs Evaluating the Between-Language Difference in F1 and F2 High Round Vowels, by Speaker in Group 2.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Turning now to compare how learners produced the front unround vowel in their L1 and L2, Table 5 shows the Pillai scores and associated p values within-speaker MANOVAs comparing F1 and F2 values of French and English /i/ for Group 1 learners.
Output of MANOVAs Evaluating the Between-Language Difference in F1 and F2 High Unround Vowels, by Group 1 Speaker.
Both learners in Group 1 produced French and English /i/ in the same F1-F2 space. Table 6 provides the Pillai scores and associated MANOVA p values comparing F1 and F2 values of Group 2’s French and English /i/. Most learners in Group 2 did not produce a difference in F1 and F2 for French and English /i/.
Output of MANOVAs Evaluating the Between-Language Difference in F1 and F2 High Unround Vowels, by Group 2 Speaker.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
FL6 was the only learner in either group who did not produce L2 French /i/ with the same F1 and F2 as their English /i/. FL6 was also the Group 2 speaker who produced a significant acoustic difference between French and English /u/. This learner will be discussed further in Section 5.
4.1.2 Lip rounding results
The degree of lip rounding for the native speakers’ French vowels are compared to show how these speakers produced the lip rounding contrasts between /i/-/y/-/u/. Figures 7 and 8 illustrate the size of the lip opening produced by NF1 and NF2. Because the distance from the camera to each speaker’s lips differed, the area of lip rounding is not normalized across speakers and results are only given for individual speakers.
Area of lip opening (mm2) for French high vowels produced by native French speaker NF1.
Area of lip opening (mm2) for French high vowels produced by native French speaker NF2.
A one-way ANOVA for NF1 confirms that there is a significant difference in area of lip opening between the three high vowels, F(2, 33) = 184.6, p < .001***, and a post hoc Tukey HSD test shows that both /y/ and /u/ differed significantly in lip opening from /i/, but not from each other. A one-way ANOVA for NF2 shows the same pattern; the three high vowels have a significantly different opening, F(2, 33) = 36.01, p < .001***, and a post hoc Tukey HSD test shows that /y/ and /u/ differed significantly from /i/ but not from each other.
Having established that the two native French speakers had a similar degree of lip rounding for the high round vowels, which both differed from the unround vowel, learner productions of French /i/, /y/, and /u/ were compared. A one-way ANOVA for each speaker comparing area of lip opening for each of the French vowels shows that learners in Groups 1 and 2 produced French /y/ and /u/ with the same degree of lip rounding, and produced /i/ differently from /y/ or /u/ (similar to the native speaker results). Within-participant results of the one-way ANOVAs for Group 1 learners are shown in Table 7, and Group 2 learner results in Table 8.
Output of Within-Speaker One-Way ANOVAs and Tukey HSD Tests Comparing Area of Lip Opening for the High French Vowels /i y u/ for Group 1 Speakers.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Output of Within-Speaker One-Way ANOVAs and Tukey HSD Tests Comparing Area of Lip Opening for the High French Vowels /i y u/ for Group 2 Speakers.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Tables 7 and 8 show that all Group 1 and 2 learners in this study produced rounding on high vowels similar to the native-speaker controls. All learners rounded French /y/ and /u/ to a similar degree (given the results of the Tukey HSD test showing no significant difference in lip opening area between /y/ and /u/ for any speaker), and all learners rounded the high round vowels /y/ and /u/ more than French /i/. This pattern is further shown for two representative speakers in Figures 9 and 10. Results for all other speakers are provided in Supplement Appendix B.
Area of lip opening (mm2) for French and English high vowels produced by learner FL5.
Area of lip opening (mm2) for French and English high vowels produced by learner FL2.
Looking at the between-language lip rounding patterns, all learners show a similar pattern for the French and English high vowels: French vowels tend to be more rounded than English vowels of the same quality which can be seen in Figure 9. A second one-way ANOVA run for each speaker comparing the area of lip opening for all French and English vowels shows that the pattern is only significant for French and English /i/ for most speakers. The output of each ANOVA and relevant post hoc Tukey HSD results are presented in Table 9. Almost every learner had the same pattern, in that French /i/ tended to be significantly more rounded than English /i/ (shown for learner FL5 in Figure 9). In addition, although French /u/ appeared to be more rounded for each speaker, there is no significant difference in degree of lip rounding between English /u/ and French /y/ and /u/.
Output of F-Values From Within-Speaker One-Way ANOVAs and Tukey HSD Tests Comparing Area of Lip Opening for English and French High Vowels.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
FL2 (shown in Figure 10) was the only learner examined here who did not produce a significant difference in lip rounding between French and English /i/. However, FL2 followed the same trend as the other learners, in that French vowels tended to be more rounded than English vowels, even though this difference was not significant.
4.1.3 Tongue position results
We turn now to the results of the ultrasound tongue imaging to compare how learners and native French speakers use tongue position to distinguish the high vowels. First, two separate SSANOVAs were run for NF1 and NF2 comparing tongue position at the midpoint for tokens of French /i/, /y/, and /u/. The SSANOVA plots are shown in Figures 11 and 12. For all figures presented in this section, the tip of the tongue is on the left side of the figure, the root of the tongue on the right side. The solid line represents the average tongue contour of the 12 tokens of each French and English vowel. The dashed line represents the Bayesian 95% confidence interval. Any point where the dashed lines of two vowel types do not overlap, there is a significant difference (at the .05 level) in tongue position for those two vowel types.
SSANOVA of tongue position of the French high vowels, produced by NF1.
SSANOVA of tongue position of the French high vowels, produced by NF2.
The output of the SSANOVAs shows that the two native speakers did not produce a significant difference in tongue position for the front vowels /i/ and /y/. Furthermore, neither of the front vowels overlaps with the back vowel /u/ at any point of the tongue contour. These results confirm that native French speakers produced the front vowels /i/ and /y/ with the same tongue position, and that /u/ is realized further back. The contrast between /i/ and /y/ is articulatorily distinguished by lip position. The contrast between French /y/ and /u/ is articulatorily distinguished by tongue position.
Next, we examine how learners articulated the contrast between the French high vowels. Learner FL3’s SSANOVA plot is presented in Figure 13 as representative of the two Group 1 learners. Recall that Group 1 speakers produced French /y/ and /u/ in different regions of the F1-F2 space.
SSANOVA of tongue position of the French high vowels, produced by FL3,
Group 1 speakers produced a difference between French /y/ and /u/ in the acoustic space, and produced /y/ with a more fronted tongue position compared with /u/. However, as shown in Figure 13, FL3 also produced /y/ with a different tongue position than /i/. The other Group 1 speaker (FL1) also produced a difference in tongue position between /y/ and /i/, which can be seen in Supplement Appendix C. This is different from the native speakers, who produced /i/ and /y/ with the same tongue position.
Learners FL2 and FL6’s SSANOVA plots comparing tongue position for the French high vowels are shown in Figures 14 and 15 as representative of Group 2 learners; all other learner plots can be found in Supplement Appendix C. Notably, all learners in Group 2 lacked a significant difference in tongue position between French /y/ and /u/ at least for the tongue root. For all learners in Group 2, /i/ tended to be more fronted than /y/, although learners FL2 and FL6 overlapped /i/ and /y/ along the tip of the tongue. Nevertheless, no Group 2 learners produced the contrast between /i/, /y/, and /u/ with target-like tongue positions. In particular, they differed from the native speakers who had the same tongue position for front vowels /i/ and /y/ versus a more posterior position for back /u/.
SSANOVA of tongue position of the French high vowels, produced by FL2.
SSANOVA of tongue position of the French high vowels, produced by FL6.
Finally, learners’ tongue positions for French and English vowels were compared to address the question of whether learners transferred L1 articulatory gestures to L2 phones that are both similar and new. Two SSANOVAs were run for each speaker comparing between-language high vowel productions: one for front vowels, another for round vowels. Learner productions of French /i, y/ and English /i/ were compared, as well as French /y, u/ to English /u/. SSANOVA plots for learner FL3 are presented in Figures 16 and 17 as representative of the two Group 1 learners.
SSANOVA of tongue position of the French and English high front vowels, produced by FL3.
SSANOVA of tongue position of the French and English high round vowels, produced by FL3.
Both learners in Group 1 did not produce a difference in tongue position between French /i/ and English /i/ or French /u/ and English /u/ along most of the tongue contour (although FL3 produced English /u/ slightly lower than French /u/ at the constriction location). FL3’s tongue position for the new French phone /y/ was significantly different from English /i/ and English /u/. The other Group 1 learner did not produce French /y/ with a tongue position different from English /u/ (see Supplement Appendix C for all the SSANOVA plots).
Figures 18 and 19 show the results of SSANOVA plots comparing FL6’s productions of English and French high front vowels, and English and French high round vowels, as representative of Group 2 learners. Again, all other SSANOVA plots can be found in Supplement Appendix C.
SSANOVA of tongue position of the French and English high front vowels, produced by FL6.
SSANOVA of tongue position of the French and English high round vowels, produced by FL6.
Examining transfer patterns for the similar L1 and L2 vowels, learners in Group 2 did not produce a difference in tongue position between French and English /i/ or /u/. This can be seen in the overlapped confidence intervals in Figures 18 and 19. Although learner FL6 did produce French /i/ with slightly higher tongue position than English /i/ at the maximum point of constriction, this difference is not significant along the majority of the tongue contour. For the new phone /y/, no learners in Group 2 produced a difference between French /y/ and English /u/ along the tongue root, but all learners did produce French /y/ with a more anterior tongue front than English /u/.
4.1.4 High vowels: interim summary
Section 4.1 presented the results of high vowel productions by native French speakers and L1 English-L2 French learners. Acoustic analysis showed that only two of the six learners who participated in this study produced a phonetic contrast between French /y/-/u/ in the F1-F2 space (referred to here as Group 1 learners). Four of six learners did not produce a target-like acoustic distinction between French /y/ and /u/ (referred to here as Group 2 learners). Turning to the results of the articulatory analysis, all learners were target-like in the lip rounding of French rounded vowels. For tongue position, Group 1 learners tended to back French /u/ more than /y/, but did not produce /y/ with the same tongue position as /i/, unlike the native speakers. Group 2 learners’ French /u/ and /y/, in contrast did not differ significantly along the root of the tongue. It is likely that the fronted tongue position of /u/, as well as the backed tongue position of /y/, caused the acoustic overlap between /y/ and /u/ for Group 2 learners.
Turning to learners’ transfer of high vowel acoustic and articulatory categories from their L1 to their L2, half of the learners in the present study had the same acoustic categories for French and English /u/, and all learners (except FL6) produced French and English /i/ in the same region of the F1-F2 space. For lip rounding, all learners produced both French high round vowels with the same degree of lip rounding as their English high round vowel. This differed for unround vowels, where only one speaker out of six rounded French and English /i/ to a similar degree. For the ultrasound results, most learners had the same tongue position for French and English /u/, which suggests the fronted tongue position of French /u/ was due to transfer of the L1 English tongue body gesture. No learners, even those in Group 1, produced French /y/ with the same tongue position as their English /i/. All learners overlapped their French and English /i/ along the entire contour.
The phonetic categories that are transferred from a speaker’s L1 to their L2 appear to depend on vowel quality. The learners in the present study were more likely to transfer the lip rounding posture of the high round vowel from English to French than the unround lip posture. Learners fairly consistently transferred the high tongue gesture from their L1 to their L2 for /i/ and /u/, although they were more consistent in producing French /i/ with the same tongue position as English /i/. The transfer patterns are summarized for each speaker in Table 10. The following section shows the results for learner mid vowel productions to further explore how vowel quality interacts with transfer patterns.
Summary of Transfer Patterns From English to French for Each Learner’s High Vowel Productions.
4.2 Mid vowels
4.2.1 Acoustic results
The acoustic vowel space of each native speaker’s French mid vowel productions are shown in Figures 20 and 21. Native speakers NF1 and NF2 produced French /ø/ in a different region of the F1-F2 space from French /e/ and /o/.
F1 and F2 values (Hz) of speaker NF1’s mid French vowel productions.
F1 and F2 values (Hz) of speaker NF2’s mid French vowel productions.
Acoustic vowel plots for learners FL2 and FL4 are shown in Figures 22 and 23 as representative of all learner results. All other acoustic vowel plots can be found in Supplement Appendix A.
F1 and F2 values (Hz) of speaker FL2’s mid vowel productions.
F1 and F2 values (Hz) of speaker FL4’s mid vowel productions.
As can be seen in the learner plots, there was more overlap with the French phone /ø/ and the round back vowels than seen for the native speakers. Similar to the native speakers, however, learners produced /ø/ in a different F1-F2 region than French /e/.
An MANOVA with F1 and F2 as the dependent variables show, as expected, that native speakers did not produce the mid round vowels /ø/ and /o/ in the same region of the F1-F2 space. The Pillai scores and associated p values for each speaker’s MANOVA comparing F1 and F2 for the mid round vowels are shown in Table 11.
Output of MANOVAs Evaluating the Difference in F1 and F2 Values of French Vowels /ø/ and /o/, by Speaker.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Table 11 shows that the native French speakers and most learners produced a difference in the F1-F2 space of the mid round vowels /ø/ and /o/. This is true of learners in Group 1 (who produced a contrast for the high round vowels) and learners in Group 2 (who did not produce a difference in the high round vowels). FL4 is the one speaker who did not acoustically distinguish /ø/ and /o/. This learner will be discussed further in the following sections.
Next, each learner’s French and English productions were compared. Two MANOVAs were run for each speaker, comparing F1 and F2 of French /o/ and English /o/, and French /ø/ and English /o/. Pillai scores and associated p values of the MANOVAs are presented, by speaker, in Table 12.
Output of MANOVAs Evaluating the Between-Language Difference in F1 and F2 of Mid Round Vowels, by Speaker.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Four out of six learners did not produce French and English /o/ with the same F1 and F2 values at the vocalic midpoint. Two learners (FL4 and FL6) did produce French and English /o/ in the same region of the F1-F2 space. For the new French phone /ø/, every learner except FL4 produced a significant difference in F1 and F2 from English /o/. Thus, FL4 is the only learner who produced French /o/, English /o/, and French /ø/ in the same region of the F1-F2 space. Recall from Section 4.1 that FL4 was the only learner who produced French /u/, English /u/, and French /y/ with the same F1 and F2 values.
Finally, F1 and F2 values of French and English front mid unround vowels are compared. Separate ANOVAs were run for each speaker comparing F1 and F2 for French and English /e/. Pillai scores from the MANOVA and p values are presented in Table 13. The results from the within-speaker MANOVAs show that no learners overlapped French /e/ and English /e/ in the F1-F2 space at the vocalic midpoint.
Output of MANOVAs Evaluating the Between-Language Difference in F1 and F2 Mid Unround Vowels, by Speaker.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
The following sections present the results concerning how learners articulate the contrast between French mid vowels and use L1 articulatory gestures for L2 productions.
4.2.2 Lip rounding results
First, the degree of lip opening for native French speakers’ French vowels /e/, /ø/ and /o/ is compared using two one-way ANOVAs. The purpose of this analysis is to show how native speakers produced the contrast between the French mid vowels, which will then be compared to how learners produced the contrast. Figures 24 and 25 show the area of lip opening for each vowel, produced by NF1 and NF2, respectively. Results from the one-way ANOVA show that NF1 produced a significant difference in lip opening between the three mid vowels, F(2, 39) = 45.11, p < .001***, and a post hoc Tukey HSD test shows that the round vowels /o/ and /ø/ differed significantly from /e/ but not from each other. The results for NF2 are similar: The three vowels differed in area of lip opening, F(2, 39) = 30.45, p < .001***, and a Tukey HSD test reveals that /o/ and /ø/ did not differ from each other, but differed significantly from /e/. Thus, for native French speakers, mid front and back round vowels had the same degree of lip rounding, and were more round than /e/.
Area of lip opening (mm2) for French mid vowels produced by native French speaker NF1.
Area of lip opening (mm2) for French mid vowels produced by native French speaker NF2.
Next, the degree of lip rounding for learner productions of the French mid vowels was compared to determine whether these speakers produced a target-like lip rounding contrast. A one-way ANOVA was run for each speaker, with area of lip opening as the dependent variable, compared for the French vowels /e/, /ø/, and /o/ (see Table 14, along with corresponding Tukey HSD tests, which show how each vowel pair differed in terms of lip rounding).
Output of Within-Speaker One-Way ANOVAs and Tukey HSD Tests Comparing Area of Lip Opening for the Mid French Vowels /e ø o/.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Using the native speakers as a baseline for comparison, all learners were target-like in producing French /o/ with significantly more lip rounding than French /e/. Most learners rounded the new front round mid vowel /ø/ significantly more than French /e/, with the exception of FL1. Three out of six learners (FL1, FL2, and FL4) rounded French /o/ significantly more than /ø/. The area of lip opening for learners FL4 and FL5 is shown in Figures 26 and 27, respectively, with FL4 producing the back round vowel more rounded than the front round vowel and FL5 producing the mid round vowels with similar degrees of lip rounding. All other lip rounding area plots can be found in Supplement Appendix B.
Area of lip opening (mm2) for French and English mid vowels produced by learner.
Area of lip opening (mm2) for French and English mid vowels produced by learner FL5.
Figures 26 and 27 also show that, similar to results for the high vowels, learners tended to round French vowels more than their English counterparts. French /e/ was more rounded for all learners than English /e/ and four of six learners rounded French /o/ more than English /o/. However, the front mid round vowel /ø/ was systematically less round than French and English /o/. A one-way ANOVA comparing lip opening area for all French and English vowels was run for each speaker, and the relevant findings from each Tukey HSD test are shown in Table 15.
Output of Within-Speaker One-Way ANOVAs and Tukey HSD Tests Comparing Area of Lip Opening for English and French Mid Vowels.
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Four out of six learners (FL1, FL3, FL4, and FL5) rounded French /e/ significantly more than English /e/ and no learners produced a significant difference in rounding between French and English /o/. Finally, only two learners (FL1 and FL2) produced a significant difference in rounding between French /ø/ and English /o/. Both of these learners had a tendency to unround /ø/.
Overall, learners tended to round the French mid /e/ more than English /e/, but did use the same lip rounding position for French and English /o/. Only half of the learners examined here were target-like in rounding the new French phone /ø/ to the same degree as French /o/, while the other learners tended to produce /ø/ with unrounded lips.
The next, and final, results section assesses each speaker’s tongue positions in productions of the French mid vowel contrasts, and English mid vowels.
4.2.3 Tongue position results
Two separate SSANOVAs were run for native speakers NF1 and NF2 measuring the overlap in tongue positions for the French mid vowels /e/, /ø/, and /o/ along the tongue contour. The SSANOVA plots for NF1 and NF2 are provided in Figures 28 and 29, respectively, and again, tongue tip is to the left of the figure.
SSANOVA of tongue position of the French mid vowels, produced by NF1.
SSANOVA of tongue position of the French mid vowels, produced by NF2.
Both of the native speakers produced a significant difference between the French mid vowels along the tongue contour. For both NF1 and NF2, /e/ is produced with the tongue further forward than /ø/, which is produced with the tongue further forward than /o/. Although both speakers produced some overlap between the front vowels /e/ and /ø/ along the tongue body, /e/ was significantly further forward along the majority of the tongue contour.
Learners’ tongue positions for the productions of the three French mid vowels /e/, /ø/, and /o/ were compared next. Recall from Section 4.2.1 that all learners except FL4 produced French /ø/ and /o/ in different regions of the F1-F2 space. For the tongue position analysis, SSANOVAs for each learner were run, comparing the position at the midpoint of each French vowel. SSANOVA plots for learners FL1 and FL3 are shown in Figures 30 and 31 as representative of the learners, and all other SSANOVA plots can be found in Supplement Appendix C.
SSANOVA of tongue position of the French mid vowels, produced by FL1.
SSANOVA of tongue position of the French mid vowels, produced by FL3.
Half of the learners (FL2, FL3, and FL6) produced the contrast between the mid vowels similarly to how the native speakers produced the contrast, in that /e/ was more fronted than /ø/, which was more fronted than /o/ (see Figure 31). The other half of learners produced /o/ with an anterior constriction location that overlaps with /ø/ along the tongue body and tongue root (see Figure 30 for learner FL1 results).
A second set of SSANOVAs were run to compare learners’ tongue positions of their L2 French mid vowels with their L1 English mid vowels. Two SSANOVAs were run for each speaker: The first SSANOVA compared learner tongue positions for the English front vowel /e/, the French front vowel /e/, and the French front round vowel /ø/; the second SSANOVA compared French /o/ and French /ø/ with English /o/. Within-speaker SSANOVA plots for learners FL3 and FL6 are presented in Figures 32–35, and once again, all other learner plots can be found in Supplement Appendix C.
SSANOVA of tongue position of the French and English mid front vowels, produced by FL3.
SSANOVA of tongue position of the French and English mid round vowels, produced by FL3.
SSANOVA of tongue position of the French and English mid front vowels, produced by FL6.
SSANOVA of tongue position of the French and English mid round vowels, produced by FL6.
First, examining how learners produced L1 and L2 /e/, recall from Section 4.2.1 that no learners in the present study produced French and English /e/ in the same region of the F1-F2 space. However, learners such as FL3 (Figure 32) did not produce French and English /e/ with different tongue positions along most of the tongue contour. This contrasts with learner FL6 (shown in Figure 34) who did produce French and English /e/ with significantly different tongue positions along the tongue front.
For the similar back vowels French /o/ and English /o/, recall from Section 4.2.1 that FL4 and FL6 were the only learners who produced these vowels in the same regions of the F1-F2 space. Although the SSANOVA plots do not show perfect overlap between French and English /o/ for all speakers, L1 English and L2 French mid back vowels were produced with a similar tongue position for most learners. Learner FL6 (seen in Figure 35) was the only learner who produced a significant difference between French and English /o/ along much of the tongue contour. This result will be discussed further in Section 5, in light of the acoustic finding that FL6 produced French and English /o/ in the same region of the F1-F2 space.
Finally, learner tongue positions for the new French phone /ø/ were compared with English /e/ and /o/. Learners tended to produce a significant difference in tongue position between French /ø/ and both English vowels /e/ and /o/. Two learners produced French /ø/ with no significant difference from English /o/ along the tongue root (FL4 and FL5, seen in Supplement Appendix C). However, most learners produced French /ø/ with a different tongue position than their English vowels.
4.2.4 Mid vowel results summary
This section compared the productions of the French and English mid vowels for each learner, and found a range of individual differences in productions. Each learner’s transfer patterns between English and French are summarized in Table 16.
Summary of Transfer Patterns From English to French for Each Learner’s Mid Vowel Productions.
Overall, many learners did not transfer L1 acoustic categories to produce any of the French mid vowels. All of the learners produced French /o/ with the same degree of lip rounding as English /o/, and the majority of learners produced French /ø/ with the same degree of lip rounding as English /o/. Many learners did not produce French /e/ with the same lip rounding as English /e/. For tongue position, learners tended to transfer the tongue body position from English /o/ to French /o/, and many produced French /e/ and English /e/ with similar positions. Most learners did not produce French /ø/ with a similar tongue position to English vowels.
5 Discussion
The results of this study show a wide variety of learner strategies used to produce the contrast between the French high vowels /i y u/ and the French mid vowels /e ø o/. Section 1 outlined two research questions, which we return to here.
5.1 How is the French round vowel contrast articulated by L2 learners?
The first research question asked how learners produce the acoustic and articulatory contrast between the French front round and back round vowels. In their articulation of the French round vowels, Group 1 learners were similar to native speakers in the lip rounding of /y/ and /u/, but did not overlap the tongue position of French /i/ and /y/ as the native speakers did. Group 1 learners also tended to front the tongue position of French /u/. Group 2 speakers did not produce the contrast between French /y/-/u/ in the F1-F2 region, but similar to Group 1 learners, were target-like in terms of lip rounding of the French high round vowels. Group 2 learners tended to produce French /y/ and /u/ with overlapped tongue position, notably producing /u/ with an anterior constriction location. This result suggests that the apparent merged French /y/-/u/ category for Group 2 learners is caused by producing French /u/ as English /u/, which had a more anterior tongue position. This may be because of the phonetic nature of English /u/ that has fronted variants in coronal contexts similar to [y]. Furthermore, /y/ is produced by Group 2 speakers with a more posterior tongue body position compared with /i/, which also leads to the apparent acoustic merger between /y/ and /u/. It can thus be concluded that learners who did not produce an acoustic contrast between French /y/ and /u/ did so because of a non-target-like tongue body position, not lip rounding.
For the mid vowels, only one learner did not accurately produce the contrast between the front round vowel /ø/ and French /o/ (FL4). In examining whether FL4’s acoustic merger of /ø/ and /o/ was caused by lip unrounding or tongue fronting, we find different results from the high vowels: it does not appear that all learners were target-like in their lip position for the French mid round vowels. Half of the learners here (FL1, FL2, and FL4) rounded French /ø/ less than French /o/, unlike the native speaker controls. In addition, these three learners tended to produce French /ø/ and French /o/ with overlapped tongue position. This shows that for FL1, FL2, and FL4, the contrast between the mid round vowels was one of rounding, rather than tongue position. Following from the results that FL1, FL2, and FL4 produced /ø/ with less rounded lips than French /o/, it is likely that learners were mapping this new phone to their L1 English schwa category.
The learners in this study were more likely to produce high front round vowels with lip rounding comparable with back round vowels than they were to produce mid front round vowels with lip rounding comparable with back round vowels. This may be because of the way in which the speakers were perceptually mapping the front round vowels to L1 categories. Previous research examining how naïve L1 English listeners map French front round vowels to back round vowels has found that listeners perceive French /y/ and /ø/ to be poor exemplars of back round vowels, but did not find a difference in perceptual assimilation patterns between mid and high vowels (Strange et al., 2009). This could be because learners were mapping French /ø/ to English /ǝ/, which is unround. However, because there is no phonemic high central vowel in English, learners either created a new phonological category for /y/ or produced French /y/ as English /u/, and thus we see that learners produced /y/ as rounded.
A further note must be made here regarding learner FL4. The within-speaker results show considerable individual variation, which highlights the complicated nature of learners’ interlanguage grammars. FL4 was the only speaker who had acoustically merged French /y/-/u/ and /ø/-/o/. This learner also produced French /u/, English /u/, and French /y/ with the same acoustic values, and French /o/, English /o/, and French /ø/ with the same acoustic values. FL4 also consistently overlapped the tongue position for the mid front and back round vowels, and high front and back round vowels. These phonetic results suggest that this learner had not formed new phonological categories for the round vowels in French production, but rather likely perceptually mapped French round vowels to the English round vowels of similar height. Both the acoustic and the ultrasound production results are consistent with this finding.
5.2 Do learners use L1 articulatory gestures or acoustic categories to produce L2 phones?
The second research question this experiment investigated is whether learners use L1 articulatory gestures or acoustic categories to produce L2 phones. Based on the present results, the nature of category transfer depends on vowel quality, and the following generalization can be made about category transfer for L2 vowels. Learners tended to transfer L1 acoustic categories for French /i/ and /u/. Although only half of learners produced English /u/ and French /u/ in the same region of the F1-F2 space, other learners produced English /u/ and French /y/ in the same region of the F1-F2 space, thus leading to the conclusion that learners did, in fact, transfer L1 English /u/ acoustic values to French. Again, this variation between learners may be due to the fact that English /u/ can be produced as [y] in some contexts, and this phonetic property of /u/ causes some learners to perceive French /y/ as phonetically similar to English /u/, thus assimilating /y/ to the English /u/ phonological category. Learners tended to transfer lip rounding gestures for the round vowels /u/ and /o/ to the French vowels /u/, /y/, and /o/. In contrast, learners tended to transfer tongue positions for /i/, /u/, /o/, and sometimes /e/ to similar French vowels. Learners who did not transfer L1 tongue body gestures to produce French /e/ may perceive the phonetic differences between the monophthongal French /e/ and diphthongal English /e/, thus creating a new phonetic category for French /e/. Learners did not transfer L1 tongue positions for the high front round vowel /y/, and very few transferred L1 tongue positions for French /ø/. Because both tongue fronting and tongue backing were consistently transferred, the tongue body gesture for front and back vowels may be defined by the “extreme” or peripheral vowels in the articulatory space. However, learners were more likely to transfer tongue position for L2 phones that are similar to L1 phones, rather than for new phones.
The most clear finding from these results is that there appears to be a rounding gesture that is transferred from the L1 to L2. The visual cue of lip rounding likely influences this consistent pattern. Learners are producing a difference in lip position between round and unround vowels, but the lack of transfer of a lip posture from the L1 to the L2 for only the unround vowels indicates that unround vowels do not have an unround target in their interlanguage. It is, therefore, proposed that unround vowels are unspecified for rounding, whereas rounded vowels are specified with a lip rounding gesture.
A secondary finding of this study is that French vowels tended to be rounder than English vowels. These learners may have acquired different “articulatory settings” (Gick et al., 2004; Honikman, 1964) for French, in which the lips are more rounded when at rest. This explanation is compatible with the proposal that unround vowels are unspecified for rounding because if the unround vowels were produced with active spreading, we would expect French and English unround vowels to be produced with similar lip postures. If French has a more rounded neutral position, it would follow that the vowels unspecified for rounding would be produced as more rounded in French than English. Future research could test the predictions that stem from the underspecification of unround vowels, and whether French has a rounding resting posture that leads unround vowels to be produced as more rounded than English vowels.
Finally, it should be noted again that the lip rounding measurements were taken at the maximum point of constriction, while the ultrasound and acoustic measurements were made at the vocalic midpoint. Although both measurements are anticipated to correspond to the articulatory targets of vowels, there may be differences across the measurements that are an artifact of this analysis. Taking the measurement at the maximum point of constriction may show clearer transfer patterns, as we see with lip rounding. The acoustic midpoint was selected for the ultrasound images to make processing the tongue splines take a reasonable amount of time, and because it can be difficult to identify the maximum point of constriction of the tongue. If measuring at the maximum point of constriction is a more accurate representation of the articulatory goal of a speech sound, however, we may expect learners to more consistently transfer the tongue body position to the similar L2 phones. Future research is suggested to more directly compare these two measurement points.
5.3 Representation of vowel targets
The overarching research goal of this study was to examine the targets of vowels in L2 speech production. This was done by testing the hypotheses of theories that differ in terms of how speech sounds are represented. The findings showed that the targets of the vowels depend on vowel quality, which is not necessarily predicted by either theory.
One major limitation of the SLM in light of the findings presented here is that it argues that acoustic-phonetic categories are transferred for perceptually similar L1 and L2 phones (Flege, 1987). This experiment found that acoustic-phonetic categories were only transferred for the high vowels, and rarely for mid vowels. There is a possibility that the learners in the present study were not mapping the mid vowels to their L1 vowel phonological categories because they perceived the phonetic difference between English mid vowels /e/ and /o/ (which are typically produced as diphthongs) and the European French mid vowels (which are produced as monophthongs in the variety of French to which they were exposed). However, the articulatory findings of the present study did show that many L1 English speakers were producing the L2 French mid vowels with the same tongue position as their English vowels even when the acoustic results did not show transfer. This was the case for /e/ for FL1 (and marginally for FL3, FL4, and FL5), and for /o/ for FL1, FL3, FL5 (and marginally FL2).
One possibility for the mismatch between the transfer patterns of the high vowels and the mid vowels may be that English mid vowels tend to be more diphthongal than English high vowels (see Nearey, 2012 for a discussion on vowel inherent spectral change in North American English vowels). Learners initially classify French /i e u o/ as their L1 English phonological categories, but as they are exposed to more instances of /e/ and /o/, the phonetic differences between French and English /e/ and /o/ cause learners to create new phonetic categories for French /e/ and /o/. However, if this were the case, we would expect to see that learners do not use their L1 categories to produce both /e/ and /o/, but learners did produce French /o/ with the same tongue body gesture as English /o/, and some also produced French /e/ with a similar tongue body gesture as English /e/. This lends support for the claim that these learners were mapping the French mid vowels to their L1 counterparts. SLM fails to predict the mismatch between mid and high vowel category transfer.
PAM-L2 similarly predicts that if two L2 phones are mapped to the same L1 category, the two phones will be considered equivalent to the L1 phone, and there will be no further perceptual learning (Best & Tyler, 2007). Because PAM-L2 considers the articulatory gesture to be the unit of perceptual analysis, it can thus be assumed that L2 phones that are perceptually mapped to L1 categories will be produced with L1 articulatory gestures. Assuming learners perceptually assimilate the French categories /i e o/ to their English counterparts (and leaving aside the complicated mapping of French /u/ to English /u/), we expected both lip rounding and tongue body gestures to be transferred from English to French. We did see transfer of the tongue body gesture for /i e o/ for many learners but only saw transfer of the lip rounding gesture from English to French /u/ and /o/. We did not see transfer of lip posture for unround vowels. PAM-L2 does not make explicit predictions about the articulatory gestures transferred in L2 productions, and therefore does not predict the mismatch between transfer of round and unround lip posture.
Despite the findings that many learners transferred rounding and tongue body gestures from their L1 to produce L2 phones, there is evidence that learners also transferred acoustic categories for the high vowels. Perhaps the more peripheral vowels have a clearer acoustic target that can be transferred between their L1 and L2. Mid vowels do not seem to have as obvious acoustic transfer. It appears that category transfer depends on vowel quality. The apparent difference in phonetic transfer of mid and high vowels is similar to the cardinal vowels proposed by Jones (1956). The cardinal vowel space is defined by the “extreme” vowels. Although Jones (1956) originally proposed the vowel space in articulatory terms, there is a clear connection to the acoustic space. From the present data, it appears that the vowels that are on the periphery of the acoustic space (in this case, /i/ and /u/) have an acoustic target, whereas the mid vowels do not. The mid vowels may perhaps be defined by their relative distance from the high vowels, rather than having a target themselves. Asymmetries between “peripheral” and “non-peripheral” vowels have also been found in speech perception, and it is argued that more peripheral vowels /i/, /u/, and /a/ act as natural referent vowels and shape a speaker’s perceptual system (Polka & Bohn, 2011).
One limitation about this finding from the present study is that no low vowels were included in the analysis. To further test the hypothesis that the peripheral vowels have an acoustic target while the mid and central vowels are defined in relation to these vowels, an analysis of low vowels is suggested. Despite this limitation, there is a difference between the phonetic transfer of acoustic categories for the peripheral vowels and the mid vowels that is reminiscent of the cardinal vowel space proposed by Jones (1956).
The articulatory data show that learners did tend to transfer both front and back tongue body positions from their L1 to L2 for similar vowels and produced contrastive front and back round vowels articulatorily. This serves as preliminary evidence for tongue body gestural targets. Learners did not appear to transfer articulatory gestures to produce new non-native vowels, as evidenced by the lack of tongue body transfer for the learners who had created new categories for French /y/ and /ø/. There was also evidence for a lip rounding target, but not a lip unrounding target.
5.4 Use of ultrasound for L2 research
This study has made novel use of ultrasound tongue imaging and lip rounding data to investigate individual paths to L2 category formation. The use of articulatory data showed transfer patterns that were obscured when only using acoustic data, which underlines the importance of considering both articulatory and acoustic data in studies to inform theories of L2 category formation. For example, Group 1 learners acoustically produced the contrast between French /y/ and French /u/, which may lead to the conclusion that these learners created a new phonological category for /y/ as a front vowel. However, the tongue position results showed that these learners did not produce /y/ with the fronted tongue position of /i/, which suggests that these learners created a new category that perhaps did not have the same front tongue body gesture as /i/. Rather for many learners, it appears that /y/ had a central tongue body gesture. Furthermore, the acoustic results showed a lack of target transfer for vowels where the tongue body target appears to be the same. For example, learners FL1, FL2, FL3, and FL5 did not produce French and English /o/ with the same acoustic values, but did use the same tongue body gesture for these vowels. This is evidence that speakers are transferring articulatory targets to produce these vowels.
The use of ultrasound, and articulatory methods more generally, contribute new insights into how learners use their L1 categories to produce L2 speech sounds. Furthermore, L2 articulatory data contribute to our understanding of targets in speech production.
5.5 Future directions
The results of this experiment show a mismatch between articulation and acoustics in certain contexts that merits further exploration. As mentioned above, there were instances where speakers used L1 articulatory gestures to produce non-native vowels but did not use L1 acoustic categories to produce those same vowels (such as French and English /o/). This study described acoustic-phonetic categories in the F1-F2 space, and described articulatory gestures by tongue body positions and lip posture. This is because this study was designed to examine the articulatory gestures that are assumed to be relevant in creating the phonological contrast between the target vowels (namely, tongue backness/height and lip rounding). There are other articulatory movements, however, that can cause the apparent mismatch between acoustic and articulation. For example, nasality can cause changes to F1, and tongue curvature can cause changes to F2. Future studies gathering more articulatory information may explain why transfer of articulatory gestures does not necessarily imply acoustic-phonetic category transfer. This would help to understand the phonetic strategies learners are using to produce L2 vowels.
Second, the individual variation in the production results may be related to different perceptual mapping between speakers. It is hypothesized that learners who produced French /y/ as English /u/ perceived these two as phonetically similar and therefore assimilated French /y/ to English /u/. Perception data would bring more light to the differences between speakers and would test this hypothesis empirically. Furthermore, perception data could help to answer whether learners who do not produce French /ø/ with similar lip rounding to English /o/ are perceptually assimilating French /ø/ to English /ǝ/. Future studies that examine the relationship between articulation and perception are likely to make fruitful contributions to our understanding of category formation and individual variation.
Finally, this study found that speakers consistently transferred lip rounding postures from their L1 to their L2 but did not produce L1 and L2 unround vowels with similar lip rounding postures. This evidence is suggestive of a lip rounding gestural target, while unround vowels may be unspecified for rounding. This proposal has implications for L2 speech production research, but also features theories more generally, and can be tested with a wide variety of future experiments. For example, if unround vowels are unspecified for rounding, we would expect to see a lack of phonological processes that depend on a [-round] feature. Typological studies that test this assumption are suggested.
6 Conclusion
This experiment collected articulatory and acoustic data from L1 English-L2 French vowel productions, in an effort to investigate the nature of vowel targets in second-language speech. Considerable between-speaker variation was attested, revealing the complex nature of L2 category formation that is not necessarily apparent with acoustic data alone. Nevertheless, several clear patterns emerged across the learners examined here. First, learners who were not acoustically target-like in producing the contrast between French front round and back round vowels tended to front L2 back vowels, resulting in an apparent merger. Second, learners were consistent in transferring lip rounding for round vowels from their L1 to an L2, but were less likely to transfer unround lip positions. Learners also tended to transfer tongue positions for front and back vowels for “similar” L1 and L2 phones, but not for “new” phones. Finally, learners were more likely to transfer acoustic categories from their L1 to their L2 for high than mid vowels. This experiment finds evidence that more peripheral vowels have clear acoustic and articulatory targets in L2 speech production, while less peripheral vowels do not.
Supplemental Material
sj-pdf-1-las-10.1177_00238309241259748 – Supplemental material for The Articulatory and Acoustic Representation of Second-Language French Vowels
Supplemental material, sj-pdf-1-las-10.1177_00238309241259748 for The Articulatory and Acoustic Representation of Second-Language French Vowels by Madeleine Oakley in Language and Speech
Footnotes
Appendix A
Acknowledgements
I would like to thank all the participants who contributed their data to this project. Thank you to Elizabeth Zsiga, Jennifer Nycz, Lisa Davidson, and Jeff Mielke for their feedback on this work.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is partially funded by Language Learning Dissertation Grant.
Supplemental material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
