Skip to navigation – Site map

HomeAnglophonia30Voice assimilation of morphemic -...

Voice assimilation of morphemic -s in the L2 English of L1 French, L1 Italian and L1 Spanish learners

Leonardo Contreras Roa, Paolo Mairano, Marc Capliez and Caroline Bouzon

Abstracts

This study investigates the progressive voice assimilation rule in word-final morphemic -s in L2 English. We have analyzed data from the IPCE-IPAC corpus of learner oral productions, by measuring periodicity for all learner realizations of morphemic -s. We have compared three groups, namely 15 L1 French learners, 15 L1 Italian learners, and 10 L1 Spanish learners. Given the different distributions and status of [s] and [z] in the participants’ L1s and based on SLM (Speech Learning Model) and MDH (Markedness Differential Hypothesis), we hypothesized that L1 French learners and L1 Italian learners would find it easier than L1 Spanish learners to reproduce the outcome of the voice assimilation rule, but our predictions are only partially confirmed by the results: L1 French learners (who have /s/ and /z/ in their L1 as phonemes occurring in word-final position) are indeed the most successful in producing the expected patterns of periodicity. However, L1 Spanish learners outperform L1 Italian learners in producing periodicity in voiced contexts. We compare these results with our previous analysis of voicing for non-morphemic /s/ and /z/ by the same speakers, and we discuss our findings in relation to the markedness of these two sounds. We propose that the /s/ ~ /z/ voicing opposition may constitute an exception to the markedness hierarchy of voice contrasts for obstruents (word-initial < word-medial < word-final), whereby the word-final position is not more marked than the other positions. Additionally, if taken together, the results of this and our previous study reveal differences in the voicing of morphemic vs non-morphemic realizations for sibilants in L2 English (similar to L1 English). This may have repercussions on models of L2 phonology acquisition, which do not presently take into account an interaction between L2 sounds and their morphemic status.

Top of page

Author's notes

We would like to thank all learners who agreed to take part in the IPCE-IPAC corpus. Additionally, we are grateful to Valentina De Iacovo (University of Turin) for helping during the recruitment and recording of Italian participants, and to Audrey Gros-Bonfiglioli and Ioana Trifu-Dejeu (University of Paris 10 Nanterre) for sharing their data of 5 French participants with us.

Full text

1. Introduction

1/s/ and /z/ are a pair of frequent phonemes in English (Cruttenden; Mines et al.) opposed by [+/-voice] with a high functional load (Munro and Derwing; Brown; Catford). When occurring at the end of a word, these phonemes can be classified into two main categories: they can be non-morphemic, i.e. they are part of a stem (e.g. bus, class, appearance), or they can be morphemic, i.e. they are realizations of morphemes expressing the grammatical features of the plural (e.g. cars), genitive (John’s) or genitive plural in nouns (e.g. students), the 3rd person singular in verbs (e.g. eats), or clitic forms of has and is. When non-morphemic, these phonemes can appear in various phonotactic contexts and constitute a phonological opposition producing numerous minimal pairs (e.g. bus - buzz). However, in the case of morphemic -s, these two sounds are usually regarded as one single phoneme /z/ with two allophones, [z] and [s], the former of which can be subject to a vowel insertion if preceded by a sibilant: [ɪz] (Br. Eng.) or [əz] (Am. Eng.) (Finegan: 123), i.e. a total of three possible allomorphs. The choice between [z] or [s] is determined by a progressive assimilation rule, and therefore depends on the preceding segment. For instance, it will be pronounced as [s] in meats due to the preceding segment (/t/) being voiceless, and as [z] in meads due to the preceding segment (/d/) being voiced. These three forms of morphemic -s have been traditionally treated similarly in studies of phonology and morphology (Chomsky and Halle).

2The production of these two sounds regardless of their morphological status, as well as their differentiation through [+/-voice] has been shown to be a problem for learners of L2 English, both at a production and perception level (Flege; Fullana and Mora; Barreiro; Tyler et al., Mairano et al. 2019). In this paper, we analyze the voicing patterns of morphemic -s in English produced by learners whose native language (L1) is French, Italian, or Spanish, with data from the IPCE-IPAC corpus (Herry-Bénit et al.). To our knowledge, very few studies have focused on the differences in voicing patterns of fricatives from an empirical and comparative perspective with learners of different L1s. In this article, we will not deal with the generic /s/ ~ /z/ contrast, since this is analyzed in a companion paper by Mairano et al. (forthcoming). Instead, we focus specifically on morphemic -s, where voicing is determined by an assimilation rule. Part of the interest in investigating morphemic -s comes from the fact that voiced sounds are universally more marked in word-final position than in word-medial or word-initial position (Eckman 2008), and two of the native languages under scrutiny in this paper do not allow for /z/ in this position. Exploring the behavior of these sounds in word-final position can therefore serve as a basis to better understanding the acquisition of L2 English phonology.

3In the next sections, we shall present the characteristic of final /s/ and /z/ in English and in the native languages of our learners. This contrastive analysis will allow us to formulate hypotheses based on models of L2 phonology acquisition., which will be tested on L2 English productions of our L1 French, L1 Italian and L1 Spanish learners, as described in sections 3 and 4. A final discussion on our findings will close the article.

2. Final [s] and [z]

2.1 Morphemic -s voicing in English

4Morphemic -s has three allomorphs in English, namely /z/, /s/, and /ɪz/ or /əz/. While the basic form is /z/, the other two allomorphs are determined by a progressive voice assimilation rule after voiceless consonants, and by a schwa insertion rule after sibilants (Finegan 12324). This can be summarized as follows:

  • 1 Cruttenden explains that, in final position, the fricatives /v, ð, z, ʒ/, although voiceless, re (...)

5From a phonetic point of view, the voicing of morphemic -s is not only governed by the phonological progressive assimilation rule, but can also undergo partial or total regressive assimilation to the following segment. In a study concerning fricatives, Stevens et al. found that glottal vibration – the articulatory correlate of voicing – in syllable-final voiced fricatives tends to be more influenced by the voicing of following consonants rather than that of preceding ones, i.e. regressive assimilation due to coarticulation can be stronger than phonological progressive assimilation. This confirms previous acoustic findings (e.g. Haggard) as well as phonetic descriptions (Cruttenden: 19394) which report that voiced fricatives can be completely voiceless in final position1. It has also been observed that voiced fricatives show a high degree of periodicity, one of the acoustic correlates of voicing, in intervocalic position (Hardcastle and Clark; Cruttenden). Conversely, Stevens et al. observed that utterance-final voiceless fricatives are likely to present a higher proportion of periodicity as well as a higher duration than voiceless fricatives in other contexts.

6The main acoustic correlate of voicing is periodicity, but duration has also been identified as a secondary cue to voicing. Previous studies have analyzed the phonetic voicing patterns of final /s/ and /z/, regardless of their morphemic status. For example, duration has been found to be higher for voiceless fricatives than for voiced ones in prepausal position (Crystal and House; Jongman et al.). Additionally, this devoicing of /z/ increases gradually along with the metric domain, as it was observed in an acoustic and articulatory study (B. L. Smith), which found that the periodicity of /z/ in American English decreases gradually, while duration increases gradually. Thus, syllable-final devoicing is lower than word-final devoicing, which, in turn, is lower than utterance-final devoicing. Other contextual features such as word stress are less likely to be predictors of word-final devoicing of /z/. Secondary articulatory features like differences in airflow have also been identified for /z/ from /s/, and are used by English speakers to maintain the distinction between these two sounds (C. L. Smith). Finally, another important cue to voicing is the length of the vowel preceding final obstruent pairs: vowels preceding voiced obstruents are longer than those preceding voiceless fricatives in word-final position (Raphael; Peterson and Lehiste; House).

7Recent studies have revealed durational differences between morphemic and non-morphemic /s/ and /z/ in pairs traditionally considered as homophones (e.g. non-morphemic a lapse vs. morphemic two lap+s) (Walsh and Parker; Plag et al.; Yung Song et al.). However, results are somewhat contradictory: while Plag et al. and Yung Song et al. observed longer durations in non-morphemic /s/ and /z/, Walsh & Parker observed the opposite (but this study was carried out on just 10 speakers, did not take into account phonetic covariates, and did not perform any inferential statistical analyses). Durational differences between morphemic and non-morphemic -s seemed to be acquired early by L1 English children: Yung Song et al. showed that children produce longer frication to differentiate morphemic vs non-morphemic -s. Given that native English speakers seem not to produce morphemic and non-morphemic -s in the same way phonetically, similar considerations may hold for L2 English learners. For this reason, we inspected realizations of /s/ and /z/ in two separate analyses; the first analysis looked at the /s/ ~ /z/ contrast in non-morphemic -s, where it is fully contrastive, and is presented in a companion study (Mairano et al., forthcoming); the second analysis, presented here, looks at /s/ and /z/ exclusively for morphemic -s, where voicing is determined by the previous segment.

2.2 Final [s] and [z] in French, Italian and Spanish

8In this section, we sketch the distribution and the behavior of /s/ and /z/ in the L1s of the three groups of learners that we have analyzed. This is done to provide a starting point for an analysis that aims to determine the presence of transfer – broadly defined as the influence of a learner’s L1 features – in their L2 production. The Speech Learning Model (SLM - James E. Flege) states that L2 phonemes are likely to be easily learned when they are absent from the learner’s L1 phoneme inventory or when they are dissimilar from the learner’s L1 phonemes. Conversely, L2 phonemes are less likely to be easily learned when they are similar to one or more of the learner’s L1 phonemes. The Perceptual Assimilation Model (PAM - Best) approaches the same idea from a perceptual point of view and states that learners perceive L2 segments by establishing gestural “constellations” (p. 193) throughout the “universal phonetic domain” (p. 191) between non-native sounds and the sounds of their L1. However, although the same articulatory gestures used to produce phonemes might be present in both the learner’s L1 and L2, the phonological role of these sounds is not necessarily the same in both languages, as we will illustrate hereafter. Depending on the degree of proximity of these gestures and the role that they play in each language, learners might (1) assimilate new sounds to one of their native categories, (2) assimilate them as uncategorizable speech sounds, or (3) assimilate them as non-speech sounds (p. 195).

9Although these two models alone can be used to make predictions about the difficulty of acquisition of an L2 phoneme based on its degree of difference with a learner's L1 phonemes, including their degree of markedness can lead to more precise hypotheses. By markedness, we will refer to the degree of distribution of a structure within a language (an unmarked structure being more widely distributed in a language than a marked one) or across languages (an unmarked structure being more widely distributed in the world’s languages). It has been established that voice contrasts in coda position, and therefore in word-final position, are typologically marked, i.e. universally marked cross-linguistically (Greenberg; Moulton). In particular, voiced obstruents are marked with respect to voiceless obstruents: if a language has voicing contrasts for obstruents in coda position, it will necessarily have them in word-initial and word-medial positions too, but the opposite does not apply (Eckman 97). We therefore expect voiced realizations for morphemic -s to be particularly challenging for our learners, given that morphemic -s is by definition word-final. Of course, we expect the difficulty to be higher for learners whose L1 does not have /z/ as a phoneme, or as an allophone.

10In order to assess the degree of difficulty for L1 French, L1 Italian and L1 Spanish learners to produce word-final morphemic -s in English, we will analyze the presence of these phonemes in these languages following three criteria: (1) the presence or absence of these sounds in the learner's L1, (2) the presence or absence of these sounds in word-final position in the learner's L1, (3) their phonological status in word-final position (phonemic or allophonic), and (4) their degree of relative markedness. The study of the behavior of these phonemes in the native systems of our three groups of learners will ultimately serve as a way of formulating hypotheses to guide our analyses.

2.2.1 Final [s] and [z] in French

11Both [s] and [z] can appear at a syllable- and word-final position in French and maintain their phonemic status. They can occur either as part of a stem (e.g. masse /mas/, bêtise /bɛtiz/) or as an inflectional morpheme (e.g. marking plurality as in les enfants /le zɑ̃fɑ̃/, de grands enfants /də gʁɑ̃ zɑ̃fɑ̃/, des nouvelles inquiétantes /de nuvɛl zɛ̃kjetɑ̃t/ or, more rarely, signaling subjunctivity qu’il paraisse /kil paʁɛs/, que je finisse /kə ʒə finis/). Some minimal pairs that illustrate this contrast in French: il hausse [ilos] / il ose [iloz], il laisse [illɛs] / il lèse [illɛz], ma laisse [malɛs] / malaise [malɛz]. We therefore predict that French learners are sensitive to the voicing of these two phonemes, including at a word-final position.

12The realization of final morphemic (more particularly: inflectional) /z/ (but not /s/) is common in French, often as a marker of plurality. However, its realization is subject to external sandhi rules during connected speech, a phenomenon often referred to as liaison. When liaison occurs, /z/ surfaces after the word boundary when the following word begins with V, e.g. les camions /le#kamjɔ̃/ vs. les amis /le#zami/. Even though the /z/ is resyllabified into the beginning of the following word as represented above, we will consider it as morphologically belonging to the previous word and therefore as a case of final /z/. Some cases of /z/ sandhi are categorical and therefore obligatory, especially between determiners and nouns, verbs and clitics, and in cases of suffixes expressing person or number (Bybee 2005; Rosoff; Bybee 2001). Other cases are variable and might depend on register or on the syntactic cohesion of a phrase. Some authors even argue that liaison is disappearing and that only the most frequently occurring contexts are maintained (Bybee2001; Ågren; Delattre).

13Regardless of the inflectional or non-inflectional status of /s/ and /z/, regressive voice assimilation of these and other consonants is attested in French across word boundaries (Hallé and Adda-Decker 2007; Hallé and Adda-Decker 2011). According to this experimental study, assimilations appear to be most likely in [+fricative]#[+fricative] and [+fricative]#[+stop] contexts.

2.2.2 Final [s] and [z] in Italian

14Italian is characterized by having predominantly CV-type or open syllables. The coda position undergoes severe constraints (Bertinetto and Loporcaro), where normally only /l, r, n, s/ can be found. Some exceptions apply for loanwords. The same constraints apply for word-final position, although /s/ occurs exceptionally and mostly in loanwords, non-morphemically (e.g. lapis, autobus). /z/ is phonologically absent in word-final position, and its behavior in other positions is subject to geographic variation. In Standard Italian, /s/ and /z/ are contrastive word-medially (e.g. chiese /ˈkjɛse/ ‘asked’ vs chiese /ˈkjɛze/ ‘churches’), but this opposition is neutralized in many regional varieties of Italian. In most Northern varieties, [z] is an allophone of /s/ occurring before a voiced consonant or between two vocalic segments (e.g. chiese /ˈkjɛse/ [ˈkjɛze] for both words mentioned above). However, voicing contributes to enhance the /s/ ~ /sː/ gemination contrast in Northern Italian, whereby casa (/ˈkasa/ [ˈkaːza], meaning ‘home’) and cassa (/ˈkasːa/ [ˈkasːa], meaning ‘cashier’) are phonetically distinct by duration and voice.

15Regressive assimilation applies intervocalically and across morphemes, but not across phonological word boundaries, e.g. dis[z]+dire but lapis[s]#nero (Nespor: 17274; Nespor and Vogel: 12529). This type of assimilation is obligatory and categorical, unlike Spanish where regressive assimilation is variable and gradient and also applies across word boundaries (see below).

16In loanwords containing final obstruents, Italian tends to add an epenthetic final or paragogic vowel, e.g. weekend[ə] (Broniś; Domokos). However, epenthesis is considerably less frequent with final /s/ (Nespor: 17879; Krämer: 153), especially in non-oxytonic words (Broniś). In this sense, some authors suggest that /s/ behaves like a sonorant in word-final position (Huszthy; Baroni).

17One study that has compared the progressive assimilation of word-final /z/ in English with the regressive assimilation of word-initial /s/ in Italian as a means of studying transfer in Italian learner’s English is Bosisio (2010). The author reports most of the learners she studied – third-year university students from Northern Italy specializing in English – systematically devoiced morphemic and non-morphemic word-final /z/, either partially or fully (p. 183). She concludes that learners' low performance at producing native-like assimilation patterns of final morphemic -s in English may be due at least in part to the representation of both sounds with the grapheme <s>, as well as to the relatively small threat to comprehension.

2.2.3 Final [s] and [z] in Spanish

18While /s/ is a frequent phoneme occurring in all syllabic positions in Spanish, [z] does not have a phonemic status in this language. It only occurs as an allophone of /s/ in a syllable-final position before a voiced consonant due to regressive assimilation of voice. This can happen both within and across word boundaries (Hooper; Navarro):

s [+voice] / ____ (#) C[+voice]

Thus, the following cases are possible:

desde: /desðe/ [dezðe]
las vacas: /las βakas/ [laz#βakas]
dos huesos: /dos wesos/ [doz#wesos]

19In most varieties, voicing does not occur before vowels, despite the fact that the latter have a higher proportion of periodicity than voiced consonants (but see Bradley 2005 for Highland Ecuadorian Spanish). Furthermore, [z] never appears in onset position in Spanish.

20The regressive voicing of /s/ is far from being categorical, and the phenomenon was said to be “stylistically determined, gradient, and variable” (Bradley and Delforge: 22) in an optional complementary distribution. Some authors attribute it to articulatory tension (Torreblanca; Whitley; Navarro), since /s/ tends to remain voiceless in careful, tense speech and its voicing appears in presence of a more relaxed articulation and in colloquial speech.

21It should be noted that /s/ is subject to extensive variation in Spanish, and not only in terms of voicing. Some varieties (e.g. Andalusian, Chilean, Rioplatense, among others) present a phenomenon called s-debuccalization or s-aspiration, which consists in replacing syllable- and word-final /s/ with /h/, e.g. desde: /desðe/ → /dehðe/. In such cases, a phenomenon akin to s-voicing can occur before C[+voice], through which /h/ disappears and the following C[+voice] becomes geminated: desde /dehðe/ → [deððe], rasgo /rahɣo/ → [raɣɣo] (Morris).

2.2.4 Comparison

22The descriptions sketched above highlight differences across the three L1s of our learners. Table 1 summarizes the distribution of word-final [s] and [z] in the three above-mentioned romance languages:

23Table 1. Distribution of word-final [s] and [z] in French, Italian and Spanish

French

Italian

Spanish

[s]

Present (contrastive)

Rare

Present

[z]

Present (contrastive)

Absent

Present (non-obligatory allophone)

2.3 Aim of the study

24By analyzing a corpus of learner productions, we seek to determine if there exist differences in the voicing of word-final morphemic -s in English produced by L1 speakers of French, Italian, and Spanish. We will not analyze realizations of non-morphemic /s/ and /z/, as these are the object of a separate study focusing specifically on this phonemic distinction (Mairano et al. 2019). Voicing will be conceived as the proportion of periodical signal within the boundaries of each occurrence of morphemic -s, and will therefore be expressed as a relative value ranging from 0 (no voicing) to 1 (full voicing). For the sake of our analysis, periodicity will be the main acoustic correlate of voicing studied; vowel duration was taken in consideration as a secondary acoustic cue, but other acoustic correlates such as formant transitions and amplitude will not be considered.

25As shown above, the presence and distribution of final /z/ in English are different from those in French, Italian, and Spanish. Based on previous results in non-morphemic contexts (Mairano et al., forthcoming), and following the tenets of the SLM, the PAM, and the MDH, we posit the following hypotheses. We predict that:

26H1: Speakers of different L1s will present different voicing patterns.

27H2: Voicing patterns will be determined by the distribution patterns of word-final [s] and [z] in the speakers' L1s.

H2a: L1 French learners will be able to correctly produce the outcome of the voice assimilation rule in English since both sounds are present and have phonemic status in these contexts in their L1. Additionally, L1 French learners of L2 English have been found to produce a voicing contrast between /s/ and /z/ in a non-morphemic context (Mairano et al., forthcoming), and this reinforces our hypothesis for morphemic -s.
H2b: L1 Italian learners should be able to correctly produce the outcome of the voice assimilation rule since both sounds are present as phonemes in Standard Italian and as obligatory allophones in Northern Italian. Moreover, L1 Northern Italian learners of L2 English have already been shown to be able to reproduce the English /s/ ~ /z/, voice contrast in non-morphemic context, including in the more marked word-final position (Mairano et al., forthcoming), even though their L1 does not have [z] (and only rarely [s]) in word-final position.
H2c: L1 Spanish-speaking learners should find it difficult to correctly produce the outcome of the voice assimilation rule in word-final contexts since only /s/ has a phonemic value in L1 Spanish, while [z] is a non-obligatory contextual variant subject to a non-obligatory voice assimilation rule which applies in the opposite direction with respect to English (regressive instead of progressive). Moreover, L1 Spanish learners have been found not to be able to correctly reproduce the /s/ ~ /z/ voice contrast in a non-morphemic context (Mairano et al., forthcoming).

3. Methodology

28To study the voicing of morphemic -s in L2 English, we analyzed data from the IPCE-IPAC corpus (InterPhonology of Contemporary English, or InterPhonologie de l’Anglais Contemporain in French). The data corresponded to speech recordings of L2 English produced by L1 speakers of French, Spanish and Italian. Section 3.1 briefly presents the IPCE-IPAC corpus, while 3.2 describes the profile of the participants and the tasks performed. Section 3.3 explains the methodology followed to extract and compare periodicity values from all occurrences of word-final morphemic /s/ and /z/ across groups of speakers. Finally, 3.4 presents details about the statistical analysis.

3.1 The IPCE-IPAC corpus

29The IPCE-IPAC corpus is a collection of recordings of L2 English speakers with different L1 backgrounds. The recording protocol encompasses two read-aloud tasks (a newspaper article and a word list), a repeat-aloud task of a word list, a semi-guided formal interview with a teacher or native English speaker, an informal conversation with another non-native English speaker. Furthermore, all participants answer a questionnaire aimed at collecting information about their linguistic and acquisitional background. Herry-Bénit et al. (to appear) provides a full account of the IPCE-IPAC protocol.

30This study analyzes data from the reading aloud task. The speech obtained through this task is perfectly comparable across speakers since it forces them to utter the same words. A two-page-long text adapted from a newspaper article was used. Learners could scan through the article before reading it out loud, in order to become familiar with the topic, vocabulary and syntax. This was done to ensure speech fluidity for the study of connected speech phenomena, which could potentially have an influence on the degree of voicing in word-final position due to coarticulation with adjacent segments.

3.2 Participants

31Recordings produced by 40 learners of L2 English were analyzed. 15 of them were L1 speakers of metropolitan French (12 F, 3 M, age = 24, SD = 6.59, hereafter FR). 10 of them were enrolled at the University of Lille and the remaining 5 studied at the University of Paris 13 Nanterre and came from (Trifu-Dejeu) and (Gros-Bonfiglioli). Their CEFR proficiency level varied from B1 to C1. Their age of first contact with English (AOFC) was of 10.1 years on average (SD = 3.36). Aside from English, some of them spoke Spanish (n = 4), German (n = 2), Arabic (n = 1), Berber (n = 1), Italian (n = 1), and Portuguese (n = 1).

3215 participants were L1 speakers of Italian (11 F, 4 M; age = 22.5, SD = 2.38) and were students at the University of Turin at the time of the recording. 10 were from the Turin area, 2 were from other areas in northern Italy, 2 were from southern Italy, and 1 was from Sardinia. CEFR proficiency levels ranged from B1 to C1, and AOFC was on average of 10.6 years (SD = 3.36). All of them spoke other languages: French (n = 5), Spanish (n = 5), German (n = 5), Arabic (n = 2), Russian (n = 2), Catalan (n = 1), Japanese (n = 1), Portuguese (n = 1).

3310 participants were L1 speakers of South American Spanish (3 F, 7 M; age = 30.2, SD = 6.98), from Peru (n = 5), Colombia (n = 3), and Chile (n = 2). CEFR proficiency levels ranged from B1 to C1, and AOFC was of 7.4 years on average (SD = 2.84). At the time of recording, most of them lived in Paris or Lille and spoke French (n = 9), and additionally Italian (n = 1), and Portuguese and German (n = 1). One participant lived in Chile and did not speak any other L2 aside from English. Even though most L1 Spanish participants lived in France and spoke French, which might be considered as a potential confound in our study, this did not have an impact on our previous results on the /s/ ~ /z/ voice contrast for non-morphemic -s (Mairano et al., forthcoming).

34The recording conditions and equipment varied depending on the location. Nevertheless, all recordings took place in a quiet environment in order to ensure audio quality fit for acoustic analyses.

3.3 Data annotation and acoustic analysis

35We first proceeded to transcribe the recordings orthographically in a .txt file. This orthographic transcription was used to automatically create a phonetically annotated and aligned TextGrid file using WebMAUS (Kisler et al.). We manually verified the result of this process and corrected only errors produced by the automatic transcriber. We preserved a phonological approach and left the target L1 phonemes in our annotations of L2 speech instead of using symbols to indicate phonetic deviations in the learners' realizations. For instance, the /z/ at the end of knows was transcribed as /z/ – its target L1 English form – regardless of whether the speaker produced it as [s], [z], or [z̥]. WebMAUS allows the user to choose different regional varieties of English for the transcription. We used the British variety as a norm for our transcriptions since most Italian and French learners claimed to use it as a reference. We took this decision despite the fact that most Latin American participants followed a North American English pronunciation as a reference, because we favored the comparability of transcriptions for the subsequent extraction of acoustic parameters (cf. 3.3). This does not hinder our analysis of /s/ and /z/, since the two regional varieties do not present phonological differences at this level.

36Via a Praat script, we extracted the duration and the proportion of periodical signal of every realization of morphemic -s. This proportion is measured as relative to the duration of the whole segment and ranges from 0 (complete absence of periodic signal) to 1 (periodicity throughout the whole phoneme).

37Since phonological voice assimilation of /z/ is determined by the preceding segment, we distinguished three possible left contexts: /z/ preceded by a C[‑voice], /z/ preceded by a C[+voice], /z/ preceded by a V (including cases where morphemic -s is realized with the allomorph /ɪz, əz/, such as glasses and masses, whereby of course /z/ is preceded by a V). Additionally, since the periodicity of a phoneme can be influenced by adjacent phonemes due to coarticulation, and since regressive assimilation due to coarticulation in continuous speech has been found to be pervasive in English (cf. Stevens et al.; Haggard), we differentiated the analysis depending on whether a [+voice] segment (C or V) or a [-voice] segment (C or pause) followed. Table 2 below shows the number of realizations analyzed by group of speaker and position.

3.4 Statistical analysis

38Extracted data was saved in CSV format and analyzed in R (version 3.6.1, R Core Team). Visualization charts and diagrams were generated with the ggplot2 package (version 3.1.0, Wickham, 2016). Analyses of periodicity by group and position were carried out with mixed-effects models using the lme4 package (version 1.1.21, Bates et al.). Separate models were built for periodicity (cf. 4.1) and duration (cf. 4.3). We took into account the following fixed effects: Left Context (C[+voice], C[-voice], V), Right context (voiced segment, voiceless segment), Group (French, Italian, Spanish), and all their interactions. The following variables were considered as random effects: Speaker (to account for speaker-specific differences within each group) and Word (to account for possible lexical effects on voicing). The structure of the random effects was that of maximal specification (Barr, et al.), including random intercepts for all random effects, by-speaker random slopes for Left context and Right context, and was gradually simplified to reach convergence. All variables were coded with sum-to-zero contrasts in order to obtain meaningful intercept values (Schnielzeth, 2010) and the statistical significance of our predictors was determined by calculating p values via Satterthwaite’s approximation for the estimation of the degrees of freedom, as implemented in the lmerTest package (version 3.1.0, Kuznetsova el al.). We performed post-hoc tests as per the emmeans package (version 1.3.4, see Length et al., with fdr (false discovery rate) correction.

39Finally, we performed a random forests analysis using the party package (version 1.3.4., Strobl et al.), a machine learning technique that allows us to obtain reliable estimations of the individual contribution of each predictor, even in cases of multicollinearity, i.e. when multiple predictors are correlated to one another. This technique is gaining interest within various fields of linguistics and especially phonetics (see for instance Tagliamonte and Baayen; Winter and Grawunder). It allows us to assess the relative importance of the variables used for predicting voicing in word-final morphemic -s. We took into account the same variables as in the mixed-effects model analysis and fitted random forests with the cforest command, 500 decision trees and all other parameters set to default. Variable importance was estimated via the varimp command and results were plotted to charts for interpretation.

4. Results

40In order to study the voicing patterns of word-final morphemic -s produced by L2 English learners, we performed a quantitative and qualitative analysis of all realizations in terms of periodicity and duration. Unsurprisingly, the pronunciation of a few occurrences was omitted in the pronunciation of some learners. The number of such omitted pronunciations amounted to 121 (with respect to expected occurrences based on the text of the read-aloud task). These missing pronunciations could obviously not be analyzed in terms of periodicity or duration. Additionally, realizations of >0.4 seconds (n = 2) were discarded as mistakes or hesitations. In total, we analyzed 1324 tokens, of which 878 occurrences marking the plural, 404 occurrences marking the 3rd person, and 42 occurrences of the clitic form of ‘is’ in the word ‘it’s’ (not an inflectional morpheme, but still subject to voice assimilation, as already discussed in section 3.3); no /z/ marking the genitive occurred in the text. Table 2 shows the number of tokens analyzed by group, left context, and right context.

41Table 2. Number of occurrences analyzed by group, left and right context

Left context

Group

Right context

[+voice]

[-voice]

C[-voice]

L1 French

42

43

L1 Italian

72

12

L1 Spanish

46

8

C[+voice]

L1 French

37

135

L1 Italian

80

104

L1 Spanish

60

74

V

L1 French

71

152

L1 Italian

115

120

L1 Spanish

73

80

42In our analysis, we proceeded by separately inspecting the periodic proportion of each realization (sections 4.1 and 4.2) and its duration (section 4.3).

4.1 Quantitative analysis of periodicity

43We started the analysis by inspecting the average proportion of periodic signal within target segments (0 indicates no periodicity in the target segment, 1 indicates periodicity throughout the segment). In the analysis, we considered the left context (C[-voice], C[+voice], V) and the right context ([+voice], [-voice]) within which each target segment occurred. The result can be seen in Figure 1.

44The chart reveals that learners of all groups follow a similar trend, whereby the periodic proportion of the target segment is influenced by the preceding segment: more specifically, a preceding C[-voice] triggers a virtually non-periodic realization, a preceding C[+voice] triggers some periodicity in the realization of the target segment, and a preceding vowel triggers an even higher periodic proportion in the realization of the target segment. However, the amount of increase in periodicity is not the same across groups: L1 French learners show the greatest amount of increase, followed by L1 Spanish learners, followed by L1 Italian learners. Finally, it is interesting to note that the same pattern is reflected whether the right context is a voiced or a voiceless segment; a voiceless segment to the right does cause lower proportions of periodicity altogether, but we still observe a gradual (although more moderate) increase in periodicity due to the voice assimilation with the left phoneme.

Figure 1. Average periodic proportion for word-final morphemic -s by type of preceding phoneme (C[-voice], C[+voice], V) and type of following segment ([+voice], [-voice]) as pronounced by the three groups of learners

Figure 1. Average periodic proportion for word-final morphemic -s by type of preceding phoneme (C[-voice], C[+voice], V) and type of following segment ([+voice], [-voice]) as pronounced by the three groups of learners

45In order to test statistically these results, we fitted a mixed-effects model predicting the periodic proportion of the target segments on the basis of Left context (C[+voice], C[-voice], V), Right context ([+voice], [-voice]), Group (L1 French, L1 Italian, L1 Spanish) and their interactions as specified in section 3.5. Participant and Word were entered as random effects. By-speaker random slopes for all fixed effects were initially added to the model but had to be removed due to convergence issues. The model summary and other details can be found in Appendix 1, while the significance of the predictors is reported in Table 3.

46The main effect of our main predictor Left context was highly significant (p < .001), and post-hoc tests with fdr correction revealed that all pairwise comparisons were highly significant (all p values < .001). This confirms that, globally, the periodic proportion for word-final morphemic /z/ realizations was significantly higher if preceded by a V than if preceded by a C[+voice], and also significantly higher if preceded by a C[+voice] than if preceded by a C[-voice]. The main effect of the Right context was also highly significant, confirming that the periodic proportion of target segments is globally higher if a voiced segment follows (p = .002).

47Table 3. Statistical significance for the fixed effects in the model predicting the periodicity of target segments. Probability values were extracted from the ANOVA table (Satterthwaite’s method). See Appendix 1 for details about the model

Fixed effect

p value

Left context

<.001

Right context

.002

Group

.012

Left context x Right context

.158

Left context x Group

<.001

Right context x Group

.102

Left context x Right context x Group

.986

48Coming to differences across L1 groups, we found that the main effect of the Group was also significant (p = .012). Post-hoc pairwise comparisons revealed that, globally, the periodic proportion of target segments was significantly higher for L1 French vs. L1 Italian learners, while it was only descriptively but not statistically higher for L1 French vs. L1 Spanish learners, and for L1 Spanish leaners vs. L1 Italian learners. More interestingly, the Group interacted highly significantly with the Left context (p < .001), indicating that learners of different L1s did not increase periodicity according to the left context in the same way: in fact, we can observe in Figure 1 that the amount of increase is different across groups. In order to investigate this interaction in detail, post-hoc pairwise comparisons with fdr correction were carried out with the aim to observe differences of periodicity across groups and across types of left context (full post-hoc comparisons are reported in Appendix 1). Such tests revealed that: (a) for L1 French learners, periodicity increases significantly due to the left context from C[-voice] to C[+voice] and from C[+voice] to V (all p values < .001); (b) a similar situation is observed for L1 Spanish learners (all p values < .030); (c) for L1 Italian learners, periodicity due to the left context increases only descriptively from C[-voice] to C[+voice] (p = .268) but significantly from C[+voice] to V (p = .002). Furthermore, (d) if the left context is a C[+voice], periodicity does not differ across the three groups (p = .928 for all comparisons); (e) if the left context is a C[-voice], French learners show a higher periodic proportion in the target segments than Italian learners (p = .002), but differences for French vs. Spanish-speaking learners and for Spanish vs. Italian learners do not reach statistical significance (p = .099); (f) if the left context is a V, French learners show a higher periodic proportion in the target segments than Italian learners (p < .001) and Spanish-speaking learners (p = .014), while differences between Italian and Spanish-speaking learners are merely descriptive (p = .188). All this seems to suggest that all learners produce (at least a partial) voice assimilation of word-final morphemic -s to the left segment. Clearly, L1 French learners are the ones who produce the highest difference of periodicity based on the left context. Furthermore, it is remarkably interesting to note that L1 Spanish learners seem to produce voicing differences as well (at least partially). This comes as a surprise, given that these very same L1 Spanish learners do not seem to produce the /s/ ~ /z/ voicing contrast in positions in word-internal and word-final positions in contexts where the sound is not morphemic (Mairano et al., forthcoming). Further studies would be of interest in order to assess if the periodicity ratios produced by learners is comparable to those produced by native English speakers in cases of the voice assimilation rule.

49Finally, we wanted to determine the contribution of each variable in predicting the periodicity of morphemic -s in L2 English. In order to do so, we fitted a random forests model with our data and then ran a variable importance analysis. Figure 2 ranks variables in order of importance for predicting periodicity of the target segment. Interestingly, we can observe that Participant is the variable showing the strongest impact on predictions, suggesting that different participants may have different voicing patterns in otherwise equal conditions. The other variable expressing random effects in the linear mixed-effects analysis, that is Word, shows a moderate impact on the predictions resulting from the random forests analysis. The relevance of these two variables is further investigated in a more qualitative analysis in the following section. Among the variables that were considered as fixed effects in the linear mixed-effects analysis, Left context is by far the one showing the greatest impact on random forests predictions. This probably reflects the fact that learners of all groups produced differences in periodicity depending on whether a C[-voice], C[+voice], or V preceded the target segment. Right context showed a far smaller impact on predictions, probably reflecting the fact that voice assimilation resulting from coarticulation with the next segment is far more limited than voice assimilation resulting from a phonological transformation rule, even in L2 English. Finally, the impact of L1 Group on random forests predictions is also fairly limited, probably reflecting the fact that learners of all L1s produce periodicity for word-final morphemic -s with similar patterns.

Figure 2. Variable importance for predicting the periodicity of word-final morphemic -s in L2 English

Figure 2. Variable importance for predicting the periodicity of word-final morphemic -s in L2 English

4.2 Qualitative analysis of periodicity

50Since the impact of Participant and Word was found to be relevant for determining the periodicity for word-final morphemic -s, we decided to investigate the effect of these variables adopting a more qualitative approach. In order to do so, we observed periodicity as a function of Word and Participant in a by-item and a by-participant analysis.

51The by-item analysis (Figure 3) shows the periodic proportion for morphemic -s for the words in the text. The chart above illustrates cases in which the target segment is preceded by a C[-voice] and shows no relevant difference across words: learners of all groups produced virtually no periodicity in all of them. The chart in the middle illustrates cases in which the target segment is preceded by a C[+voice] and reveals few surprises. We can note that L1 Italian learners produce extremely little periodicity in all words, while L1 Spanish and L1 French participants show higher proportions of periodicity. The only relevant peculiarity concerns the inflectional /z/ in buckles, which tends to be realized as voiceless by all groups. Finally, the chart below illustrates cases in which the target segment is preceded by a V and, again, reveals few surprises: L1 French learners produce higher proportions of periodicity for the target segment than L1 Spanish and L1 Italian learners, with the exception of neighbours (which may be due to many speakers producing a rhotic consonant).

Figure 3. By-item analysis of the periodic proportion for word-final morphemic -s as pronounced by the three groups of learners

Figure 3. By-item analysis of the periodic proportion for word-final morphemic -s as pronounced by the three groups of learners

52The by-participant analysis reveals differences among learners of each group, providing interesting insights in the process of acquisition of this phonological process. As for the L1 French group, we can observe that most learners reproduced the expected pattern for the progressive voicing assimilation: whenever the target segment is preceded by a C[-voice] (red line), it is realized as entirely aperiodic, while whenever it is preceded by a C[+voice] (green line) or a V (blue line), we find periodicity. FR14 is the only L1 French learner who does not seem to make any voice assimilation to the left context, while FR03, FR04 and FR15 do so very slightly. Additionally, FR01 behaves unexpectedly, with higher periodicity for the target segment in presence of C[-voice] and V, and no periodicity in presence of C[+voice]. These patterns somewhat reflect data by the same learners for realizations of the /s/ ~ /z/ contrast in contexts where voicing is not governed by a phonological voicing rules (Mairano et al., forthcoming): in that condition, FR14 also does not produce differences in periodicity between /s/ and /z/, while FR03, FR04 and FR15 produce but a small difference.

53As for the L1 Italian group, one learner (IT11) stands out by nicely reproducing the expected pattern. For most other speakers, the assimilation rule seems to apply only partially: target segments are aperiodic when preceded by a C[-voice] and periodical when preceded by a V, and this corresponds to the expected pattern. However, for most speakers the target segment is unexpectedly aperiodic even in the case where it is preceded by C[+voice]. Additionally, some speakers (IT04, IT06, IT07, IT10, IT12) produce no or little difference in periodicity depending on the left context. This partially reflects results obtained for the same learners for realizations of the /s/ ~ /z/ contrast in contexts where voicing is not governed by a phonological rule (see the companion paper Mairano et al., forthcoming): in that condition, IT04, IT06 and IT12 also did not produce differences in periodicity between /s/ and /z/.

54Finally, most L1 Spanish learners seem to be able to produce the expected outcome of the voice assimilation rule. This is surprising, given that these same participants failed to correctly produce the /s/ ~ /z/ voicing contrast in other contexts (cf. (see the companion paper Mairano et al., forthcoming). The only L1 Spanish learner who does not seem to reproduce the progressive voice assimilation rule is SP01, for whom periodicity is comparable in all three conditions. At any rate, the fact that L1 Spanish learners reproduce the voice assimilation rule for morphemic -s despite failing to produce the /s/ ~ /z/ contrast in other contexts is unexpected and puzzling.

Figure 4. By-participant analysis of the periodic proportion for word-final morphemic -s as pronounced by the three groups of learners

Figure 4. By-participant analysis of the periodic proportion for word-final morphemic -s as pronounced by the three groups of learners

4.3 Analysis of duration

55Having analyzed the primary acoustic cue of voicing, we also investigated one of its secondary cues, namely duration. We did this because in our study on the /s/ ~ /z/ contrast (Mairano et al., forthcoming), L2 English learners were found to use shorter durations for /z/ than for /s/, like native English speakers. We carried out the analysis of duration in the same way as we did for periodicity, i.e. by plotting measurements of duration for the target segments, differentiated by left context (C[-voice], C[+voice], V) and right context ([+voice], [-voice]). The results, shown in Figure 5, do not suggest that duration plays a relevant role in the phonetic implementation of the progressive voice assimilation rule in the speech of L2 English learners. The only clear effect on duration seems to be due to the right context, rather than the left context.

Figure 5. Average periodic proportion for word-final morphemic -s by type of preceding phoneme (C[-voice], C[+voice], V) and type of following segment ([+voice], [-voice]) as pronounced by the three groups of learners

Figure 5. Average periodic proportion for word-final morphemic -s by type of preceding phoneme (C[-voice], C[+voice], V) and type of following segment ([+voice], [-voice]) as pronounced by the three groups of learners

56In order to statistically test these results, we fitted a linear mixed-effects model for predicting durations of word-final morphemic -s on the basis of Group, Left context, Right context, and their interactions, with Participant and Word as random effects. In this case, we were able to include by-participant random slopes for Left context, Right context and their interaction without experiencing convergence issues. The significance of the fixed effects is shown in Table 4, while the summary of the model and other details are reported in Appendix 2. The only significant effect found was that of the Right context: word-final morphemic -s occurrences followed by a voiceless segment tend to be longer than if followed by a voiced segment. The effect of Group and Left context were not significant, suggesting that durations play a small (if any) role in the phonetic implementation of the progressive voice assimilation for morphemic -s in the speech of L2 English learners. For this reason, we did not further pursue the analysis of durations.

57Table 4. Statistical significance for the fixed effects in the model predicting the duration of target segments. Probability values were extracted from the ANOVA table (Satterthwaite’s method). See Appendix 2 for details about the model

Fixed effect

p value

Left context

.737

Right context

<.001

Group

.374

Left context x Right context

.828

Left context x Group

.121

Right context x Group

.122

Left context x Right context x Group

.911

5. Final discussion

58In this paper, we studied the behavior of English word-final morphemic -s in the speech of L2 learners whose L1 is French, Italian or Spanish. The progressive voice assimilation rule triggered by the Left context is realized by learners of all L1s, but to different degrees. This confirms our first hypothesis (H1) and suggests that many of our learners have acquired the devoicing constraints of morphemic -s. However, statistical analyses show that voicing patterns vary per group and per type of left context. Namely, while periodicity differences due to the left context are significant for all groups for C[-voice] vs. V, the same cannot be said for C[-voice] vs C[+voice]. The latter are statistically significant only for L1 French and (to a lesser extent) L1 Spanish learners, but not for L1 Italian learners.

59Our results also validate hypothesis H2a about French-speaking learners, who have been found to produce a voicing alternation following the English voice assimilation rule for morphemic -s. Instead, hypotheses H2b and H2c about Italian and Spanish-speaking learners have not been validated by our data. We had initially hypothesized that L1 Italian learners would find it easier than L1 Spanish learners to produce the voicing patterns of morphemic -s. This was based on the following assumptions: H2b) L1 Italian learners would be able to apply the English voice assimilation rule due to the existence of [s] and [z] as obligatory allophones in their L1, and due to their ability to produce a difference in periodicity for /s/ and /z/ in L2 English in other contexts, including word-final position for non-morphemic -s (Mairano et al., forthcoming); H2c) L1 Spanish-speaking learners would find it difficult to reproduce the outcome of the English voice assimilation rule due to the existence of only /s/ as a phoneme ([z] being a non-obligatory allophone) and due to the fact that they failed to produce a difference in periodicity for /s/ and /z/ in L2 English in other contexts (Mairano et al., forthcoming), and due to the interference of a voice assimilation rule with opposite directionality affecting morphemic -s in their L1.

60The results of morphemic -s voicing for L1 Spanish learners of L2 English challenge predictions based on the MDH. According to the MDH, word-final /z/ should be difficult, due not only to a negative transfer of the L1 (no /z/ in L1 Spanish), but also to /z/ being more marked in word-final position than in word-medial position. It is therefore puzzling to observe that L1 Spanish learners fail to produce a voicing contrast between /s/ and /z/ in word-medial position (as well as in a word-final non-morphemic context, cf. Mairano et al., forthcoming), but that they successfully reproduce the output of the voice assimilation rule for morphemic -s in word-final position. This may potentially be linked to the existence of a regressive voice assimilation for final -s in Spanish. Initially, we had hypothesized that such rule would have a negative effect on L2 English productions, given the opposite directionality (regressive vs progressive). But it is possible that the simple existence of a voice assimilation rule (although gradient, non-obligatory, and with opposite directionality) may enhance L1 Spanish learners’ sensitivity to voicing in the context where it applies. In other words, although /z/ is more marked in word-final position, L1 Spanish learners may be more sensitive to voicing in this context due to the existence of a voice assimilation rule in their L1 applying to morphemic -s in Spanish.

61It is also puzzling to observe that Italian learners are outperformed by Spanish learners in terms of periodicity. The former do not produce a significant difference in morphemic -s voicing when it is preceded by C[-voice] and C[+voice], and globally they use a far smaller proportion of periodicity for morphemic -s than L1 Spanish learners. It is also worth mentioning that these acoustic results converge with impressionistic judgments by Bosisio, who found that Northern Italian learners of L2 English mostly did not produce /z/ for morphemic -s. At first sight, one may think that the results for L1 Italian learners are explained by phonotactic constraints of the target sounds in the learners’ L1: since /z/ is nonexistent in Italian in word-final position (while /s/ is marginal but existent), word-final /z/ in L2 English may simply be realized by Italian learners following their L1 phonotactic constraints, namely as [s]. This explanation would be in line with the MDH, given that [z] is more marked in word-final position, both universally and cross-linguistically (i.e. with respect to L1 constraints). However, this does not reflect our previous finding that the same L1 Italian learners produce voiced realizations for /z/ in word-final position in a non-morphemic context (Mairano et al., forthcoming). It thus appears that the difficulty in producing /z/ for our L1 Italian learners of L2 English is restricted to morphemic -s, rather than being a generalized problem in producing voicing in word-final position. This then supports the claim that morphemic -s has a different status from non-morphemic -s in learners’ phonological representations (see below) and opens research avenues to be explored in the future.

62We believe that our results have repercussions on MDH, and more widely on models of L2 acquisition. Concerning MDH, our findings support our previous claim that the /s/ ~ /z/ opposition may be a partial exception to the following markedness hierarchy for voiced obstruents: word-initial < word-medial < word-final. In neither study did we test the word-initial position, but the present study tested the word-final morphemic position, and the previous study tested the word-medial and word-final (non-morphemic) positions. Taken together, our results suggest that: (i) producing word-final /z/ is not more difficult than in word-medial intervocalic position for L1 Spanish learners of L2 English, and even seems to be easier in word-final position in the case of morphemic -s; (ii) producing word-final is more difficult than in word-medial position for L1 Italian learners, but only for morphemic -s, suggesting that the difficulty may not be due to producing voicing in the final position but may instead interact with the morphemic nature of the sound. These findings seem to suggest that word-final /z/ may be more marked than word-medial /z/ and thereby cast doubts to the hierarchy mentioned above for obstruent voicing. Other considerations seem to support this fact. First, word-initial /z/ (where voicing is supposed to be least marked) is rare in English (but also in other languages, including French), and clearly rarer than in word-medial or word-final position (results by Kessler and Treiman show that /z/ is more than four times rarer in syllable onset than in syllable coda). Additionally, /s/ and /z/ have often been found to be exceptions within phonological theories, for instance within the sonority hierarchy for syllable structure (cf. Venneman). It may well be the case that these two sounds constitute an exception to the markedness hierarchy for voicing as well. Of course, our results can only be considered as preliminary, and will need to be further tested in the future in order to be validated. Whether such claims are confirmed by future investigations or not, an important consideration can be inferred from the present results. It is remarkable that voicing patterns for word-final morphemic -s do not seem to reflect the voicing patterns observed for word-final non-morphemic -s in productions by the same speakers. As already mentioned, L1 Italian learners exhibited a voicing alternation for non-morphemic /s/ ~ /z/ but a limited voicing difference in morphemic -s, while Spanish learners exhibit the opposite, that is to say no voicing alternation for word-final non-morphemic -s, but considerable voicing differences for word-final morphemic -s. These results may be connected to recent findings by Plag et al. for L1 English that morphemic and non-morphemic -s are not homophonous. Although Plag et al. revealed differences in duration rather than voicing, our data seem to confirm that morphemic and non-morphemic -s are produced with phonetically different patterns by learners of L2 English. Of course, these findings need to be taken with due caution, considering that they come from a corpus-based analysis where not all covariates could be controlled. However, if more specific investigations should replicate such results in the future, they could not only support discoveries on L1 English, but also have important repercussions on models of L2 phonology acquisition that do not currently consider the role of morphological status in the pronunciation of L2 phonemes.

Appendix 1

63We report here the summary for the linear mixed-effects model predicting periodicity for realizations of morphemic -s, as presented in section 4.1. Variables were contrast coded as follows: Left Context (C[-voice] = 1, 0; C[+voice] = 0, 1; V = -1, 1), Right Context (voiced = -1; voiceless = 1), Group (L1 French = 1, 0; L1 Italian = 0, 1; L1 Spanish = -1, -1). Model formula: Periodicity ~ Left context * Right context * Group + (1 | Participant) + (1 | Word).

64NB: we tried including by-participant random slopes for Left context and Right context, but we experienced model convergence issues and had to remove them.

Parameters

Fixed effects

Random effects

Part.

Word

Est.

SE

t

p

SD

SD

Intercept

0.184

0.028

6.54

<0.001

0.153

0.051

Left context 1

-0.134

0.023

-5.72

<0.001

Left context 2

0.010

0.017

0.57

0.571

Right context 1

0.031

0.010

3.06

0.002

L1 1

0.097

0.036

2.71

0.010

L1 2

-0.090

0.036

-2.49

0.017

Left context 1 x Right context 1

-0.029

0.017

-1.74

0.081

Left context 2 x Right context 1

0.008

0.013

0.62

0.536

Left context 1 x L1 1

-0.078

0.020

-3.81

<0.001

Left context 2 x L1 1

0.021

0.016

1.28

0.202

Left context 1 x L1 2

0.059

0.023

2.58

0.010

Left context 2 x L1 2

-0.027

0.016

-1.66

0.097

Right context 1 x L1 1

0.025

0.013

2.02

0.044

Right context 1 x L1 2

-0.018

0.014

-1.37

0.172

Left context 1 x Right context 1 x L1 1

-0.007

0.021

-0.32

0.749

Left context 2 x Right context 1 x L1 1

0.008

0.016

0.51

0.614

Left context 1 x Right context 1 x L1 2

-0.003

0.023

-0.12

0.905

Left context 2 x Right context 1 x L1 2

-0.003

0.016

-0.17

0.862

65The following tables give post-hoc pairwise comparisons for the Lext context x L1 Group interaction: (a) levels of Left context within every L1 Group (i.e., Left context | L1 Group) above, and (b) L1 Groups within every Left context (i.e., L1 Group | Left context) below. Probability values were computed with fdr (false discovery rate) correction for multiple comparisons.

Contrast (Left context | L1 Group)

estimate

SE

p

Group = L1 French

C[-voice] - C[+voice]

-0.243

0.045

<0.001

C[-voice] - V

-0.393

0.042

<0.001

C[+voice] - V

-0.151

0.036

<0.001

Group = L1 Italian

C[-voice] - C[+voice]

-0.058

0.052

0.268

C[-voice] - V

-0.168

0.051

0.002

C[+voice] - V

-0.109

0.031

0.002

Group = L1 Spanish

C[-voice] - C[+voice]

-0.1322

0.060

0.029

C[-voice] - V

-0.216

0.060

0.001

C[+voice] - V

-0.083

0.035

0.029

Contrast (L1 Group | Left context)

estimate

SE

p

Left context = C[+voice]

L1 French - L1 Italian

0.049

0.074

0.928

L1 French - L1 Spanish

0.008

0.084

0.928

L1 Italian - L1 Spanish

-0.041

0.089

0.928

Left context = C[-voice]

L1 French - L1 Italian

0.234

0.063

0.002

L1 French - L1 Spanish

0.118

0.070

0.099

L1 Italian - L1 Spanish

-0.116

0.069

0.099

Left context = V

L1 French - L1 Italian

0.275

0.061

0.001

L1 French - L1 Spanish

0.185

0.068

0.014

L1 Italian - L1 Spanish

-0.090

0.068

0.188

Appendix 2

66We report here the summary for the linear mixed-effects model predicting duration for realizations of morphemic -s, as presented in section 4.3. Variables were contrast coded as follows: Left Context (C[-voice] = 1, 0; C[+voice] = 0, 1; V = -1, 1), Right Context (voiced = -1; voiceless = 1), Group (L1 French = 1, 0; L1 Italian = 0, 1; L1 Spanish = -1, -1). Model formula: Duration ~ Left context * Right context * Group + (Left context * Right context | Participant) + (1 | Word).

Parameters

Fixed effects

Random effects

Part.

Word

Est.

SE

t

p

SD

SD

Intercept

0.119

0.004

27.50

<0.001

0.019

0.014

Left context 1

-0.004

0.005

-0.78

0.438

0.007

Left context 2

0.002

0.004

0.54

0.589

0.006

Right context 1

-0.007

0002

-4.18

<0.001

0.003

L1 1

0.006

0.005

1.21

0.234

L1 2

-0.005

0.005

-1.15

0.258

Left context 1 x Right context 1

-0.002

0.003

-0.59

0.557

0.010

Left context 2 x Right context 1

0.001

0.002

0.34

0.734

0.008

Left context 1 x L1 1

0.007

0.003

1.97

0.055

Left context 2 x L1 1

-0.003

0.003

-0.96

0.340

Left context 1 x L1 2

0.003

0.004

0.78

0.440

Left context 2 x L1 2

-0.004

0.003

-1.40

0.166

Right context 1 x L1 1

0.004

0.002

2.08

0.041

Right context 1 x L1 2

-0.002

0.002

-0.83

0.410

Left context 1 x Right context 1 x L1 1

0.002

0.004

0.48

0.636

Left context 2 x Right context 1 x L1 1

<0.001

0.003

0.13

0.894

Left context 1 x Right context 1 x L1 2

-0.002

0.004

-0.37

0.710

Left context 2 x Right context 1 x L1 2

<0.001

0.003

0.23

0.821

Top of page

Bibliography

Ågren, John. Étude Sur Quelques Liaisons Facultatives Dans Le Français de Conversation Radiophonique : Fréquences et Facteurs. Uppsala: Acta Universitatis Upsaliensis, 1973.

Baroni, Antonio. “Element Theory and the Magic of /S/.” In Crossing Phonetics-Phonology Lines, edited by Cyran, Eugeniusz and Jolanta Szpyra Kozłowska. Newcastle: Cambridge Scholars Publishing, 2014: 3–30.

Barr, Dale J., et al. “Random effects structure for confirmatory hypothesis testing: Keep it maximal.” Journal of memory and language 68.3 (2013): 255–278.

Barreiro Bilbao, Silvia Carmen. “Perception of Voicing in English Fricatives by Spanish Listeners.” Círculo de Lingüística Aplicada a La Comunicación, vol. 69 (2017): 3466, doi:10.5209/CLAC.55313.

Bates, Douglas et al. “Fitting Linear Mixed-Effects Models Using Lme4.” Journal of Statistical Software, vol. 67, no. 1 (2015): 1–48, doi:10.18637/jss.v067.i01.

Bertinetto, Pier Marco and Michele Loporcaro. “The Sound Pattern of Standard Italian, as Compared with the Varieties Spoken in Florence, Milan and Rome.” Journal of the International Phonetic Association, vol. 35, no. 2 (2005): 132–51, doi:10.1017/S0025100305002148.

Best, Catherine. “A Direct Realist View of Cross-Language Speech Perception.” In Speech Perception and Linguistic Experience: Issues in Cross-Language Research, edited by Winifred Strange. Timonium, MD: York Press, 1995: 171–204.

Bosisio, Nicole. “S-Assimilation in English and Italian: Implications for Foreign Language Learning and Teaching.” In Current Issues in English Language Teaching and Learning: An International Perspective, edited by Cal Varela, Mario et al. Newcastle: Cambridge Scholars Publishing, 2010: 175–88.

Bradley, Travis G. Sibilant voicing in highland Ecuadorian Spanish.” Lingua (gem), vol 2, no. 2 (2005): 942.

Bradley, Travis G. and Ann Marie Delforge. “Systemic Contrast and the Diachrony of Spanish Sibilant Voicing.” In Historical Romance Linguistics: Retrospective and Perspectives, edited by Randall Gess and Deborah Arteaga. Amsterdam: John Benjamins, 2006: 19, doi:10.1075/cilt.274.04bra.

Broniś, Olga. “Italian Vowel Paragoge in Loanword Adaptation. Phonological Analysis of the Roman Variety of Standard Italian.” Italian Journal of Linguistics, vol. 28, no. 2 (2016): 25–68.

Brown, Adam. “Functional Load and the Teaching of Pronunciation.” In Teaching English Pronunciation: A Book of Readings, edited by Adam Brown. New York: Routledge, 1991.

Bybee, Joan. “Frequency Effects on French Liaison.” In Frequency and the Emergence of Linguistic Structure, edited by Joan Bybee and Paul Hopper. Amsterdam: John Benjamins Publishing Company, 2001: 33760.

Bybee, Joan. “La Liaison : Effets de Fréquence et Constructions.” Langages, vol. 39, no. 158 (2005): 24–37, doi:10.3406/lgge.2005.2660.

Catford, John Cunnison. “Phonetics and the Teaching of Pronunciation: A Systemic Description of the Teaching of English Phonology.” In Current Perspectives on Pronunciation: Practices Anchored in Theory, edited by J. Morley. Washington, D.C.: TESOL, 1987: 83–100.

Chomsky, Noam and Morris Halle. The Sound Pattern of English. New York: Harper & Row, 1968.

Cruttenden, Alan. Gimson’s Pronunciation of English. London: Routledge, 2014 (8th edition).

Crystal, Thomas H. and Arthur S. House. “Segmental Durations in Connected-Speech Signals: Syllabic Stress.” Journal of the Acoustical Society of America, vol. 83, no. 4 (1988): 1574–85, doi:10.1121/1.395912.

Delattre, Pierre. Studies in French and Comparative Phonetics. The Hague, London, Paris: Mouton, 1966.

Domokos, György. “Anglicismi Nella Lingua Italiana.” Verbum, vol. 2 (2001): 295–305.

Eckman, Fred R. “Typological Markedness and Second Language Phonology.” In Phonology and Second Language Acquisition, edited by J. G. Hansen Edwards and M. L. Zampini. Amsterdam: John Benjamins Publishing Company, 2008: 95–115.

Finegan, Edward. Language: Its Structure and Use. Boston: Thomson Wadsworth, 2007 (5th edition).

Flege, James E. “Second Language Speech Learning: Theory, Findings, and Problems.” In Speech Perception and Linguistic Experience: Issues in Cross-Language Research, edited by Winifred Strange. Timonium, MD: York Press, 1995: 233–77, doi:10.1111/j.1600-0404.1995.tb01710.x.

Flege, James E. “The Effect Of Linguistic Experience on Arabs’ Perception of the English /s/ vs . /z/ Contrast.” Folia Linguistica, vol. 18, no. 1-2 (1984): 11738.

Fullana, Natalia and Joan C. Mora. “Production and Perception of Voicing Contrasts in English Word-Final Obstruents: Assessing the Effects of Experience and Starting Age.” In Recent Research in Second Language Phonetics/Phonology: Perception and Production, edited by Michael Watkins, Andreia Rauber and Barbara Baptista. Newcastle: Cambridge Scholars Publishing Citeseer, 2009: 97–117.

Greenberg, Joseph. Language Universals. The Hague: Mouton, 1976.

Gros-Bonfiglioli, Audrey. L’utilisation du linking en anglais : le cas d’apprenants francophones dans IPCE-IPAC. PhD Thesis, University of Paris 10 Nanterre, in preparation.

Haggard, Mark. “The Devoicing of Voiced Fricatives.” Journal of Phonetics, vol. 6, no. 2, Elsevier Masson SAS (1978): 95–102, doi:10.1016/s0095-4470(19)31101-5.

Hallé, Pierre André and Martine Adda-Decker. “Voicing Assimilation in Journalistic Speech.” 16th International Congress of Phonetic Sciences, 2007: 493–96.

Hallé, Pierre André and Martine Adda-Decker. “Voice Assimilation in French Obstruents: A Gradient or a Categorical Process?” In Tones and Features: A Festschrift for Nick Clements, edited by John Goldsmith, Elizabeth Hume and Leo Wetzels. Berlin: De Gruyter, 2011: 149–75.

Hardcastle, W. J. and J. E. Clark. “Articulatory, Aerodynamic and Acoustic Properties of Lingual Fricatives in English.” Speech Research Laboratory Work in Progress, vol. 3 (1981): 51–78.

Herry-Bénit, Nadine et al. “IPCE-IPAC Interphonology of Contemporary English: Methodological Issues.” International Journal of Learner Corpus Research, to appear.

Hooper, Joan B. “The Syllable in Phonological Theory.” Language, vol. 48, no. 3 (1972): 525, doi:10.2307/412031.

House, Arthur S. “On Vowel Duration in English.” The Journal of the Acoustical Society of America, vol. 33, no. 9 (1961): 1174–78, doi:10.1121/1.1908941.

Huszthy, Bálint. “The ‘Untamed’ / s / of Italian Dialects An Overview of the Singular Behaviour of Italo-Romance Sibilants.” Verbum, vol. 1, no. 2 (2017): 191–216.

Jongman, Allard et al. “Acoustic Characteristics of English Fricatives.” The Journal of the Acoustical Society of America, vol. 108, no. 3 (2000): 1252, doi:10.1121/1.1288413.

Kessler, Brett and Rebecca Treiman. “Syllable structure and the distribution of phonemes in English syllables.” Journal of Memory and language, vol. 37, no. 3 (1997): 295–311.

Kisler, Thomas et al. “Multilingual Processing of Speech via Web Services.” Computer Speech & Language, vol. 45 (2017): 326–47.

Krämer, Martin. The Phonology of Italian. Oxford: Oxford University Press, 2009.

Kuznetsova, Alexandra et al. “lmerTest package: tests in linear mixed effects models.” Journal of statistical software 82.13 (2017): 1–26.

Lenth, Russell, et al. “Emmeans: Estimated marginal means, aka least-squares means.” R package version 1.3.4 (2019).

Mairano Paolo et al. “Acoustic distances, Pillai scores and LDA classification scores as metrics of L2 comprehensibility and nativelikeness.” Proceedings of the International Congress of Phonetic Sciences. Melbourne, 2019: 1104–08.

Mairano, Paolo et al. “The /s/ ~ /z/ voice contrast in L1 French, L1 Spanish and L1 Italian learners of L2 English.” Language, Acquisition and Interaction (forthcoming).

Mines, M. Ardussi et al. “Frequency of Occurrence of Phonemes in Conversational English.” Language and Speech, vol. 21, no. 3 (1978): 221–41.

Morris, Richard E. “Constraint Interaction in Spanish / s / -Aspiration : Three Peninsular Varieties.” Proceedings of the Third Hispanic Linguistics Symposium, edited by Héctor Campos et al. Somerville: Cascadilla, 2000: 14–30.

Moulton, William G. The Sounds of English and German. Chicago: The University of Chicago Press, 1962.

Munro, Murray J. and Tracey M. Derwing. “The Functional Load Principle in ESL Pronunciation Instruction: An Exploratory Study.” System, vol. 34, no. 4 (2006): 52031.

Navarro Tomás. Manual de Pronunciación Española. Madrid: Consejo Superior de Investigaciones Científicas, 1977 (19th edition).

Nespor, Marina. Fonologia. Bologna: Il Mulino, 1993.

Nespor, Marina and Irene Vogel. Prosodic Phonology. Berlin: De Gruyter Mouton, 1986.

Peterson, Gordon E. and Ilse Lehiste. “Duration of Syllable Nuclei in English.” In The Journal of the Acoustical Society of America, vol. 32, no. 6 (1960): 693–703, doi:10.1121/1.1908183.

Plag, Ingo et al. “Homophony and Morphology: The Acoustics of Word-Final S in English.” Journal of Linguistics, vol. 53, no. 1 (2017): 81–216, doi:10.1017/S0022226715000183.

R Core Team. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, 2014, http://www.r-project.org/.

Raphael, Lawrence J. “Preceding Vowel Duration as a Cue to the Perception of the Voicing Characteristic of Word-Final Consonants in American English.” Journal of the Acoustical Society of America, vol. 51, no. 4B (1972): 1296–303, doi:10.1121/1.1912974.

Rosoff, Gary H. “The Function of Liaison as a Correlate of Plurality in Spoken French.” Canadian Modern Language Review, vol. 30, no. 4 (1974): 357–61, doi:10.3138/cmlr.30.4.357.

Schielzeth, H. (2010). “Simple means to improve the interpretability of regression coefficients.” Methods in Ecology and Evolution, 1(2), 103–113.

Smith, Bruce L. “A Phonetic Analysis of Consonantal Devoicing in Children’s Speech.” Journal of Child Language, vol. 6, no. 1 (1979): 19–28, doi:10.1017/S0305000900007595.

Smith, Caroline L. “The Devoicing of /z/ in American English: Effects of Local and Prosodic Context.” Journal of Phonetics, vol. 25, no. 4 (1997): 471–500, doi:10.1006/jpho.1997.0053.

Stevens, Kenneth N. et al. “Acoustic and Perceptual Characteristics of Voicing in Fricatives and Fricative Clusters.” The Journal of the Acoustical Society of America, vol. 91, no. 5 (1992): 2979–3000, doi:10.1121/1.402933.

Strobl, Carolin, et al. “Conditional Variable Importance for Random Forests.” BMC Bioinformatics, vol. 9, no. 307 (2008): doi:10.1186/1471-2105-9-307.

Tagliamonte, Sali A. and R. Harald Baayen. “Models, Forests, and Trees of York English: Was/Were Variation as a Case Study for Statistical Practice.” Language Variation and Change, vol. 24, no. 2 (2012): 135–78, doi:10.1017/S0954394512000129.

Torreblanca, Maximo. “El Fonema /s/ En La Lengua Española.” Hispania, vol. 61, no. 3 (1978): 498, doi:10.2307/341080.

Trifu-Dejeu, Ioana. The Interphonology of /r/ in French Learners of English in the IPCE-IPAC Corpus. PhD Thesis, Université Paris Nanterre, in preparation.

Tyler, Michael, et al. “Perceptual Assimilation of English Dental Fricatives by Native Speakers of European French.” Proceedings of the 19th International Congress of Phonetic Sciences, 2019.

Vennemann, Theo. “On the theory of syllabic phonology.” Linguistische Berichte, vol. 18 (1972): 1–18.

Walsh, Thomas and Frank Parker. “The Duration of Morphemic and Non-Morphemic /s/ in English.” Journal of Phonetics, vol. 11, no. 2, Elsevier Masson SAS (1983): 201-06, doi:10.1016/s0095-4470(19)30816-2.

Wickham, Hadley. ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag, 2016.

Whitley, Melvin Stanley. Spanish/English Contrasts: A Course in Spanish Linguistics. Washington, D.C.: Georgetown University Press, 1986.

Winter, Bodo and Sven Grawunder. “The Phonetic Profile of Korean Formal and Informal Speech Registers.” Journal of Phonetics, vol. 40, no. 6 (2012): 808–15, doi:10.1016/j.wocn.2012.08.006.

Yung Song, Jae, et al. “Durational Cues to Fricative Codas in 2-Year-Olds’ American English: Voicing and Morphemic Factors.” The Journal of the Acoustical Society of America, vol. 133, no. 5 (2013): 2931–46, doi:10.1121/1.4795772.

Top of page

Notes

1 Cruttenden explains that, in final position, the fricatives /v, ð, z, ʒ/, although voiceless, remain lenis [v̥,ð̥,z̥,ʒ̥] (p. 193).

Top of page

List of illustrations

Title Figure 1. Average periodic proportion for word-final morphemic -s by type of preceding phoneme (C[-voice], C[+voice], V) and type of following segment ([+voice], [-voice]) as pronounced by the three groups of learners
URL http://journals.openedition.org/anglophonia/docannexe/image/3711/img-1.png
File image/png, 156k
Title Figure 2. Variable importance for predicting the periodicity of word-final morphemic -s in L2 English
URL http://journals.openedition.org/anglophonia/docannexe/image/3711/img-2.png
File image/png, 95k
Title Figure 3. By-item analysis of the periodic proportion for word-final morphemic -s as pronounced by the three groups of learners
URL http://journals.openedition.org/anglophonia/docannexe/image/3711/img-3.png
File image/png, 407k
Title Figure 4. By-participant analysis of the periodic proportion for word-final morphemic -s as pronounced by the three groups of learners
URL http://journals.openedition.org/anglophonia/docannexe/image/3711/img-4.png
File image/png, 393k
Title Figure 5. Average periodic proportion for word-final morphemic -s by type of preceding phoneme (C[-voice], C[+voice], V) and type of following segment ([+voice], [-voice]) as pronounced by the three groups of learners
URL http://journals.openedition.org/anglophonia/docannexe/image/3711/img-5.png
File image/png, 161k
Top of page

References

Electronic reference

Leonardo Contreras Roa, Paolo Mairano, Marc Capliez and Caroline Bouzon, Voice assimilation of morphemic -s in the L2 English of L1 French, L1 Italian and L1 Spanish learnersAnglophonia [Online], 30 | 2020, Online since 20 December 2020, connection on 30 April 2025. URL: http://journals.openedition.org/anglophonia/3711; DOI: https://doi.org/10.4000/anglophonia.3711

Top of page

About the authors

Leonardo Contreras Roa

LIDILE (EA 3874)
lcontrerasroa@gmail.com

Paolo Mairano

Université de Lille, STL UMR8163
paolo.mairano@univ-lille.fr

By this author

Marc Capliez

Université de Lille, STL UMR8163
marc.capliez@univ-lille.fr

Caroline Bouzon

Université de Lille, STL UMR8163

 


caroline.bouzon@univ-lille.fr

By this author

Top of page

Copyright

CC-BY-NC-ND-4.0

The text only may be used under licence CC BY-NC-ND 4.0. All other elements (illustrations, imported files) are “All rights reserved”, unless otherwise stated.

Top of page
Search OpenEdition Search

You will be redirected to OpenEdition Search