Orthography-Induced Length Contrasts in the Second Language Phonological Systems of L2 Speakers of English: Evidence from Minimal Pairs

1.1 Orthographic effects on L2 speech production

The orthographic forms, that is to say spellings, of L2 words and sounds can help L2 speakers perceive and produce L2 speech. For instance, learning the phonological form of words in a novel language is easier if spoken input is accompanied by written input (Davidson, 2010; Erdener & Burnham, 2005), and written representations can help even proficient L2 speakers to better distinguish difficult L2 contrasts in pseudowords (Escudero, Hayes-Harb, & Mitterer, 2008; Escudero & Wanrooij, 2010). However, L2 orthographic forms can also lead to non-targetlike production, for instance when L2 speakers add a sound corresponding to a so-called “silent letter,” such as a [l] in walk (/wɔːk/ in Standard British English; Bassetti & Atkinson, 2015). The most frequently reported orthographic effect is sound substitution, for instance when Spanish_L1 speakers of American English_L2 pronounce a [t] or a [d] in words spelled with letters <t> or <d>, which native speakers produce as flaps (Vokic, 2011). Substitutions are often associated with incongruences between grapheme-phoneme correspondences in the first and second language, whereby the two languages map the same grapheme (single letter, digraph or trigraph) onto different phonemes. For instance, English learners of L2 Spanish may produce the Spanish letter <v> as [v], following the grapheme-phoneme correspondences of their first language, although in Spanish the phoneme /v/ does not exist and <v> represents /b/ (Zampini, 1994). The next section reports recent research that argued that English orthographic forms can result in Italian speakers of English_L2 making a phonological contrast in their English_L2 production that does not exist in the English language because of English orthographic forms.

1.2 Orthographic effects on the duration of consonants and vowels in Italian_L1 speakers’ English_L2 production

1.3 Effects of orthography on Italian_L1 speakers’ production of English_L2 consonants and vowels

Previous research (Bassetti, 2017) found that native Italian (Italian_L1) learners of English as a Second Language (English_L2) with 10 years’ learning experience produce the same English consonant as long or short depending on the number of consonant letters in its spelling, for instance producing the [t] in kitty (/ˈkɪti/ in Standard British English) as longer than the [t] in city (/ˈsɪti/), whereas there are no differences in duration in native English controls. This inter-orthographic effect must happen because Italians are transferring L1 grapheme-phoneme correspondences to L2 English, thereby interpreting English double consonant letters as representing geminates, as in their native orthography. This means that Italian speakers of English_L2 produce consonant length differences even though such differences do not exist in the target language English, and rely on the number of consonant letters in English orthographic forms to determine the duration of English consonants.

With regard to orthographic effects on vowel duration, Piske, Flege, MacKey, and Meador (2002) first suggested that Italian native speakers rely on orthography for the pronunciation of some vowels in the production of English_L2 pseudowords, for instance pronouncing English [ɪ] as [i] because the letter <i> is pronounced as [i] in Italian. Bassetti and Atkinson (2015) found that Italian_L1 learners of English_L2 with 10 years’ learning experience produce the same vowel as longer when spelled with a double vowel letter than a singleton vowel letter, for instance producing [uː] as longer in moon than in June (both /uː/ in Standard British English: /muːn/ and /dʒuːn/), whereas there are no duration differences in English native controls. This happens because Italians correctly recode double vowel letters such as <oo> as a long vowel, but also recode the <u> in June as a short vowel, without realizing that the grapheme <u_e> represents a long vowel too. This means that these Italian_L1 learners of English_L2 (1) use length to distinguish English vowels—even though contrastive vowel length does not exist either in their native phonology or in the English phonology—and (2) rely on the number of vowel letters in L2 orthographic forms to determine the duration of vowels, overgeneralizing the rule that a single vowel letter represents a short vowel to words where the single vowel letter is part of a <V_e> grapheme that represents a long vowel. This paper will then expand on Italians’ use of English orthographic forms in determining the length of target consonants and vowels.

1.4 This study

This study was aimed at testing whether the alternation between long and short consonants and vowels in the production of Italian_L1 speakers of English_L2 should be considered as a phonological contrast in the L2 phonological systems of these speakers. We achieved this by investigating whether they produce English homophonic word pairs as minimal pairs distinguished by a short or long sound (consonant or vowel) when the same target sound is spelled with a single letter in one word and a digraph in the other.

There is previous evidence that Italian_L1 speakers of English_L2 produce the same target sound as short when spelled with one letter and long when spelled with two letters (for consonants, see Bassetti, 2017; for vowels, see Bassetti & Atkinson, 2015). However, these studies used orthographic near-minimal pairs. In such pairs, the two words’ spellings represent the same target consonant or vowel with one letter or with two letters (e.g., /t/ in city and kitty), and the sounds immediately preceding and immediately following the target sound were controlled (/ɪ_i/ in both city and kitty), but the two words had differences other than the spelling of the target consonant (initial /s/ or /k/). With such near-minimal pairs the phonological environment is not as tightly controlled as with minimal pairs, so the effect found may theoretically be due to factors other than spelling. For instance, the words in a near-minimal pair may differ in the number of phonemes within a consonant cluster or syllable, and this may affect the duration of the target sound. This is because the duration of speech subunits decreases as the number of subunits in a larger unit increases, such as the number of consonants in a consonant cluster and the number of phonemes in a syllable (Lindblom, Lyberg & Holmgren, 1977). The use of minimal pairs eliminates these potential confounding variables. The aim of this study was then to test previous findings of orthographic effects on the duration of consonants and vowels in semi-minimal pairs by using orthographic minimal pairs. Furthermore, the study includes more pairs than previous studies by Bassetti and colleagues (Bassetti, 2017; Bassetti & Atkinson, 2015). Finally, research that investigates orthographic effects on L2 production using L2 minimal pairs has mostly used pseudowords (e.g., Escudero et al., 2008; Piske et al., 2002). Research on pseudowords is important in illuminating how orthography affects phonology at the point of word learning; however, using real words is important for understanding what happens beyond initial learning.

This study was then aimed at extending previous research on effects of number of letters on target sound length by using word pairs in which the target consonant or vowel occurs in a fully controlled phonological environment, by using real words, and by using a larger number of items than previous studies. Materials consisted of 33 orthographic minimal pairs, that is to say English homophonic word pairs where the same consonant or vowel is spelled with one letter in one word and a digraph in the other, so that if number of letters affects sound length the two homophonic words would be produced as a minimal pair distinguished by a long and a short consonant or vowel. With regard to consonants, we predicted that Italian_L1 speakers of English_L2 would produce for instance the homophonic words finish and Finnish (both /ˈfɪnɪʃ/ in Standard British English) as [ˈfɪnɪʃ] with a short [n] and [ˈfɪnːɪʃ] with a geminate [nː] respectively, because they would produce the double letters <nn> as a geminate [nː], according to the grapheme-phoneme correspondences of their native language. With regard to vowels, we predicted that the number of vowel letters would lead participants to apply a phonological contrast between long and short vowels to homophonic word pairs, for instance producing the homophones scene and seen (both /siːn/) with a short and a long vowel respectively, because the vowel is spelled with a single letter in one word and with a digraph in the other. Results would then demonstrate that English orthographic forms—number of letters—lead Italian_L1 speakers of English_L2 to establish a singleton-geminate consonant contrast that does not exist in the English language, and to distinguish two English homophonic words by means of a short-long vowel contrast that does not exist in either their L1 or L2.

The second main aim of the study was to investigate whether naturalistic exposure reduces the impact of orthography on L2 speech production. A shortcoming of current research on orthographic effects on L2 speech production is a focus on naïve learners of novel languages (e.g., Davidson, 2010; Escudero et al., 2008; Hayes-Harb, Nicol, & Barker, 2010; Pytlyk, 2011) and beginner learners, typically with one or two years of L2 exposure (e.g., Young-Scholten & Langer, 2015; Zampini, 1994). There has been little research on learners with longer length of study (e.g., 10 years in both Bassetti, 2017, and Bassetti & Atkinson, 2015) and almost no research on late bilinguals who have finished studying the language (with the exception of research using pseudowords, e.g., Piske et al., 2002; but see Vokic, 2011). This study then aimed at testing whether exposure to the target language in a naturalistic environment can reduce orthographic effects in speech production. We achieved this by comparing instructed L2 learners with late bilinguals who had been living in an English-speaking environment for an average of six years.

Finally, we investigated the relationship between the duration of short and long consonants and vowels in L2 English and in L1 Italian. We achieved this by measuring the duration of consonants and vowels in Italian minimal or near-minimal word pairs containing consonants and vowels similar to those tested in the English task, such as vano-vanno and di-zii.

2.1 Design

Orthographic effects were tested using orthographic minimal word pairs, in which the same target sound (a consonant or a vowel) was spelled with a single letter or a digraph, and calculating the ratio between the duration of the consonant spelled with double letters and the duration of the consonant spelled with a single letter. Language background was a between-group factor: the learner group were Italian_L1 learners of English_L2, the bilingual group were Italian_L1-English_L2 late bilinguals, and the native group were English native speakers.

2.2 Participants

Participants were 30 Italian high-school learners of English (learners: males = 18, mean age = 16.7 years, SD = 0.5), 30 Italian-English late bilinguals (bilinguals: males = 10, mean age = 32.5, SD = 7.3), and 30 British English native speakers (natives: males = 18, mean age: 30.8, SD = 13.5). No participant reported visual, reading, or language difficulties.

The learner group were native speakers of the Roman variety of Standard Italian who were instructed late learners of English. In this variety, geminates are generally twice as long as singleton consonants (Esposito & Di Benedetto, 1999). Two learners were bilingual from birth with a language that has contrastive vowel length (Hebrew or Tagalog), but their performance in the vowel production task did not differ from other participants (p = 0.637). Nobody had studied a second language with contrastive length. Participants were attending the fourth year at a liceo classico or scientifico (state-run non-vocational high school) in Rome. They were enrolled in eight different classes at three different schools. They were studying English language and literature as a compulsory school subject, with three 1 h sessions a week, using British English textbooks. They had been studying English on average for 10 years and four months (SD = 25 months) with a median age of onset of acquisition of six years. Their proficiency was measured using the Oxford Placement test (Allan, 1992), which tests grammar and listening comprehension. Most test-takers (54%) performed at intermediate level (B1 or B2), another 35% were beginners (A1 or A2), and three participants were proficient (C1; four participants were absent). They had little exposure to native teachers (the median reported length of study with native teachers was 6.2 months, ranging 0–156 months). Most participants (n = 18) had never been in an English-speaking country, and the others’ median stay abroad duration was 1.4 months (range 10–60 days). Respondents reported spending twice as much time listening to English (including music and movies) than reading English (Mdn_listening = 4 h per week, range: 0.5–28; Mdn_reading = 2, range: 0–28). All respondents but one considered a native-like pronunciation important or very important. Participants were almost equally keen on a British English and an American pronunciation, with a slight preference for the former (on a 7-point scale where 7 = “very much,” M_British = 4.9, SD = 1.4; M_American = 4.6, SD = 1.2).

The bilingual group were instructed late learners of English who had been living in the UK for an average of six years and seven months (range: 3–24 years), and were recruited in the UK (London or the Midlands). They were native speakers of various Southern or Central varieties of Italian, for which gemination is well described and uniformly realized (Giordano & Savy, 2012), and they had lived in Italy until at least high-school graduation. Three participants had studied an L2 with contrastive consonant length at beginner level (Japanese or Corsican), but their performance in the consonant production task did not differ from other participants in the same group (p = 0.630). The bilinguals had studied English as a school subject in Italy for a median of eight years (range 3–13), with a median age of onset of acquisition of 11, mostly with Italian native teachers (four had also studied with an English native teacher). Their English proficiency was measured using self-ratings on the Common European Framework of Reference for Languages for five competencies (writing, spoken production, spoken interaction, reading, and listening). Most participants rated themselves as “proficient” (C1 or C2; ranging from a minimum of 59% of participants in spoken production to a maximum of 83% in reading), with the rest being intermediate (B1 or B2). Respondents reported spending three times longer reading and listening to English than Italian (Reading: Mdn_English = 15 h per week, range: 1–84, Mdn_Italian = 5, range: 1–15; Listening: Mdn_English = 14, range: 0–84, Mdn_Italian = 4 h, range: 0–50). They reported similar amounts of speaking in the two languages (Mdn_English = 25, range: 5–70, Mdn_Italian = 20, range: 2–80). Seventy-nine percent of respondents (n = 24, six answers were missing) considered a native-like pronunciation important or very important, 21% did not.

The native group were native speakers of British English recruited in the UK (London or the Midlands), who mostly reported their native variety as Standard or Received Pronunciation (n = 15) and Southern (n = 10); the other varieties were Yorkshire (n = 3) and Midlands (n = 2). One participant was bilingual from birth with a language that has contrastive consonant length (Gujarati), and two further participants were bilingual from birth with a language that has contrastive vowel length (Arabic or Serbo-Croatian); a few had studied an L2 with contrastive consonant length (n = 2) or vowel length (n = 6). However, none of these English speakers performed differently from the rest of the group.

Participation was voluntary and compensated. Consent was obtained from all participants, and from parents in the case of participants under the age of 18. The study received ethical approval by the Humanities and Social Sciences Research Ethics Committee of the University of Warwick.

2.3 Materials

Materials for the English word reading-aloud task were 33 orthographic minimal or near-minimal pairs, and materials for the Italian word reading-aloud task were 18 minimal or near-minimal word pairs. Materials are provided in the Appendix.

2.4 Task and procedure

Participants were tested individually in a sound-attenuated or quiet room in a school in Rome or at a university in the UK. They received a printed list of English phrases or sentences, each containing an underlined target word. Phrases and sentences were used to help readers identify the target words, and were taken from the British National Corpus Online service. For example, the word knows appeared inside the phrase He knows I never argue. Participants read the target word aloud within the carrier sentence I say ___ and pause three times. The carrier sentence presented the target word in nuclear position within the intonational unit, which helped to obtain consistent speech rates. The three repetitions were used to obtain a mean duration, as a single production may be unreliable (Flege, 1995). At the end of the session, Italian participants also performed the Italian word reading task. In this task, target words were presented in a short phrase or sentence extracted from the CoLFIS corpus (Bertinetto et al., 2005) or created on purpose. Each word was repeated three times in the carrier sentence Dico__senza pausa (“I say __ without pausing”). In both English and Italian word reading tasks, items were presented in one of four different orders. Responses were recorded using a Zoom H4N Pro digital recorder connected to a Shure SM10A headset microphone or an AKG HSD171 headset with dynamic microphone, or using an Alesis Multimix 12 Firewire mixer connected to a Røde NT2-A microphone.

2.5 Data analysis

Sounds that could not be acoustically analyzed were eliminated from statistical analysis (English consonants: 4.4%, or 151 out of 3,420 items; English vowels: 1.3%, or 34 out of 2,520; Italian sounds: 0.4%, or 4 out of 1,080). This was due to sudden noises (such as a chair moving, or a participant moving a sheet of paper), or to deviant pronunciations (when a segment other than the target one was pronounced differently within the word pair). Additionally, outliers that were 2.5 SDs away from the participant’s mean (see Selst & Jolicoeur, 1994) we eliminated from the analysis (English consonants: 1.6%; English vowels: 0.4%; Italian consonants: 0.2%: no Italian vowel data was eliminated).

The duration of each target consonant or vowel was measured using the Praat software (Boersma & Weenink, 2016). Consonant duration was measured following procedures in Turk, Nakai & Sugahara (2006). For vowels, the beginning was marked as the onset of a clearly defined formant pattern in correspondence of coherent periodicity of the waveform. The end was marked where the formant pattern (especially F1 and F2) suddenly disappeared or became remarkably weaker. For each word, the mean duration of the target sound was obtained from measures of the three repetitions. Acoustic measurements were performed by four trained phoneticians (including two authors), whose intraclass correlation coefficient was 0.96, 95% CI [0.95, 0.97], p < 0.001.

3.1 English consonant duration

For each homophonic word pair, a CC:C ratio was calculated by dividing the mean duration of the consonant spelled with double letters by the mean duration of the consonant spelled with a single letter. Results are shown in Table 1.

OPEN IN VIEWER

Table 1.

Mean CC:C ratio for consonant duration as a function of Language Background.

Language Background	Position		Overall (SD)
	Legal in Italian (SD)	Illegal in Italian (SD)
English native speakers	1.04 (0.21)	1.07 (0.30)	1.06 (0.26)
Italian learners of EnglishL2	1.39 (0.49)	1.12 (0.29)	1.27 (0.44)
ItalianL1-EnglishL2 bilinguals	1.38 (0.45)	1.11 (0.34)	1.26 (0.42)

SD: standard deviation.

A one-way analysis of variance (ANOVA) with Language Background (Learner, Bilingual, Native) as a between-group variable revealed that the mean CC:C ratio differed across groups, F₁(2, 87) = 30.46, p < 0.001, ges = 0.41, F₂(2, 36) = 14.57, p < 0.001, ges = 0.26. Post hoc comparisons with Tukey adjustment revealed that the ratio was larger in the bilingual group than in the monolingual control group, and in the learner group than in the monolingual group, both ps < 0.001. This reflects the fact that C- and CC-consonants had almost the same duration in the English group (mean CC:C ratio = 1.06, SD = 0.26). The two L2 speaker groups (bilinguals and learners) did not differ, p = 0.857, reflecting the fact that they had almost the same CC:C ratio (M_learners = 1.27, SD = 0.44; M_bilinguals = 1.26, SD = 0.42).

To test whether the legality of the position of the target consonant in Italian affected gemination in the Italian groups, a mixed ANOVA was run with Language Background (Learner, Bilingual, Native) as a between-group factor and Consonant Position (Legal, Illegal) as a within-group factor (see Table 1). The main effect of Language Background, F₁(2, 87) = 29.22, p < 0.001, ges = 0.25, F₂(2, 34) = 23.52, p < 0.001, ges = 0.39, and the main effect of Consonant Position, F₁(1, 87) = 54.31, p < 0.001, ges = 0.24, F₂(1, 17) = 16.69, p < 0.001, ges = 0.35, were qualified by an interaction, F₁(2, 87) = 18.35, p < 0.001, ges = 0.17, F₂(2, 34) = 12.05, p < 0.001, ges = 0.25. Post hoc comparisons with Tukey adjustment revealed that the CC:C ratio was larger in legal than illegal positions in the learner group, p < 0.001, and in the bilingual group, p < 0.001, showing that Italians produce longer consonants in positions where gemination is attested in Italian native words than in positions where it is not. Target consonant position had no effect on duration ratio in the English native group, p = 0.226.

3.2 English vowel duration

For each homophonic word pair, a VV:V ratio was calculated by dividing the mean duration of the vowel spelled with a vowel digraph by the duration of the vowel spelled with single letter. Table 2 shows the results.

OPEN IN VIEWER

Table 2.

Mean VV:V ratio for vowel duration as a function of Language Background.

Language Background	Type of Digraph		Overall (SD)
	VV (SD)	V1V2 (SD)
English native speakers	1.05 (0.17)	1.04 (0.25)	1.05 (0.23)
Italian learners of EnglishL2	1.13 (0.38)	1.11 (0.31)	1.12 (0.33)
ItalianL1-EnglishL2 bilinguals	1.20 (0.53)	1.08 (0.27)	1.11 (0.37)

SD: standard deviation; VV: double vowel letter ; V₁V_2: two vowel letters.

A one-way ANOVA with Language Background (Learner, Bilingual, Native) as a between-group variable revealed that the mean VV:V ratio differed across groups in the by-subject analysis, F₁(2, 87) = 6.02, p = 0.004, ges = 0.12, F₂(2, 26) = 2.94, p = 0.07, ges = 0.08. Post hoc comparisons with Tukey adjustment revealed that the ratio was larger in the bilingual group than in the monolingual control group, p = 0.007, and in the learner group than in the monolingual group, p = 0.003. This reflects the fact that V- and VV-vowels had almost the same duration in the English group (mean VV:V ratio = 1.05, SD = 0.23). The two L2 speaker groups (bilinguals and learners) did not differ, p = 0.950, as they had almost the same VV:V ratio (M_learners = 1.12, SD = 0.33; M_bilinguals = 1.11, SD = 0.37).

To test whether there was a difference in the duration of vowels spelled with a double vowel letter (VV, e.g., seen) and vowels spelled with a digraph composed of two different letters (V₁V₂, e.g., soul), a mixed ANOVA was run with Language Background (Learner, Bilingual, Native) as a between-group factor and Type of Vowel Digraph (double letters or VV, two letters or V₁V₂) as a within-group factor (see Table 2). The main effect of Language Background was significant in the by-subject analysis and approached significance in the by-item analysis, F₁(2, 87) = 5.32, p = 0.007, ges = 0.64, F₂(2, 24) = 3.09, p = 0.064, ges = 0.09. The main effect of Type of Digraph was significant in the by-subject analysis, F₁(1, 87) = 5.43, p = 0.022, ges = 0.03, F₂(1, 12) = 0.82, p = 0.382, ges = 0.04. However, both main effects were qualified by an interaction in the by-subject analysis, F₁(2, 87) = 3.16, p = 0.048, ges = 0.03, F₂(2, 24) = 1.64, p = 0.214, ges = 0.05. This shows that VV-vowels were longer than V₁V₂ vowels in both the learner group (p < 0.001) and the bilingual group (p = 0.006). Type of vowel spelling had no effect on native speakers’ vowel duration, p = 0.995.

3.3 Duration in Italian_L1 and in English_L2

Durations of Italian consonants and vowels were measured in the same way as English ones. For each Italian word pair, a long sound-to-short sound ratio (CC:C or VV:V) was calculated by dividing the mean duration of the target sound spelled with a digraph by the duration of the single-letter counterpart. The mean CC:C ratio in Italian was 2.04 (SD = 0.97; 2.47 in intervocalic position; 1.17 in word-final position). The mean VV:V ratio was 1.60 (SD = 0.49).

3.4. Orthographic consistency

To disentangle the effects of the orthographic form of words from the effects of the phonological form of related words in the native language, two repeated-measures t-tests compared the CC:C ratios in L1–L2 orthography-congruent cognate or loanword pairs with the ratio of the other word pairs containing the same target consonant or vowel. With regard to consonants, there were four pairs containing a target sound spelled <VnnV>: one contained an orthographically congruent word (Finnish, It. finnico, /ˈfinːiko/), one contained an incongruent word (manners, It. maniere, /maˈnjɛre/), and two contained no cognates (tenner, banned). Although level of L1–L2 orthographic congruency affected consonant duration, F(2, 176) = 18.27, p < 0.001, ges = 0.08, results were mixed, as the orthography-congruent pair had a larger ratio than both the incongruent pair and the unrelated pairs (both ps = 0.001), but the incongruent pair did not differ from the unrelated pairs (p = 0.448). Also, no differences were found between the pair containing the orthography-incongruent carrot (It. carota, /kaˈrɔta/) and the other <VrrV> pair, p = 0.063, and the incongruent word pair had a descriptively larger CC:C ratio than the unrelated pair. With regard to vowels, the VV:V ratio was smaller in the pair containing the loanword jeans (/dʒins/) than the other word pair containing <CeaC>,  p < 0.001.

4.1 Effects of consonant spelling on consonant duration in orthographic minimal pairs

The Italian_L1 speakers of English_L2 produced the same consonant 1.26–1.27 times as long when spelled with double letters than when spelled with a single letter, and specifically 1.38–1.39 times as long in intervocalic contexts. No effects of orthography were found in native speakers, whose mean CC:C ratio was close to one. These results are in line with previous findings that compared near-minimal pairs such as city-kitty (Bassetti, 2017), confirming that Italian native speakers decode English double consonant letters as geminate consonants, in line with the grapheme-phoneme correspondence rules of their native orthography, where <CC> corresponds to geminate /Cː/.

In line with literature on the phonetic correlates of gemination in Italian, when tested in their native language, this study’s native speakers of the Roman variety of Italian produced geminates that were twice as long as singleton consonants. These speakers’ geminate-singleton ratio is then larger in Italian_L1 than in English _L2. There may be various reasons for this smaller ratio in English, including probably the interference between orthographic input that contains double consonant letters and native spoken input where longer consonants are not attested.

Crucially, this study found this difference in duration in orthographic minimal pairs, that is to say homophonic words that contain the same target sound spelled with a single letter or a digraph. Controlling the phonological environment by means of the use of minimal pairs supports the claim that the effect can only be attributed to spelling. The relationship between the Italians’ two categories and the native speakers’ category were slightly different from previous findings. Bassetti (2017) reported that Italian_L1 speakers of English_L2 have established two phonological categories, one shorter and one longer than the native English category. Results from this study revealed that consonants spelled with double letters were produced as longer in the Italian participants (M_bilinguals = 103 ms, SD = 45; M_learners = 100, SD = 39) than the native speakers (M = 89, SD = 45), but consonants spelled with a single letter in the Italian participants were not shorter than in the native speakers (M_bilinguals = 89, SD = 45; M_learners = 85, SD = 37; M_natives = 87, SD = 46). Whatever the phonetic durations, it is safe to conclude that, in the phonological system of these English_L2 speakers, words such as finish and Finnish are in fact minimal pairs—/ˈfɪnɪʃ/ versus /ˈfɪnːɪʃ/—distinguished by a singleton or geminate consonant. Because the two words only differ in the number of letters in the spelling of the target sound, this is an orthography-induced phonological contrast.

Apart from the effects of L1 and L2 orthography, L1 phonological constraints also played a role. Some of the target consonants appeared in contexts where gemination is legal in native Italian words—intervocalic position—and others in positions that are illegal in Italian native words—word-final or syllable coda. The mean CC:C ratios in Italians’ English_L2 production were 1.39 in legal positions and 1.12 in illegal position. The ratio in intervocalic position is then similar to that reported in Bassetti’s study of near-minimal pairs, where all targets were intervocalic (Bassetti, 2017). The ratio is also smaller than the Roman participants’ Italian CC:C ratio, which was 2.47 in intervocalic position. English_L2 consonants in positions that are illegal in Italian were also lengthened, but to a much smaller extent than those in legal position. This reflects a similar pattern in Italian participants’ production of Italian geminates, whose CC:C ratio in illegal position was just 1.17. It appears that the legality or illegality—according to L1 phonotactic rules—of the resulting geminate consonant in the first language modulates a gemination process in L2 English, which appears to occur much less or not at all in contexts where the geminate consonant would not be legal in a native language word. All these phenomena show that the orthographic effect is not a blanket effect, but there is interaction among L2 orthography, L1 orthography-phonology correspondences, and L1 phonology.

Although the legality of the position of gemination in Italian phonology affected Italians’ English_L2 geminate durations, other relations between L1 and L2 gemination were not found, as follows. First, gemination occurred both in cognate words and in unrelated words, and comparisons of cognate and non-cognate words produced mixed results. This confirmed previous findings (Bassetti, 2017) that gemination in an English word is not simply caused by gemination in a native cognate word, although the presence of a cognate word with gemination may reinforce the orthographic effect. Second, participants’ English_L2 CC:C ratio was smaller than their Italian_L1 CC:C ratio. It appears that Italians’ gemination in English is present, but not as strong as in their native language, probably because such durational differences are not present in the spoken input.

Length of exposure to the target language in a naturalistic context did not modulate the effects of orthography on consonant production, as instructed L2 learners and late bilinguals with an average residence of over six years in a British English-speaking environment had almost identical CC:C ratios. It was expected that orthographic forms would affect learners more than bilinguals, because instructed learners’ L2 input includes both large amounts of orthographic input and spoken input produced by other non-native speakers (classmates and teachers) who may also produce geminates. The late bilinguals instead had been exposed to a native-speaking environment for years, and native speakers do not produce longer or shorter consonants in relation to their spelling, as shown both by English participants in this study and elsewhere (Bassetti, 2017). However, it is also possible that current learners of English are exposed to more spoken input than previous generations, and indeed learners reported spending twice as long listening to than reading English, as this included pop songs, movies and the internet. The late bilinguals in this study learnt English earlier, and this likely stronger reliance on written input during the learning process may be one reason why orthographic effects were found despite lengthy naturalistic exposure.

The fact that learners and bilinguals did not differ is also interesting because targets were real words. Research shows that experienced L2 speakers are affected by orthographic forms when learning pseudowords (e.g., Escudero et al., 2008). Finding such effects in real words shows that orthographic effects extend well beyond the initial stage of learning and continue to influence the production of known words even after years of exposure to a native-speaking environment. It appears that once the phonological contrast between geminates and singletons is established during the learning process, Italian speakers of English continue to make this contrast despite its absence in the spoken input they are exposed to.

4.2 Effects of vowel spelling on vowel duration in orthographic minimal pairs

Although orthography did not affect vowel duration in English native speakers, the Italian_L1 speakers of English_L2 showed orthographic effects on vowel duration. Italian participants produced the same English target vowel on average 1.11–1.12 times as long when spelled with a vowel digraph than when spelled with a single vowel letter. Effects of orthography on duration were much weaker than in consonants, possibly reflecting the facts that vowel length is not contrastive in Italians’ native phonology, that English vowels differ qualitatively as well as quantitatively, and that the correspondence between number of letters and English vowel length is far less regular than the correspondence between number of letters and Italian consonant length.

In line with predictions, Italian_L1 speakers of English_L2 produced a longer vowel in English words spelled with a double vowel letter than in homophonic words spelled with a single vowel letter, for instance producing a longer vowel in seen than in scene. Italian participants’ long vowels were on average 1.11–1.12 times as long as short vowels, in line with previous findings of 1.14 times longer vowels with near-minimal pairs (Bassetti & Atkinson, 2015). This confirms that Italian native speakers use vowel length contrastively in L2 English, even though vowel length is not contrastive in English, and produce the same vowel as longer or shorter depending on the number of letters present in the spelling of the word. As with the consonant data above, these results confirm and extend previous findings, as the use of minimal instead of near-minimal pairs allows us to better attribute the cause to orthography, rather than other potential confounding variables.

Italian participants, however, did not apply orthographic rules uniformly, but their production seemed to reflect a number of other factors, possibly including their use of statistical probabilities in English grapheme-phoneme correspondences, and effects of Italian phonology, as explained below. First, the Italians’ English VV:V ratio was significantly larger in vowels spelled with double letters than with vowel digraphs composed of two different vowel letters, probably reflecting the fact that <VV> mostly represents a long vowel /V:/ in English orthography, whereas correspondences between other types of digraphs and vowel length are less regular. Second, many of the single-letter vowels in this study in fact contained the digraph <V_e>, or “silent e,” as in rode (/rəʊd/). Personal conversations with Italian school teachers confirmed that Italian learners are not aware of this non-linear correspondence. However, they may have acquired the correspondence implicitly. Third, in Italian words <VV> sequences correspond to quickly re-articulated vowels (Bertinetto & Loporcaro, 2005), as in sci-scii (/ʃi/, “ski,” and /ˈʃii/, ski-PRS.2.SG, “[you] ski”) discussed above. This orthography-phonology correspondence in Italian could reinforce Italians’ tendency to recode English_L2 <VV> sequences as long vowels, or possibly as quickly re-articulated vowels. The comparison between the VV:V ratio in Italian participants’ production of L1 Italian and L2 English provides further clues. Italians’ VV-vowels were 1.6 times as long as V-vowels in their native language Italian, compared with 1.11–1.12 times in English production.

Finally, as with consonant results, length of exposure to the target language in a naturalistic context did not modulate orthographic effects on vowel production, as no differences were found in the VV:V ratio between learners and bilinguals.