Identification of consonant length in L2 Italian: the role of cross-linguistic influence

Italian contrasts singleton and geminate consonants, a distinction that often poses difficulties for second language (L2) learners. The present study examines how first-language (L1) background shapes the perception of Italian consonant length by comparing learners from five linguistic backgrounds. A total of 104 adult learners of Italian (Finnish, Czech, German, Slovak, Spanish) and 21 L1 Italian listeners as a control group completed a perception identification task with 54 pseudowords controlled for consonant type and stress position. Participants categorized each auditory stimulus by selecting between the singleton and geminate orthographic response options (e.g., sàppolo vs. sàpolo). The results reveal high overall identification accuracy (81%), but substantial differences across learner groups. Learners whose L1 encodes consonantal and/or vocalic quantity contrasts (Finnish, Czech, Slovak) produced identification patterns closer to those of Italian listeners, while learners from languages without systematic length contrasts (Spanish, German) showed less accuracy. The observed hierarchy of accuracy confirms the importance of cross-linguistic influence in the perception of duration-based contrasts. While stress position and proficiency played more limited roles, consonant type also influenced performance. In particular, sonorants were identified more accurately than obstruents, and several consonant-specific tendencies reflected differences in the learners' L1 phonological inventories. A comparison with a previous discrimination study, conducted with the same participants, revealed higher accuracy in identification across all groups, suggesting that identification performance reflects phonological categorization processes and may additionally have been facilitated by the written responses available during the task. The findings highlight the role of cross-linguistic influence in the perception of Italian consonant length and contribute to a broader understanding of how L2 learners map continuous acoustic variation onto phonological categories.

The present study investigates whether listeners with different language backgrounds (Finnish, Czech, Slovak, German, and Spanish) successfully identify singleton–geminate contrasts in Italian and whether the resulting patterns resemble those observed in a previous discrimination task involving the same participants (Pešková and Chládková, 2026).

Perception plays a central role in language acquisition because phonological categories are established through the interpretation of acoustic input. During L1 language acquisition, infants gradually attune their perception to the contrasts that are relevant in their first language(s) (L1), while sensitivity to non-native distinctions decreases over time (e.g., DeCasper and Spence, 1986; Werker and Lalonde, 1988; Tsushima et al., 1994; Dehaene-Lambertz et al., 2006; Kisilevsky et al., 2009). These perceptual categories subsequently guide speech production and form the basis for later language learning. In the domain of (consonantal) quantity, sensitivity to durational contrasts emerges relatively early and contributes to the acquisition of phonological length distinctions (e.g., Vihman and Majorano, 2017).

Second language (L2) acquisition differs fundamentally from the L1 developmental trajectory. Instead of building phonological categories in an initially open system, L2 learners must establish new categories against the background of previously acquired L1 structures. As noted by Trubetzkoy (1939/1969), foreign sounds are filtered through the phonological system of the L1—a phenomenon often described as the “phonological sieve.” Major models of non-native speech perception adopt similar assumptions. The Perceptual Assimilation Model (PAM: Best et al., 1988; PAM-L2: Best and Tyler, 2007), as well as the Speech Learning Model (SLM, Flege, 1995; SLM-revised, Flege and Bohn, 2021), propose that L2 sounds are perceived in relation to existing L1 categories. According to these models, the more similar an L2 sound is to an existing L1 category, the more likely it is to be assimilated to that category. In SLM, this also reduces the likelihood of establishing a new L2 category. Because such comparisons are assumed to operate primarily at the level of language-specific phonetic categories, the models predict an advantage for learners whose L1 contains quantity contrasts similar to those of Italian, thereby facilitating positive L1-to-L2 transfer.

However, transfer may operate at a more abstract level. According to the Feature Hypothesis (McAllister et al., 2002), the acquisition of L2 contrasts depends not only on whether a particular sound exists in the learner's L1, but also on whether the relevant phonological feature—such as duration or quantity—is employed contrastively. McAllister et al. (2002), for example, showed that Estonian listeners, whose L1 encodes rich quantity distinctions, identified Swedish vowel length contrasts more accurately than English and Spanish listeners. Similarly, Pajak and Levy (2014) demonstrated that L1 experience with vowel length enhances listeners' ability to discriminate non-native consonant length contrasts, suggesting that perceptual sensitivity to duration may generalize across segment classes, including both vowels and consonants. More generally, evidence from L2 speech perception implies that perceptual learning may extend beyond individual segments and generalize across sounds that share a common phonological feature (De Jong et al., 2009). Taken together, these findings indicate that experience with duration in one segment class may facilitate the perception of quantity contrasts in another.

Within the Second Language Linguistic Perception model (L2LP; Escudero, 2005; van Leussen and Escudero, 2015; Escudero and Yazawa, 2024), learners must not only detect relevant acoustic cues. They have to acquire their language-specific weighting. Applied to the perception of Italian gemination, this means that learners must assign greater perceptual weight to consonant duration than to other potentially relevant acoustic cues, such as consonant quality, or prosodic position. In an identification task, this perceptual process may additionally have been supported by the orthographic response options.

The findings of the previous discrimination study (Pešková and Chládková, 2026) already suggested that the perception of Italian consonant length is shaped by multiple factors. The present identification study, including the same participants, provides an opportunity to further examine whether the learners rely primarily on L1 categories or whether transfer may also operate at the level of a more abstract phonological feature such as duration. In addition, the study considers the possible influence of orthography on the identification of L2 consonant length. Orthographic input has been shown to shape learners' phonological representations (e.g., Bassetti, 2017; Nimz, 2016; Pešková et al., 2017), and this effect may also be particularly relevant in Italian, where lexical geminates are transparently represented by double consonant letters.

A growing body of research indicates that L1 phonological structure plays an important role in the acquisition of quantity contrasts (e.g., Han, 1992; Hayes, 2002; Mah and Archibald, 2003; Hayes-Harb, 2005; Hardison and Saigo, 2010; Guillemot, 2018; Lee and Mok, 2018; Tsukada and Yurong, 2022; Tsukada and Hajek, 2023; McGill, 2024). At the same time, learners may develop at least partial sensitivity to durational contrasts even when such distinctions are absent from their L1, indicating that sensitivity to duration is learnable. In the context of L2 Italian, existing empirical studies on the acquisition of Italian geminates have examined learners from several linguistic backgrounds. Altmann et al. (2012), for example, investigated German learners of Italian using a speeded same–different discrimination task and reported an asymmetry between the vowel and consonant quantity, with learners showing greater sensitivity to vowel length. De Clercq et al. (2014) found that Dutch learners were able to distinguish Italian singleton and geminate consonants with relatively high accuracy, although the durational contrast remained less clearly separated than in L1 Italian. Tsukada et al. (2018), in their study of Japanese and Italian consonant length perception by native and non-native listeners, further demonstrated that learners with experience in a quantity language may perceive consonant length contrasts even in unfamiliar languages. More recently, Feng and Busà (2022) showed that Mandarin learners could distinguish Italian geminates to some extent, although their performance did not reach native-like levels.

The present study examines the identification of Italian consonant length by adult L2 learners from five L1 backgrounds that differ considerably in the role of segmental duration in their phonological systems (Table 1). Italian contrasts singleton and geminate consonants phonemically, and consonant length plays a central role in distinguishing lexical meaning. Vowel length is not phonemic but varies predictably depending on stress and syllable structure: non-final stressed open syllables typically contain a long vowel before singletons (e.g., /′pala/ [′pa:la] ‘shovel') (e.g., Bertinetto and Loporcaro, 2005; Marotta and Vanelli, 2021). Finnish contrasts both vowel and consonant length phonemically, with quantity differences functioning as a central organizing principle of the phonological system (e.g., Lehtonen, 1970; Suomi et al., 2008). Czech and Slovak encode phonemic vowel length, but Slovak also allows long liquids in syllable nuclei (/r:/, /l:/) which may provide additional experience with phonological quantity beyond vowel length (e.g., Šimáčková et al., 2012 for Czech; Hanulíková and Hamann, 2010 for Slovak). German displays a more complex quantity system in which vowel duration interacts strongly with vowel quality and syllable structure (e.g., Hütte [′hʏtə] ‘cottage' vs. Hüte [′hy:tə] ‘hats'); the length contrasts are generally restricted to the stressed syllable (e.g., Sendelmeier, 1981; Wiese, 2000). Finally, Spanish does not employ phonemic segmental length, although a durational contrast exists in the rhotics between the tap and trill, which contrast only in the intervocalic position and differ in manner and the number of short occlusions (e.g., Quilis, 1993; Hualde, 2005; Martínez Celdrán and Fernández Planas, 2007).

The present typological variation provides a useful testing ground for examining how L1 phonological structure influences the perception of L2 quantity contrasts. In particular, it allows us to examine whether experience with phonemic duration—either in both segment classes (Finnish) or in vowels only (Czech and Slovak)—facilitates the perception of consonant length contrasts in Italian. The identification task forms part of a larger research project on the perception and production of Italian geminates in L2 acquisition (Pešková, 2025).

Based on previous research on the acquisition of quantity contrasts in L2 Italian (e.g., Celata and Cancila, 2010; Kabak et al., 2011; Altmann et al., 2012; De Clercq et al., 2014; Sorianello, 2014; Einfeldt et al., 2019) and on the results of the preceding discrimination experiment with the same participants (Pešková and Chládková, 2026), the following hypotheses are formulated.

(H1) First language. In our previous discrimination study, learners' performance largely reflected the quantity system of their L1, supporting the assumption that the L1 remains active during L2 acquisition (e.g., Westergaard et al., 2016) and shapes perception through cross-linguistic influence (e.g., Flege, 1995; Flege and Bohn, 2021). Finnish learners achieved the highest accuracy, consistent with the presence of both vowel and consonant quantity contrasts in Finnish. Czech and Slovak learners performed somewhat less accurately but still relatively well, as their languages encode systematic vowel length contrasts. As reported previously, sensitivity to vowel duration may facilitate the perception of L2 consonant length through the transfer of duration-based cues (e.g., Mah and Archibald, 2003; Pajak and Levy, 2014). Thus, while the expected advantage of Finnish learners is consistent with the SLM predictions, the relatively strong performance of Czech and Slovak learners suggests that sensitivity to duration may generalize across segment classes. In contrast, learners with L1 German, where quantity contrasts are more restricted and interact with vowel quality, and especially learners with L1 Spanish, which lacks phonemic segmental length contrasts (with the exception of rhotics), showed lower accuracy. Accordingly, we expect the same hierarchy to emerge in the identification task: Finnish > Slovak > Czech > German > Spanish.

In addition to identification accuracy, reaction times will be measured as an index of processing efficiency. We expect L1 listeners to process Italian consonant length contrasts more efficiently and therefore to respond more quickly than L2 learners (e.g., Altmann et al., 2012).

(H2) Stress position. Prosodic prominence is often assumed to facilitate the perception of durational contrasts (e.g., Dmitrieva, 2017). In the discrimination study, consonant length contrasts were perceived most reliably in post-stress (the target consonant follows a stressed vowel, e.g., sà pp olo) and pre-stress positions (the target consonant precedes a stressed vowel, e.g., sa pp ò lo), and least reliably in unstressed contexts (the target consonant was farther from the stressed vowel, e.g., sa pp olò). This pattern emerged irrespective of whether learners' L1s have variable lexical stress (German, Spanish) or fixed stress patterns (Finnish, Czech, Slovak). In particular, geminate consonants following stressed vowels were easier to perceive. This pattern may reflect common lexical structures in Italian, where geminates frequently occur after stressed vowels (e.g., Mancini and Voghera, 1994; Giannini, 2003). Based on these findings, we expect the same hierarchy to emerge in the identification task: Post-stress > Pre-stress > Unstressed.

(H3) Consonant type (manner). Both Dmitrieva (2017) and our previous discrimination experiment indicated that the perceptibility of length contrasts varies across consonant types. In particular, sonorant consonants were perceived more accurately than obstruents, suggesting that their acoustic properties provide clearer cues for duration contrasts. We therefore expect the following hierarchy: Sonorants (nasals, laterals, rhotics) > Voiced obstruents (voiced stops, fricatives, affricates) > Voiceless obstruents (voiceless stops, fricatives, affricates).

In addition, we explore whether identification accuracy may be influenced by the phonemic inventories of the learners' L1s. Consonant types that are absent, infrequent, or realized differently in the L1 may be associated with lower identification accuracy than consonants that are well represented in the learners' phonological systems.

(H4) Proficiency. In the discrimination experiment, proficiency (CEFR; Council of Europe, 2018) showed only a weak tendency toward higher accuracy among more advanced learners. Consequently, we expect only limited effects of proficiency in the identification task as well, although learners at higher proficiency levels may still perform better than learners with lower proficiency (e.g., Hardison and Saigo, 2010; Altmann et al., 2012; Tsukada and Yurong, 2022): C (advanced) > B (intermediate) > A (beginner).

(H5) Perception task. The comparison between learners' performance in the previous discrimination task and their performance in the present identification task allows us to examine potential task-related differences in perception (see, e.g., Lee et al., 2026). Identification tasks require listeners to assign stimuli to phonological categories and may benefit from access to orthographic information (double letters for geminate) presented on the screen (all participants had at least a basic level of Italian literacy). In contrast, discrimination tasks rely more strongly on fine-grained acoustic processing and place greater demands on auditory memory, as listeners must compare two successive stimuli without written response options. Nimz (2016), for example, combined discrimination, production, and identification experiments in a study on Polish learners of German vowel length. The results showed that learners had difficulty discriminating vowel length differences in a phonetic discrimination task, whereas an identification experiment with manipulated real words revealed that length manipulations were detected relatively reliably. This pattern suggests that phonetic discrimination and phonological categorization may rely on partly different perceptual processes. Accordingly, we expect higher accuracy in the identification task than in the discrimination task.

Participants

The study included 104 adult learners of Italian and a control group of 21 L1 Italian speakers from different regions of Italy, who served as a reference group in the analyses. The learner sample comprised 23 Finnish speakers (2 M, 20 F, 1 non-binary; mean age = 35), 25 Czech speakers (6 M, 19 F; mean age = 29), 8 Slovak speakers (8 F; mean age = 27), 28 German speakers (6 M, 22 F; mean age = 27), and 20 Spanish speakers (2 M, 18 F; mean age = 26). Participants represented a range of proficiency levels in Italian, which were assessed using the DIALANG vocabulary size test (Alderson, 2005; Chapelle, 2006). The majority of participants across all L1 groups were classified at proficiency level B (≈57–75%), while levels A and C were less frequent and showed relatively similar distributions across groups.1

Participants were recruited through university networks, language programs, and online platforms. Information about their linguistic background was collected through a comprehensive language history questionnaire (Pešková, 2025; Pešková and Chládková, 2026 for further details). The comparatively small number of Slovak participants reflects an ad-hoc recruitment for this group and will therefore be interpreted with caution in cross-group comparisons.

All participants indicated that they had no history of hearing problems or speech impairments. Participation was voluntary, and written informed consent was obtained from all individuals prior to the experiment.

Stimuli

The test material consisted of 54 trisyllabic Italian pseudowords (Table 2). All items were prerecorded by a male L1 speaker of Standard Italian who produced the pseudowords in isolation using a uniform declarative (falling) intonation pattern to maintain consistent prosody across recordings. The same items were also used in the previous discrimination study. All pseudowords conformed to Italian phonotactic and orthographic conventions, while avoiding resemblance to existing lexical items. This design prevented participants from relying on lexical knowledge during the task. In addition, the stimuli were designed not to correspond to real words in any of the other languages tested in the study, thus minimizing potential cross-linguistic lexical activation.

Each target pseudoword formed part of a minimal pair differing in the duration of the intervocalic consonant (e.g., sà p olo vs. sà pp olo). Acoustic measurements confirmed a robust durational contrast between the two categories: across items, geminate consonants had an average duration of approximately 185 ms, whereas singleton consonants averaged around 96 ms (for detailed measurements see Pešková, 2025; Pešková and Chládková, 2026). These values are consistent with the typical realization of singleton and geminate contrasts in Standard Italian (see, e.g., Kingston et al., 2009, p. 299).

The stimulus set covered nine consonant categories, including two subgroups of obstruents and sonorants. The obstruent group comprised voiced and voiceless stops ([b], [p]), affricates ([tʃ], [dɜ]), and fricatives ([f], [v]). The sonorant group included a lateral ([l]), a rhotic ([r]), and a nasal ([m]). Each consonant type was provided in both singleton and geminate realizations. To examine prosodic conditions, the target length was embedded in three configurations, defined by the position of the target consonant and the stressed vowel (post-stress; pre-stress; unstressed). These contexts allowed us to test whether the perception of consonant length is sensitive to prosodic structure.

Procedure and data analysis

The experiment was administered in LabVanced (Finger et al., 2017). Participants were tested individually in soundproofed laboratories or quiet experimental cubicles at the participating universities. They were asked to listen to auditory stimuli and categorized each item by selecting one of two written response options displayed on the screen. The task consisted of a total of 54 randomized target trials and was conducted on a laptop. In each trial, participants heard a pseudoword and indicated whether it corresponded to the option with a singleton (e.g., sàpolo) or the option with a geminate (e.g., sàppolo) displayed on the screen. In addition, the experiment included 109 further items based on manipulated versions of the original stimuli. These items were designed to address separate research questions and will be reported in future work.

Responses were given using a computer mouse. At the start of each trial, participants placed the cursor on a designated starting point at the center of the screen. Once the cursor reached this position, the stimulus was played. Participants then moved the cursor toward the written response option (e.g., sàpolo – sàppolo) that matched their judgment. For consistency across trials, the item containing the geminate consonant was always presented on the right side of the screen.

The experimental design allowed us to record not only the categorical identification response but also reaction times (RTs), thereby capturing the temporal dynamics of the decision process. These measures provide additional insight into processing difficulty and the underlying mechanisms of decision-making during perception. RT analyses included both correct and incorrect responses. RT outliers were excluded by removing responses exceeding 20,000 ms (8 trials, affecting 7 participants).

Before the main experiment, participants completed a short practice session with feedback to familiarize themselves with the procedure. No time limit was imposed on responses.

Statistical analysis

Identification responses of the 54 unmanipulated stimuli per participant were coded as correct (1) or incorrect (0) and analyzed using generalized linear mixed-effects models (GLMERs) with a binomial error structure in R (R Core Team, 2024) using the lme4 package (Bates et al., 2015). Predicted probabilities and 95% confidence intervals were estimated using the ggeffects package (Lüdecke, 2018). Average marginal effects and standard errors on the probability scale were computed using the marginaleffects package (Arel-Bundock et al., 2024).

The main model examined the effects of L1 background, target (G vs. S), consonant type, and stress position on the perception of consonant length contrasts, including interactions with L1.

RT data were analyzed separately using linear mixed-effects models (LMERs) with log-transformed reaction times to account for skewed distributions. These models used the same fixed- and random-effects structure as the accuracy analyses.

Model complexity was increased stepwise following the recommendations of Matuschek et al. (2017). Random intercepts for participants and items were first included, followed by random slopes justified by the experimental design. More complex random-effects structures that resulted in singular fits were simplified by removing unsupported variance components until a converged, non-singular model was obtained. The final accuracy models included random intercepts for participants and items, participant-level random slopes for Target, Sound, and Stress Position, and item-level random slopes for L1 where supported by the data. For the RT analyses, the additional item-level random slopes for L1 were not supported and a simpler random-effects structure was therefore retained. Statistical significance was evaluated at p < 0.05.

Because the Slovak group comprised only eight participants, the robustness of the results was additionally assessed by repeating all analyses after excluding this group. The resulting fixed-effect estimates and overall pattern of findings remained materially unchanged, indicating that the conclusions were not driven by the inclusion of the small Slovak group.

Accuracy according to L1 and target

Across all L1 and L2 participants, the overall identification accuracy was 81% (N = 6,750 tokens). Figures 1, 2 present the observed and predicted identification accuracy across different L1 groups for geminate (G) and singleton (S) consonants. It is important to note that while L1 Italian listeners performed at near-ceiling levels (around 95%), L2 learners showed substantially lower and more variable accuracy. The fact that native listeners did not reach perfect accuracy is not unexpected in an identification task using pseudowords and may reflect normal response variability and the increased processing demands associated with unfamiliar lexical items. Across the groups, there was no significant difference between the identification of geminates (96.2%) and singletons (95.4%). This indicates that both categories were processed with comparable accuracy at the global level.

The predicted probabilities from binomial logistic regression (for the output of the model, see Supplementary Table 1 in the Supplementary material) revealed a robust main effect of L1 background and gave the following hierarchy of accuracy: L1 Italian (96.2%) > Finnish (89.8%) > Czech (83.3%) > Slovak (83.1%) > Spanish (75.1%) > German (64%). All L2 groups differed significantly from L1 (p < 0.001). The results are largely consistent with our expectations: learners whose L1 encodes phonological quantity in consonants and/or vowels (Finnish, Czech, Slovak) tended to perform better than learners whose L1 lacks or restricts phonological length (Spanish, German).

As for RT data (Figure 3; Supplementary Table 2), L1 Italian listeners responded fastest, with mean reaction times of approximately 2,385 ms. Among the learner groups, Finnish participants showed the shortest RTs (mean 3,057 ms), followed by Spanish (mean 3,378 ms), Czech (mean 3,540 ms), German (mean 3,636 ms), and Slovak learners (mean 3,641 ms). Across most groups, singleton targets required slightly longer processing times than geminate targets. An additional comparison including both correct and incorrect responses showed that correct responses were produced significantly faster than incorrect responses (3,084 vs. 4,012 ms, p < 0.001), indicating greater processing difficulty and perceptual uncertainty for misidentified items. This pattern was observed across all L1 groups and was particularly pronounced among Slovak, Finnish, and Czech learners.

Accuracy according to L1 and stress position

Unlike in the discrimination task, the ability to identify consonant length appeared to be independent of the prosodic position (Figure 4). The interaction model (see Supplementary Table 3) showed no significant differences between the post-stress, pre-stress and unstressed conditions in the five learner groups. Thus, the expected advantage of prominent positions was not confirmed in the identification task. The post-stress position showed a tendency toward higher accuracy only in German learners. Interestingly, a statistically significant difference was detected in L1 Italian listeners (post-stress: 98.2%, pre-stress: 98.8%, unstressed: 94.6%).

Overall, the absence of a consistent stress effect suggests that phonological categorization in the identification task is relatively robust across different prosodic contexts. The findings did not confirm the hypothesis that prosodic prominence facilitates identification as was the case in the discrimination task (Pešková and Chládková, 2026), which indicates that stress does not play an important role once listeners are guided by categorical decision-making and orthographic cues. However, it is worth mentioning that the post-stress condition was processed the fastest.

Accuracy according to L1 and consonant type

Across learner groups, sonorant consonants were identified more accurately than obstruents (Figure 5). In the model estimates, predicted accuracy was 98.1% for sonorants, compared with 96.9% for voiced obstruents and 96.2% for voiceless obstruents. The advantage of sonorants was statistically significant (p < 0.01) compared to both obstruent categories. The L1-specific interaction model showed the same tendency: sonorants reached 97.3% of accuracy in Finnish (p < 0.001), 92.8% in Slovak (p < 0.05), 92.3% in Czech (p < 0.05), 73.3% in German (n.s.), and 90.2% in Spanish (p < 0.05), exceeding the corresponding values for obstruents in all learner groups. L1 Italian listeners showed high accuracy across all sound categories (sonorants: 98.2%, voiced obstruents: 94.9%, voiceless obstruents: 94.6%).

In addition, sonorant targets were associated with significantly shorter response latencies than voiceless obstruents across groups (p < 0.01), suggesting that sonorants not only yielded higher accuracy but were also processed with greater ease.

Examining the individual consonant types (Figure 6), the results suggest that manner of articulation plays an important role in the perception of consonant length and further underscores the interaction between universal perceptual tendencies and language-specific transfer. For example, the Finnish sound inventory may facilitate the perception of Italian geminates in sonorants and stops but creates difficulties with affricates [tʃ] and [dɜ], which are not present in the Finnish phonological system, and with a voiceless fricative [f], which only appears in loanwords. Spanish learners have a clear advantage with rhotics, likely due to the tap–trill contrast in Spanish. Contrary to our expectations, Slovak learners did not show a clear advantage over Czech learners in the perception of liquids. Although Slovak listeners performed somewhat better with rhotics, this tendency was not observed for laterals. Moreover, Czech and Slovak learners showed unexpectedly low accuracy for the voiceless fricative [f]. This finding is difficult to explain solely on the basis of segment frequency, as [f] is present in both languages and yielded even lower accuracy than the less frequent affricate [dɜ]. Similarly, German learners showed the lowest accuracy for the voiceless affricate [tʃ], although this sound is present in German.

In contrast to all L2 learners, L1 Italian listeners showed very balanced scores for almost all consonant types.

Accuracy according to L1 and proficiency

Overall, identification accuracy tended to increase with higher proficiency levels (Figure 7), although the magnitude of this effect varied across L1 groups and was not significant. The interaction model (Supplementary Table 4) showed improvements in all groups: Finnish (85.4% at A vs. 93.1% at C), Slovak (65.1% at A vs. 94.3% at C), Czech (85.0% at A vs. 89.2% at C), German (46.2% at A vs. 64.2% at C), Spanish (65.9% at A vs. 84.3% at C). Only German learners displayed more accuracy in B (71.0%) than C level (64.2%).

Thus, proficiency appears to play a role, but its effect is less systematic than that of L1 background. For example, especially in the Finnish group, even lower-proficiency learners showed relatively high performance in comparison to other groups, suggesting that L1 structure provides a strong advantage already at early stages of acquisition. These findings reflect earlier observations that perceptual development in L2 cannot be reduced to proficiency level alone (e.g., Feng and Busà, 2022).

Accuracy according to L1 and task

In general, identification accuracy was higher than discrimination accuracy. Model estimates indicate higher accuracy in the identification task than in the discrimination task in all groups with the exception of the Finnish learners (Figure 8; Supplementary Table 5).

This result suggests that identification tasks, which involve explicit categorical decisions and orthographic support, facilitate performance compared to discrimination tasks, which rely more strongly on fine-grained acoustic sensitivity.

The present study investigated the perception of Italian consonant quantity by adult L2 learners from five different L1 backgrounds using an identification task. The results revealed that the perception of Italian consonant length is shaped primarily by cross-linguistic influence, although the mechanisms underlying this influence appear to operate at multiple levels.

The strongest finding concerns the role of L1 phonological structure. The observed hierarchy largely reflects the presence or absence of stable phonological quantity in the learners' L1. Finnish learners, whose language contrasts both vowel and consonant quantity phonemically, performed more accurately than all other learner groups. Czech and Slovak learners, despite lacking phonemic consonant length, also showed relatively high accuracy, whereas German and Spanish learners performed considerably less accurately. These main findings are broadly consistent with the results of the preceding discrimination experiment (Pešková and Chládková, 2026) and support the first hypothesis (H1), although some unexpected patterns emerged. Slovak learners did not show any clear advantage over Czech learners despite the presence of syllabic liquids in Slovak. This may be related to the small number of Slovak participants or to the fact that Italian geminates occurred intervocalically rather than in the syllabic nucleus. German learners were expected to outperform Spanish learners, but Spanish listeners showed slightly higher identification accuracy. One possible reason is the phonetic/orthographic distinction between the Spanish tap (< r>) and trill (< rr>), which may increase learners' sensitivity to consonant duration. However, this explanation alone is unlikely, as Spanish learners also outperformed German learners in several other consonant categories, including nasals, laterals, voiced fricatives, voiceless and voiced affricates, and voiceless stops (see Figure 6). Another possible explanation is that German learners' experience with vowel quantity, reinforced by the orthographic convention whereby double consonant letters primarily signal a preceding short vowel rather than consonant length, may interfere with the acquisition of Italian consonant quantity (as discussed below). Finally, it is also possible that cross-linguistic differences in the phonetic realization of individual consonant categories contribute to these patterns. This possibility will be examined in our ongoing analyses of the production data. Given the relatively small difference between the German and Spanish groups, these explanations remain tentative and require further investigation.

Contrary to expectations, stress position had only a limited influence on identification accuracy (H2). Thus, H2 was not supported. In the previous discrimination experiment, consonant length contrasts were perceived most accurately in post-stress and pre-stress contexts and least accurately in unstressed positions (Pešková and Chládková, 2026). In the present identification task, however, these differences were considerably weaker and generally failed to reach statistical significance. Stress position had a slightly stronger effect only among German learners, who identified consonant length more accurately in post-stress contexts.

One possible explanation concerns the role of orthography. In German, double consonant spelling serves as an important cue to the quantity of the preceding stressed vowel (e.g., Ratte [′ʁatə] ‘rat' vs. Rate [′ʁa:tə] ‘advice'). German learners may therefore have exploited a familiar grapheme–phonology correspondence when distinguishing between Italian singleton and geminate forms (e.g., sàpolo vs. sàppolo), interpreting the former as containing a relatively longer stressed vowel and the latter as involving a shorter vowel followed by a longer consonant. This interpretation is consistent with the phonetic realization of Italian, whereby stressed vowels in non-final open syllables are typically shorter when preceding geminates. Both orthographic and vowel length cues may thus operate in a complementary fashion, jointly reinforcing the post-stress advantage observed in this group.

Interestingly, the post-stress position did not appear to provide any advantage for Finnish, Czech, or Slovak learners, even though these languages exhibit initial prominence and the post-stress condition examined here coincides with stress on the first syllable (e.g., sàppolo). It is plausible that stress is acoustically less salient in these languages and therefore may not provide a sufficiently salient cue to support the perception of quantity contrasts. Instead, learners seem to rely primarily on consonantal duration itself, resulting in relatively stable performance across stress conditions.

The results further revealed a clear effect of consonant type and confirmed the third hypothesis (H3). Sonorants were identified more accurately than obstruents across all learner groups, supporting previous observations that the perceptibility of quantity contrasts depends on the acoustic characteristics of individual consonant classes (e.g., Dmitrieva, 2017). Sonorants contain relatively stable and continuous acoustic energy, which seems to facilitate the perception of durational differences, whereas obstruents often provide less robust temporal cues. The present findings therefore suggest that both universal perceptual factors and language-specific phonological experience contribute to the perception of consonant quantity.

As for proficiency (H4), it showed only modest effects. Although more advanced learners tended to perform somewhat better, the influence of proficiency was considerably smaller than the effect of L1 background. However, this finding should be interpreted with caution, as the distribution of participants across CEFR levels was uneven, with relatively few learners at the A and C levels. This imbalance may have reduced the statistical power to detect potential proficiency effects.

From a theoretical perspective, the results provide an interesting opportunity to evaluate different accounts of transfer in L2 speech perception. The Speech Learning Model (Flege, 1995; Flege and Bohn, 2021) predicts that L2 categories are perceived relative to existing position-sensitive L1 categories through equivalence classification. Within this framework, the relatively high accuracy observed among Finnish learners is consistent with the assumption that Italian singleton and geminate consonants may be assimilated to established Finnish phonemic consonant quantity categories. However, SLM alone may not fully explain the observed hierarchy. Neither Czech nor Slovak contrasts consonant length phonemically, yet both groups performed considerably better than German and Spanish learners. Instead, this pattern suggests that experience with phonological quantity may facilitate the perception of Italian consonant length even when quantity is encoded in a different segment class. This interpretation seems to be more consistent with the Feature Hypothesis (McAllister et al., 2002), according to which transfer may operate at the level of abstract phonological features rather than individual segments. Under this view, experience with vowel quantity may facilitate the perception of consonant quantity because both contrasts rely on duration as a linguistically relevant cue.

At the same time, the Feature Hypothesis does not fully account for all of the observed cross-linguistic differences. The findings also suggest that the phonemic inventory of the learners' L1 plays an important role in the identification of consonant length. This pattern is consistent with Chládková et al. (2013), who showed that although duration is phonemically relevant in Dutch, its role is restricted to specific native vowel categories. From the perspective of the Second Language Linguistic Perception model (L2LP; Escudero, 2005), the observed cross-linguistic differences may further reflect language-specific differences in cue weighting and in the phonetic implementation of quantity contrasts.

This role of segment-specific experience becomes particularly evident in the present data. Finnish learners, despite showing the highest overall accuracy, performed relatively poorly with the affricates [tʃ], [dɜ] and the fricative [f], which are absent from the Finnish phonological inventory or, in the case of [f], available only in loanwords. This finding suggests that the simultaneous acquisition of unfamiliar segmental categories and consonant quantity distinctions may pose a particular challenge. Thus, experience with phonological quantity alone may not guarantee successful identification when quantity contrasts are realized in non-native segments.

Conversely, Spanish learners, who lack phonemic quantity contrasts and performed less accurately overall, showed comparatively high accuracy for rhotics. This advantage may reflect experience with the Spanish tap–trill contrast, which also involves durational, in addition to articulatory, differences.

These findings indicate that experience with phonological duration facilitates the perception of quantity contrasts at a general level, whereas the successful perception of individual consonant types remains influenced by the availability and phonetic implementation of corresponding segmental categories in the L1. Rather than supporting either a purely segment-based or a purely feature-based account, the present results suggest that both levels contribute to L2 speech perception.

The findings further support the view that consonant quantity cannot be reduced to duration alone. Although increased consonant duration represents the primary acoustic correlate of gemination, previous research has shown that length contrasts are realized through a constellation of temporal, spectral, and articulatory cues that vary across languages (Maddieson, 1984; Ladefoged and Maddieson, 1996; Ridouane, 2010; Šimko et al., 2014; Burroni et al., 2025). Consequently, successful perception of Italian geminates appears to depend both on experience with duration as a phonological feature and on familiarity with the segmental categories through which that feature is realized. Future studies should investigate how quantity is phonetically implemented across learners' native languages and how these implementation patterns influence the perception and production of Italian geminates in L2 speech.

Finally, the comparison with the discrimination experiment (Pešková and Chládková, 2026) revealed that identification accuracy was generally higher than discrimination accuracy across all learner groups, confirming H5. This finding is consistent with previous research suggesting that identification and discrimination tasks rely on partly different perceptual processes (Nimz, 2016; Lee et al., 2026). Previous studies have further shown that identification training often leads to stronger improvements in the acquisition of difficult L2 phonological contrasts than discrimination training (e.g., Lambacher et al., 2005; Iverson and Evans, 2009; Gi Law et al., 2019; Cebrian et al., 2024). Identification tasks with written responses appear to facilitate the formation of stable phonological categories, whereas discrimination tasks primarily target sensitivity to acoustic differences. In contrast, the discrimination task required listeners to compare successive stimuli without additional task-specific support, thereby increasing its cognitive demands. The higher accuracy observed in the present identification experiment therefore suggests that listeners were able to rely not only on acoustic information but also on emerging phonological representations of consonant quantity.

The role of both L1 and L2 orthography also appears to be relevant in this context. Unlike the discrimination task, the identification task provided written options containing either singleton or geminate consonants. Since Italian orthography systematically marks consonant length through double letters, participants may have relied not only on auditory but also on visual information when categorizing the stimuli. This may have provided an additional advantage for Finnish learners, whose orthography likewise uses double letters to represent long consonants (e.g., muutto ‘move, migration'), and, to a more limited extent, for Spanish learners through the orthographic distinction between the tap (< r>) and trill (< rr>) in intervocalic positions. In contrast, Czech and Slovak use double consonant letters only in restricted morphological contexts, while in German double consonant letters do not represent consonant length but primarily signal vowel quantity or occur across morpheme boundaries. Thus, the availability and function of orthographic cues differ considerably across the learners' L1s and may have influenced identification performance. Ongoing work will compare production data from reading tasks (with orthographic cues) and repetition tasks (without orthographic cues). This comparison should provide further insight into the respective contributions of orthographic and phonological information to the acquisition of Italian consonant length.

Overall, L2 learners appear to benefit more generally from experience with duration as a linguistically relevant cue, while at the same time remaining sensitive to the segmental structure of their native phonological systems. At the individual level, considerable variability was observed across all groups, suggesting that extra-linguistic factors may also shape perceptual performance. These may include prior language experience, study time, exposure to Italian outside the classroom, time spent in Italy, social contact with L1 Italian speakers, motivation, and general language-learning aptitudes (e.g., Piske et al., 2001; Colantoni et al., 2015; Derwing and Munro, 2015; Wen et al., 2023). Future research should therefore investigate how linguistic and extra-linguistic factors interact in shaping sensitivity to L2 phonological contrasts.

Ultimately, the comparison across the five listener groups highlights the importance of a cross-linguistic perspective for understanding how learners interpret durational cues in a foreign language (cf. Lee et al., 2026). Rather than relying on a single source of information, L2 listeners draw on multiple forms of prior linguistic experience to map continuous acoustic variation onto discrete phonological categories when perceiving quantity contrasts.