Impact of Speaker Accent and Listener Background on FL Learners’ Perceptions of Regional Italian Varieties

Abstract

In today’s globalized world, foreign language (FL) communication is characterized by the presence of regional variations that can impact L2 learners’ speech perception in their target language. While it is essential for FL programs to prepare their students for real-world language variation, research on learner perception of spoken regional varieties remains scarce, especially for less commonly taught languages, such as Italian. To address this, this study used a quantitative approach to explore to what extent listeners’ background factors (i.e., accent familiarity, contact with Italian speakers, heritage learner status, L1) and speaker accent impact FL learners’ perceptions of (1) intelligibility, (2) comprehensibility, and (3) acceptability of regionally accented Italian speech. A total of forty-seven FL learners of Italian listened to the speech of six Italian native speakers with different regional accents (two each from Northern, Central, and Southern Italy), rated them for comprehensibility and acceptability, and transcribed utterances for intelligibility. Mixed-effects models revealed statistically significant effects of speaker accent and listener L1 background on all three perceptual constructs. Additionally, greater contact corresponded with higher comprehensibility, while heritage learners rated speech as less acceptable for educational contexts. Listeners’ overall positive perception of regionally accented speech encourages the introduction of authentic spoken varieties in the FL classroom.

1. Introduction

Modern foreign language (FL) programs are responsible for preparing learners for the real-world communication that happens beyond the classroom, facilitating students’ entry into authentic relationships with others in a globalized world. Real-life FL communication is generally characterized by the presence of regional variation within the target language, which can impact speech perception and affect the success of second language (L2) communication (Lam & O’Brien, 2014; Schoonmaker-Gates, 2017, 2018; Wuensch & Bolter, 2020). Helping learners build realistic expectations about what L2 communication will entail is therefore an important aim among FL programs.

Both intelligibility, or the extent to which listeners actually understand a speaker’s message, and comprehensibility, or the ease or difficulty of understanding (Munro & Derwing, 1995), contribute to communicative success. Ideally, comprehensible and intelligible speech would also be considered acceptable by listeners. Acceptability can depend on attitudes or evaluations and be influenced by sociolinguistic factors and situational context (Crowther et al., 2023; Dalman & Kang, 2023; Wheeler & Kang, 2022). These perceptual measures may be impacted by speaker accent and individual listeners’ background factors (Munro, 2008). Previous studies (e.g., Kang et al., 2019a; Winke et al., 2013) have explored the impact of L1 listeners’ perceptions of L2 speakers and have established the effect of listener background variables, such as accent familiarity and attitudes, on L2 speech perception and ratings.

However, work that examines FL learners’ perceptions of the regional varieties in their target language (e.g., Saito et al., 2019; Wheeler & Kang, 2022) remains more limited, particularly among less commonly taught languages (LCTL) such as Italian. Indeed, learners of university-level Italian as a foreign language (IFL) frequently have heritage connections to the Italian language, and many IFL learners aim to use their FL skills for study abroad and professional purposes (D’Orazzi & Hajek, 2021; Ferronato, in preparation) In these real-world communication contexts, learners will encounter spoken Italian that is highly regionalized in both segmental and suprasegmental features (Gili Fivela et al., 2015), yet it is rare for IFL textbooks and other pedagogical materials to include regional variation (Cutrì, 2016), representing a potential mismatch between current pedagogical practices and real-world communicative needs. Therefore, the current study explored how IFL learners experience regionally accented speech in terms of intelligibility, comprehensibility and acceptability, with a focus on the role that listener factors and speaker accent variation play in these perceptions. Such findings in turn can inform pedagogical approaches to incorporating regionally accented speech in the FL classroom (Major et al., 2005; Stollhans, 2020; Wuensch & Bolter, 2020).

1.1. Measuring FL Learners’ Perceptions of Regionally Accented Speech

Previous work on FL learners’ perceptions of spoken regional variation in languages such as Spanish (Schoonmaker-Gates, 2017, 2018; Schmidt & Geeslin, 2022) and German (Lam & O’Brien, 2014) has demonstrated that learners can often distinguish among regional dialects in the target language and may hold varied attitudes toward different dialects. Research in this vein has been relatively limited, however, and findings have not been consistent across target languages. For instance, Masullo (2023) recently found that L2 learners in Italy struggled to detect regional differences in spoken Italian and did not appear to hold linguistic stereotypes toward regional varieties. Moreover, much of the existing literature has adopted a sociolinguistic lens for analyzing FL learners’ responses to regional variation (e.g., perceptual dialectology; Lam & O’Brien, 2014), emphasizing learners’ ability to identify different speech varieties. This focus is useful for understanding how listeners experience such speech, but in this paper, we argue that there is an opportunity to build on this by examining IFL learners in terms of (1) how well they understand regionally accented Italian utterances (i.e., intelligibility), (2) how easy/difficult they find the speech to understand (i.e., comprehensibility), and (3) how suitable they find this speech for language learning purposes (i.e., acceptability).

1.2. Listener Factors

In the examination of why individual FL learners vary in their perceptions of regionally accented speech, listeners’ linguistic profiles can offer a wealth of insight. In this paper, we explore four listener background factors that are connected to listeners’ language-related experiences: accent familiarity, contact with target language users, heritage language learner status, and L1 background.

1.2.1. Accent Familiarity

Defined as “a speech perception benefit developed through exposure and linguistic experience” (Browne & Fulcher, 2017, p. 39), accent familiarity has been shown to facilitate L2 listeners’ understanding of regionally accented speech (e.g., Schoonmaker-Gates, 2018), especially among learners of lower proficiency levels (Major et al., 2005). At higher proficiency levels, however, L2 listeners may be less susceptible to negative impacts of unfamiliar accents on their comprehension (Kang et al., 2019b). In terms of ease of understanding, previous findings are mixed, with some demonstrating that L2 listeners’ familiarity with a particular accent leads to greater comprehensibility (e.g., Winke et al., 2013) and others showing no statistically significant impact (Saito et al., 2019). One challenge in this body of work that Miao and Kang (2023) highlight is the measurement of accent familiarity; this construct has been measured through scalar ratings of accent varieties in general (e.g., How familiar are you with Australian-accented English?) or in response to a specific audio stimulus (e.g., How familiar was this speaker’s accent?), as well as through identification of a speaker’s accent (e.g., Where is this speaker from?) or shared linguistic background (see Section 1.2.4). These differences in the operationalization of accent familiarity may also contribute to the discrepancies in findings, although more work is needed.

1.2.2. Contact with Target Language Users

Related to accent familiarity is contact with target language users, which can be an indicator of a learner’s exposure to the target language. The cognitive processing benefits of exposure to unfamiliar regional and L2 accents have been demonstrated in experimental settings (Witteman et al., 2014, 2015), with exposure of even just two to four sentences helping listeners overcome initial processing slowdowns (Clarke & Garrett, 2004), although Floccia et al. (2009) found that exposure was not enough for listeners to adapt to processing the unfamiliarly accented speech at the same speed as their baseline. In the context of language learning, high variability phonetic training (HVPT) has demonstrated perceptual benefits for learners who engage in structured practice to differentiate among individual sounds in their target language (see Thomson, 2018, for a review). How well these findings might translate to self-reported measures of processing (i.e., comprehensibility), however, remains underexplored.

Often measured in terms of hours of exposure within a fixed period of time (e.g., per week, Kang et al., 2021) or frequency of exposure in media and face-to-face communication (Ockey & French, 2016), contact can reflect the amount of authentic target language input that learners receive. With regard to its role in FL accent perception, Schoonmaker-Gates (2017) found that exposure to Spanish regional dialects significantly predicted learners’ dialect perception. Among university FL learners, one common source of contact with target language users is through study abroad. In the case of Spanish, Schmidt and Geeslin (2022) found that learners with study abroad experience displayed greater differentiation in their positive and negative attitudes toward regional Spanish accents. However, work on the role of contact in IFL learners’ perceptions of regional accents remains very limited.

1.2.3. Heritage Language Learner Status

Heritage language learners (HLLs) are individuals raised in an environment where a language other than the dominant language of their country of residence was or is spoken, often within their household or community (Polinsky, 2011). While different factors affect the linguistic system of HLLs such as their “bilingual onset” (Ortega, 2020, p. 20), culture, social class, and exposure to different registers and varieties (Montrul, 2010), as adults, many HLLs seek to improve their linguistic skills in their family language as they may have achieved only a partial command of it due to the inaccessibility of education in the heritage language (Montrul, 2010)1. Several definitions and profiles are attributed to HLLs. Reynolds et al. (2009) distinguish between a narrow definition, referring to individuals who regularly use the target language at home, and a broader definition, including those who do not. This study adopted the broader definition, considering HLLs to be students exposed to the Italian language in their households (i.e., their parents or grandparents are Italian) but primarily using English in educational and external contexts in the U.S.

HLL status has been found to be a complex yet influential background factor for speech perception constructs such as acceptability, accent familiarity, and intelligibility. Wheeler and Kang (2022) reported Spanish HLL status to be a predictor of acceptability judgements about L1 Spanish speakers. Using Beinhoff’s (2014) definition of acceptability, which takes into account the listeners’ social background, and his scale of 1 (highly acceptable) to 10 (highly unacceptable), the authors found that participants identifying as HLLs rated L1 Spanish speakers as less acceptable based on what the speakers were discussing in the speech sample. Winke and Gass (2013) examined the impact of accent familiarity of heritage language (HL) and non-HL raters of L2 English speech and found that HL ratings were influenced by their interlanguage and sociocultural experiences. Similarly, Huang et al. (2016) found that Spanish and Chinese HL raters who were familiar with specific accents rated Spanish-accented English speech as “more native-like speech” compared to the ratings from non-HL raters, even though both HL and non-HL raters had the same familiarity with Spanish accents. The authors thus suggested that HL raters might carry more empathy toward L2 speakers because of their experience with accented speech or their family connections. Staggs et al. (2022) explored the effects of participants’ self-reported Spanish HL status on their intelligibility judgements of other Spanish HL speakers and learners, finding that the HL status had little impact on intelligibility.

While several HLL studies have focused on Spanish, other LCTLs are still largely unexplored especially with regard to speech perception constructs, as is the case for Italian. A few studies on Italian as an HL within the U.S. focused on the auxiliary selection in past compound tenses (Bonfatti-Sabbioni, 2023), identity and language use (De Fina, 2014), and the use of grammatical features such as gender agreement, preferred past tense forms, and clitic placement (Sabbioni, 2018). However, the Italian HLL status has not been investigated within the field of L2 speech perception yet.

1.2.4. L1 Background and Transfer

Research has shown that listeners’ L1 background has an impact on the intelligibility and comprehensibility ratings of L2 speech. Positive effects of sharing the same L1 background on L2 intelligibility, for instance, were shown by Bent and Bradlow (2003), who examined how well English L1, Chinese L1, Korean L1, and a mixed-L1 group of listeners could recognize English speech produced by L1 speakers of English, Chinese, and Korean. Their findings revealed that among various languages, English L1 listeners rated English L1 speakers as the most intelligible and that non-English L1 listeners rated the speech of proficient non-English L1 speakers as intelligible as the speech of an English L1 speaker, thus revealing the positive effects of a shared L1 between listeners and speakers on L2 intelligibility. This hypothesis, the interlanguage speech intelligibility benefit (ISIB; Bent & Bradlow, 2003), was discussed in subsequent studies that acknowledged its presence, albeit sometimes minimal. Algethami et al. (2011), for instance, found that although L2 Saudi listeners performed slightly better than native listeners did in transcribing Saudi-accented English, the difference between the groups was not statistically significant. Similarly, Hayes-Harb et al. (2008) found that Mandarin listeners of English outperformed L1 English listeners in comprehending Mandarin-accented English speech, but they did not find speakers with the same background more intelligible.

Building on this, several studies also suggested that a shared L1 background often increases comprehensibility, without necessarily improving intelligibility. For instance, Hansen Edwards et al. (2018) found that a shared L1 background of Asian Englishes significantly increases listeners’ comprehensibility and accentedness (but not intelligibility). Similarly, Saito et al. (2019) investigated how L2 raters judge the comprehensibility of Japanese speakers of English and found that raters with an L1 closer to the target language (i.e., Japanese-accented English) assigned higher comprehensibility scores. The authors discussed a connection with the ISIB phenomenon, arguing that the close distance between L1 and L2 may enhance the perceptual adaptation of listeners.

1.3. Speaker Factors

As listener-related factors have been shown to influence the perception of L2 speech, speaker-related factors are equally impactful in shaping listeners’ judgments. Studies have demonstrated how speakers’ suprasegmental features, such as speech rate, intonation and pausing (Kang et al., 2010), and accents (Munro & Derwing, 1995) contribute to listeners’ comprehensibility and/or intelligibility judgments. However, most studies have tended to explore the different suprasegmental features of native and non-native speech and how these affect listeners’ perceptions. Fewer studies have instead examined the effects of speakers’ regional variation in the target language on listeners’ perceptions. Schoonmaker-Gates (2017) found that exposure to and explicit instruction about Spanish dialectal variation for L2 Spanish learners predicted their perception of Spanish dialects. Lam and O’Brien (2014) observed that exposure to German non-standard variety helped L2 listeners to discriminate between German speakers’ dialects but did not predict their perceived intelligibility. Masullo (2023) reported on the general difficulties experienced by Italian L2 listeners in perceiving Italian regional varieties that differ from the standard.

The case of spoken Italian language presents an intriguing area of study, as speakers of Italian naturally speak a regional variety (Crocco, 2017), with distinct segmental features and intonation patterns that reflect the linguistic diversity inherent to the northern, central, and southern regions of the country (Gili Fivela et al., 2015). As for segmental features, some instances of regional linguistic diversity are the nasal consonant /n/ that can be velarized in Northern varieties when occurring at the end of a syllable or word (Maiden, 1995 in Crocco, 2017, p. 98), the intervocalic /r/ that is shortened in some Central varieties spoken in Lazio and Tuscany (Maiden, 1995 in Crocco, 2017, p. 98), and the distribution of the alveolar fricatives /s/ and /z/. When placed between vowels, /s/ and /z/ can distinguish minimal pairs such as fuso [‘fuso] (spindle) and fuso [‘fuzo] (liquified), but speakers of regional varieties neutralize this phonological contrast and have only one form with different meanings (Crocco, 2017, p. 97)2. With regard to Italian speakers’ intonation, for example, Savino (2012) and Gili Fivela et al. (2015) reported on the cross-regional variability in the ending tone of polar questions (i.e., yes–no questions) that is typically the only distinguishing feature from statements. As statements in Italian have a falling tone, intonation analyses of Northern, Central, and Southern polar questions revealed a wide variety of intonational patterns for the same questions ranging from a terminal rise to a terminal falling tone, thus confirming the extreme diversity in spoken Italian (Savino, 2012). With this regard, De Iacovo and Mairano (2024) investigated how prosodic patterns of polar questions and statements pronounced by L1 Italian speakers of different regional varieties are perceived by L1 English listeners. They found that different pitch contours (i.e., falling/raising) peculiar to specific regional varieties can challenge the recognition of sentence modality (i.e., polar question vs. statement) by L2 Italian speakers. Of particular difficulty in this study were the rise–fall contours on polar questions produced by speakers from Turin (northwestern Italy). While a more detailed discussion of regional variation in prosodic features is beyond the scope of this paper, we refer readers interested in more information to Gili Fivela et al. (2015) and De Iacovo and Mairano (2024).

1.4. Current Study

Despite the effect that regional segmental and suprasegmental features may have on FL listeners’ perceptions, to our knowledge, research investigating these perceptions remains limited. Therefore, this exploratory study aimed to address the following research question:

RQ: To what extent do listeners’ background factors (i.e., accent familiarity, contact with Italian speakers, heritage learner status, L1) and speaker accent impact FL learners’ perceptions of (1) intelligibility, (2) comprehensibility, and (3) acceptability of regionally accented Italian speech?

Because this study is exploratory, we did not hypothesize a priori about the nature of the relationships among the independent and dependent variables. However, our findings point to potential hypotheses for future confirmatory studies in this area. We also acknowledge that this study is part of a larger project that includes the investigation of the pragmatic impact of Italian intonation on the perception of yes–no questions and statements by FL learners (Yaw & Ferronato, 2024).

2. Materials and Methods

2.1. Participants

This study included two groups of participants: 47 listeners who completed the perceptual tasks and 6 speakers who recorded the audio stimuli.

Listeners were adult IFL learners (M_age = 21.41 years, SD = 7.97) who were recruited from intermediate- and advanced-level classes at three universities across the United States. The majority of listeners (70.2%, n = 33) identified as female, while 25.5% (n = 12) identified as male and 4.3% (n = 2) identified as non-binary. Listeners reported either L1 English (83%, n = 39) or Spanish (17%, n = 8) backgrounds. In spite of attempts to recruit IFL learners at both intermediate and advanced proficiency levels, nearly all participants (95.7%, n = 45) identified as intermediate level, while only two listeners (4.3%) identified as advanced. Nearly half of these learners (44.7%, n = 21) reported being heritage students of Italian (i.e., having a family connection to the Italian language, such as a grandparent who emigrated from Italy to the United States3). Additionally, nearly two-thirds (63.8%, n = 30) reported some level of contact with Italian speakers in their community and/or at home. In terms of familiarity with different regional accents, Figure 1 illustrates that listeners overall reported low levels of familiarity with the three regions included in this study (northern, central, and southern), with the greatest degree of reported familiarity appearing for the “standard” (i.e., textbook) accent.

Speaker participants in this study were six L1 Italian speakers. One female and one male were recruited from each of the regions (north, central, south) to create the audio recordings for this study. Their ages ranged from 26 to 40 (M_age = 31 years old, SD = 4.94) years, and all were educated through the university level in Italy. At the time of the recording, three of the speakers had resided for at least one year in the United States. Speakers received a USD 20 Amazon gift card as a compensation for participating.

2.2. Materials

All the data in the present study were collected online using questionnaires designed and distributed through the platform Qualtrics.com. First, a background questionnaire was administered to gather listeners’ demographic information (see Supplementary File). Data included participants’ age, gender, degree level, L1, heritage language learner status, prior knowledge of the Italian language, familiarity with Northern, Central, Southern, and Standard Italian accents (expressed through 5-point scalar ratings ranging from not familiar to extremely familiar for each accent) and prior exposure to Italian regional accents (this was self-evaluated through scalar ratings expressing the percentage of time that participants spent in contact with Italian in their home and community; the rating of the values for home contact and community contact were then averaged to represent the overall contact). To access the background questionnaire and proceed with its completion, participants were required to provide informed consent.

Next, participants completed an intelligibility task. They listened to 18 semi-scripted sentences (three sentences per speaker and six sentences per regional accent), with nine sentences being statements and nine being polar questions (see

Table 1. Examples of prompts and expected responses for the intelligibility task recordings.

Type of Utterance	Prompt	Expected Response
Statement	Ti chiedono quale materia è facile. Guarda l’immagine e rispondi.	Una materia facile è l’inglese.
	[They ask you what an easy subject is. Look at the picture and reply]	[An easy subject is English.]
Polar question	Sei al telefono con tua mamma e state organizzando una cena con i parenti. Non hai ancora avuto conferma che i tuoi nipoti verranno. Chiedi a tua mamma se hanno detto di si.	Hanno detto di si?
	[You are on the phone with your mum and planning a dinner with some relatives. Your nephews haven’t confirmed their presence yet. Ask your mum if they said yes.]	[Did they say yes?]

below for examples of the prompts presented to the speakers to record the sentences). Across all three accent conditions, all of the responses were structurally simple and consisted of a single sentence without coordination or subordination, thus appropriate for beginner-level Italian learners. The vocabulary terms selected in the prompts and expected in the responses were also covered at the beginner level of Italian classes. The order of presentation of the audio files to listeners was random, and participants could only hear each file one time. Upon listening to the audio file, they were asked to (a) transcribe what they heard using their keyboard and (b) indicate if it was a statement or a question.

Finally, participants listened to 12 short excerpts, each 30 s long, of speakers talking about their hobbies, their ideal vacation, a life adventure, and advice to someone visiting Italy for the first time (see

Table 2. Examples of prompts for the comprehensibility task recordings.

Prompts
Cosa ti piace fare nel tempo libero? [What do you like to do in your free time?]
Descrivimi la tua vacanza ideale. Dove andresti e perchè? [Describe your dream vacation. Where would you go and why?]
Qual è stata la più grande avventura della tua vita finora? Raccontami cosa è successo e perché è stata un’avventura. [What has been your greatest life adventure so far? Tell me what happened, and why this was such an adventure.]
Quali consigli daresti ad una persona che visita l’Italia per la prima volta? [What advice would you have for someone who is visiting Italy for the first time?]

for examples of the open-ended prompts presented to the speakers). These topics were selected as the vocabulary and verb tenses used to talk about them should be familiar to beginner-level students, and they also allow for the elicitation of spontaneous utterances. This selection allowed us to balance the need for controlled conditions with the authenticity of speech, which resemble real-world communication. Participants listened to three regional accents for each of the four topics. As in the intelligibility task, the audio files were presented in a randomized order, and participants could only hear each file once. Upon listening, participants were asked to judge for comprehensibility by indicating on a 5-point semantic differential scale if the speaker was easy/difficult to understand, unclear/clear, and if they required little/a lot of effort to understand (items adapted from Kang, 2010). Additionally, they rated the acceptability of the speaker for three different educational contexts by using a 5-point Likert-type scale to indicate their agreement (1) or disagreement (5) with the three statements “I consider this Italian accent acceptable for foreign language teachers”, “I consider this Italian accent acceptable for Italian textbooks and teaching materials”, and “I consider this Italian accent acceptable for e-tandem partners”.

Before any of the listening tasks, an audio check was conducted to ensure participants could perceive the stimuli clearly and at a comfortable volume. Between the intelligibility and the comprehensibility/acceptability task, participants were prompted to take a short break if needed.

2.3. Procedure

Six Italian L1 speakers were contacted as personal acquaintances of one of the researchers, with two speakers each from Northern Italy (Veneto region), Central Italy (Lazio region), and Southern Italy (Basilicata and Campania regions). To confirm their regional accents, two additional Italian L1 speakers independently listened to one randomly selected file from each speaker and identified the speakers’ geographical origins, thus validating their regional accents. Each recording session was conducted on the online meeting platform Microsoft Teams. In each meeting, the speaker was instructed to respond with complete sentences when describing a picture or responding to a scenario (prompts for intelligibility tasks) and to talk for at least one minute when responding to the open-ended questions (prompts for comprehensibility tasks). The audio recordings were subsequently downloaded and processed using the open-access software Audacity (version 3.4). All the audio files were converted to .wav format and segmented according to the prompts required for inclusion in the survey.

Once the recordings were prepared and integrated into the research instruments, listener data were gathered online using Qualtrics surveys. This study was part of a larger four-part study that spanned eight weeks. In this paper, we report on relevant findings from the first and fourth parts, which were the listener background questionnaire and speech ratings detailed in Section 2. Upon enrolling in the study, listeners received a link to the first study part, which they had two weeks to complete. Links to subsequent parts were distributed the same way in two-week increments. Each part required roughly 15–30 min to finish, and listeners were able to earn participation or extra credit points for their IFL course upon completion of all four parts.

2.4. Data Preparation and Analysis

In preparation for the analysis, all the data were exported from Qualtrics to Excel. Each listener participant was assigned a numerical code for identification and matching across tasks, and identifiers were removed. Each task was then organized into a separate spreadsheet to facilitate data management and cleaning.

As for intelligibility, we used the token sort ratio (TSR) to assess the accuracy of each listener’s audio file transcription automatically. Among several forms of fuzzy string matching, TSR has been found to be a consistent, efficient, and accurate measurement that allows the matching of transcribed and targeted long strings of words, such as sentences (Bosker, 2021). The TSR score can range from 0 (no orthographic match) to 100 (exact match). To calculate TSR for our intelligibility dataset, first we organized data in Excel in different columns corresponding to the participant (numerical code), their region of origin (northern, central, southern), their gender (male, female), the intonation type of the sentence pronounced (question, statement), the intonation type in listeners’ responses (question, statement), the match between listeners’ responses and speakers intonation (match = 1, no match = 0), the transcription of the target sentence pronounced by the speakers (target), and listeners’ responses transcription (response). Then, data were converted into .csv format and uploaded into the online open-source tool for TSR calculation (available at https://tokensortratio.netlify.app/, accessed on 11 June 2024).

For the preparation of the data for comprehensibility, the values of the comprehensibility scalar ratings were reported in columns, each corresponding to an item (easy/difficult to understand, clear/unclear, and required a lot of effort/no effort). The values of one scale (clear/unclear) were reversed to ensure the same directionality across all scales. Items’ internal consistency was calculated using Cronbach’s alpha and was relatively high (α = 0.89). As a result, we calculated one mean comprehensibility rating for each listener’s response to each audio file.

The same procedures were followed for the preparation of the data for acceptability. Each column in the Excel sheet corresponded to an item (acceptable for FL teachers, textbook, and e-tandem) and all values were reversed to ensure the same directionality across all scales and tasks. Items’ internal consistency was calculated using Cronbach’s alpha (α = 0.94). In spite of the high alpha values, we opted to analyze each acceptability item separately to explore any differences in learner perceptions across the three educational contexts (i.e., FL teachers, textbooks, and e-tandem partners).

Data were analyzed in a series of five linear mixed-effects models using the lme4 package (Bates et al., 2015) in R (Version 4.3.3; R Core Team, 2024), with both speaker and listener factors included as fixed effects (see

Table 3. Summary of the operationalization of fixed effects.

Category	Variable	Operationalization
Speaker	Accent	Regional accent (Northern, Central, Southern)
Listener	Familiarity	Reported degree of familiarity with regional accent of speaker, scale of 1 (not familiar) to 5 (extremely familiar)
	Contact	Mean reported percentage of time in contact with Italian speakers in home and community, scale of 0–100%
	Heritage status	Reported heritage learner of Italian, binary (0 = no, 1 = yes)
	L1	Reported L1, categorized into two groups—Romance (e.g., Spanish) and non-Romance (e.g., English)

). In each model, listeners were included as a random effect. Dependent variables were intelligibility (TSR scores), comprehensibility (mean ratings), and acceptability (three models—one per educational context).

Given the exploratory nature of this study, we wanted to account for the possibility of interaction effects among our independent variables. We therefore fit and compared three versions of the model for each dependent variable. The baseline models (see

Table 4. Baseline model code.

Model	Code
Intelligibility	LMEM_intel <- lmer(Intel_TSR_score ~ Spkr_Accent + Familiarity + Contact + Heritage + L1 + (1|Listener), data = df)
Comprehensibility	LMEM_comp <- lmer(Comp ~ Spkr_Accent + Familiarity + Contact + Heritage + L1 + (1|Listener), data = df)
Acceptability—e-tandem	LMEM_accept_etandem <- lmer(Accept_etandem ~ Spkr_Accent + Familiarity + Contact + Heritage + L1 + (1|Listener), data = df)
Acceptability—FL teacher	LMEM_accept_FLteacher <- lmer(Accept_FLteacher ~ Spkr_Accent + Familiarity + Contact + Heritage + L1 + (1|Listener), data = df)
Acceptability—Textbook	LMEM_accept_textbook <- lmer(Accept_textbook ~ Spkr_Accent + Familiarity + Contact + Heritage + L1 + (1|Listener), data = df)

) included no interaction terms. Using the output of the baseline, we also generated a partial interaction model with interaction terms for only the statistically significant independent variables in the baseline (i.e., speaker accent for all models, L1 for intelligibility, contact and L1 for comprehensibility, and heritage for all three acceptability models). Our third version of the models included interaction terms for speaker accent with all of the listener background variables (e.g., Spkr_Accent * (Familiarity + Contact + Heritage + L1)).

We then calculated Schwarz’s Bayesian criterion (BIC) values to aid in model selection. For all five dependent variables, the BIC values were lowest for the no interaction version of the model, as illustrated in

Table 5. BIC values used for model selection.

Dependent Variable	No Interaction	Partial Interaction	Full Interaction
Intelligibility	7838.0	7851.3	7888.8
Comprehensibility	1600.2	1622.7	1646.1
Acceptability—e-tandem	1587.1	1599.1	1630.7
Acceptability—FL teacher	1634.6	1646.8	1678.9
Acceptability—textbook	1662.4	1674.6	1704.2

. Following Kass and Raftery’s (1995) criteria, the BIC values did not provide compelling evidence to retain either the partial or full interaction models. We also note that in the case of the models with interaction terms, none of the interaction effects emerged as statistically significant. We therefore retained the most parsimonious version of each model (i.e., the baseline with no interaction terms from Table 4).

3. Results

This section presents findings according to the three speech constructs under investigation: intelligibility, comprehensibility, and acceptability.

3.1. Intelligibility

The intelligibility model focused on the effects of speaker and listener factors on listeners’ transcription accuracy as measured through TSR scores, which ranged from 0 (no orthographic match) to 100 (complete orthographic match). The model demonstrated statistically significant effects of both speaker accent (F(2, 802) = 5.408, p = 0.005) and listener L1 (F(1, 43) = 10.595, p = 0.002).

Table 6. Summary of fixed effects for intelligibility (TSR scores).

Fixed Effect	Estimate	SE	df	t	p	Mar. R2	Cond. R2
(Intercept)	65.606	3.940	81	16.652	<0.001 ***	0.086	0.321
Speaker accent—Northern	1.118	1.921	796	0.582	0.561
Speaker accent—Southern	−4.956	1.952	804	−2.539	0.011 *
Familiarity	−0.114	1.493	683	−0.076	0.939
Contact	0.237	0.163	49	1.450	0.153
Heritage	−2.096	4.434	43	−0.473	0.639
L1—Romance	18.577	5.707	43	3.255	0.002 **

* p < 0.05. ** p < 0.01. *** p < 0.001.

presents a summary of the fixed effects, which accounted for 8.6% of the variance in intelligibility scores in this model (as indicated by the marginal R² value). Additionally, the random effect of listener accounted for a further 23.5% of the variance (as indicated by the difference between the conditional and marginal R² values).

Figure 2 illustrates listeners’ TSR scores by speaker accent condition. It shows that listeners were statistically significantly less accurate in their transcription of utterances produced by speakers with a Southern accent, whereas their performance with the Northern- and Central-accented speakers was relatively similar. This indicates that the three regional accents were not equally intelligible to student listeners.

Examining the impact of listener L1, however, we noted a difference in the intelligibility scores between listeners who came from a Romance versus non-Romance L1 background.

Table 7. Estimated marginal means of intelligibility by speaker accent and listener L1.

Speaker Accent	Non-Romance L1			Romance L1
	Emmean	SE	95% CI	Emmean	SE	95% CI
Northern	67.6	2.57	[62.5, 72.8]	86.2	5.35	[75.4, 97]
Central	66.5	2.57	[61.4, 71.6]	85.1	5.35	[74.3, 95.8]
Southern	61.6	2.58	[56.4, 66.7]	80.1	5.34	[69.4, 90.9]

displays the estimated marginal mean TSR scores for these two listener groups across the three speaker accent conditions. While both groups showed the lowest scores for the Southern-accented speakers, the listeners with a Romance L1 background performed approximately 19 points higher than their non-Romance L1 counterparts did across all three accent conditions. In other words, those with a Romance L1 background found the speech statistically significantly more intelligible, suggesting that an L1 background that is linguistically closer to that of the target language may facilitate comprehension of regionally accented speech.

3.2. Comprehensibility

As for comprehensibility, the model revealed statistically significant effects of speaker accent (F(2, 499) = 15.207, p < 0.001), listener contact with Italian speakers (F(1, 50) = 9.179, p = 0.004), and listener L1 background (F(1, 41) = 6.717, p = 0.013).

Table 8. Summary of fixed effects for comprehensibility ratings.

Fixed Effect	Estimate	SE	df	t	p	Mar. R2	Cond. R2
(Intercept)	3.585	0.155	101	23.172	<0.001 ***	0.112	0.236
Speaker accent—Northern	−0.268	0.103	492	−2.617	0.009 **
Speaker accent—Southern	−0.575	0.104	502	−5.514	<0.001 ***
Familiarity	−0.074	0.071	290	−1.033	0.302
Contact	0.017	0.006	50	3.030	0.004 **
Heritage	−0.062	0.152	41	−0.406	0.687
L1—Romance	0.496	0.191	41	2.592	0.013 *

* p < 0.05. ** p < 0.01. *** p < 0.001.

reports a summary of the fixed effects from this model. In this case, the fixed effects accounted for 11.2% of the variance in comprehensibility ratings, while the random effect of listener accounted for an additional 12.4% of the variance.

Figure 3 plots listeners’ comprehensibility ratings by speaker accent condition. This illustrates a statistically significant difference among all three regional accents, with the Central-accented speakers rated as most comprehensible, followed by those with Northern accents, and finally Southern-accented speakers. Notably, however, all three accent conditions had mean comprehensibility ratings that were at or above neutral (3.0 on the scale).

In terms of listener contact, the model indicated that each percentage increase in listener contact with Italian speakers in their homes and communities improved the comprehensibility rating by 0.017. When reporting on contact, listeners were asked to consider how often they hear and/or speak Italian in a broad range of contexts, including with family, housemates, neighbors, and friends, as well as in sports clubs, places of worship, and other community spaces. This finding provides evidence that greater contact leads to greater comprehensibility among this group of Italian language learners.

Finally, related to the impact of listener L1 background,

Table 9. Estimated marginal means of comprehensibility ratings by speaker accent and listener L1.

Speaker Accent	Non-Romance L1			Romance L1
	Emmean	SE	95% CI	Emmean	SE	95% CI
Northern	3.33	0.099	[3.13, 3.53]	3.83	0.185	[3.46, 4.2]
Central	3.6	0.099	[3.4, 3.8]	4.09	0.185	[3.72, 4.47]
Southern	3.02	0.100	[2.83, 3.22]	3.52	0.184	[3.15, 3.89]

reports the estimated marginal mean comprehensibility ratings for both non-Romance and Romance L1 listeners across all three accent conditions. As with the intelligibility scores, listeners with a Romance L1 background found speakers with all three regional accents to be statistically significantly more comprehensible than listeners from a non-Romance L1 background. Indeed, comprehensibility ratings by Romance L1 listeners averaged approximately 0.5 points higher than those from non-Romance L1 listeners.

3.3. Acceptability

The final set of models demonstrated the impact of speaker accent and listener background factors on listeners’ ratings of the speakers’ acceptability in three educational contexts: as an e-tandem partner, a foreign language teacher, and a speech model in a language textbook.

Table 10. Summary of fixed effects for acceptability ratings by context.

Fixed Effect	Estimate	SE	df	t	p	Mar. R2	Cond. R2
E-tandem						0.099	0.395
(Intercept)	4.458	0.196	77	22.723	<0.001 ***
Speaker accent—Northern	−0.134	0.098	492	−1.373	0.170
Speaker accent—Southern	−0.456	0.100	499	−4.566	<0.001 ***
Familiarity	−0.042	0.075	478	−0.553	0.581
Contact	0.010	0.008	47	1.309	0.197
Heritage	−0.665	0.221	41	−3.008	0.004 **
L1—Romance	−0.172	0.279	41	−0.619	0.540
Teacher						0.078	0.391
(Intercept)	4.311	0.208	76	20.752	<0.001 ***
Speaker accent—Northern	−0.163	0.102	492	−1.600	0.110
Speaker accent—Southern	−0.492	0.104	498	−4.722	<0.001 ***
Familiarity	−0.094	0.079	484	−1.194	0.233
Contact	0.010	0.008	47	1.195	0.238
Heritage	−0.529	0.236	41	−2.248	0.030 *
L1—Romance	0.101	0.297	41	0.342	0.734
Textbook						0.065	0.368
(Intercept)	4.249	0.209	77	20.322	<0.001 ***
Speaker accent—Northern	−0.197	0.105	492	−1.874	0.062
Speaker accent—Southern	−0.424	0.107	499	−3.965	<0.001 ***
Familiarity	−0.096	0.081	475	−1.187	0.236
Contact	0.009	0.008	47	1.100	0.277
Heritage	−0.479	0.235	41	−2.036	0.048 *
L1—Romance	0.189	0.296	41	0.637	0.527

* p < 0.05. ** p < 0.01. *** p < 0.001.

summarizes the fixed effects for each of these models. There was a statistically significant effect of speaker accent across all three educational contexts, including e-tandem (F(2, 497) = 10.895, p < 0.001), teacher (F(2, 497) = 11.468, p < 0.001), and textbook (F(2, 497) = 7.864, p < 0.001). Additionally, the models revealed a statistically significant effect of listener status as a heritage language learner for the e-tandem (F(1, 41) = 9.045, p = 0.004), teacher (F(1, 41) = 5.052, p = 0.030), and textbook contexts (F(1, 41) = 4.145, p = 0.048). As indicated by the marginal R² values, the fixed effects accounted for 6.5–9.9% of the variance in acceptability ratings, while the random effect of listener accounted for an additional 29.6–31.3% of the variance in these models.

To visualize the impact of speaker accent on acceptability ratings, Figure 4 illustrates each of the three educational contexts. In all three contexts, the mean rating (marked by the white diamond) for the Southern-accented speakers was statistically significantly lower than the ratings for the Central- and Northern-accented speakers. Similar to the comprehensibility ratings, however, the majority of acceptability ratings remained above neutral (3.0) for all three accent conditions across all three educational contexts, indicating that students generally found these accents acceptable for the different educational contexts.

Regarding the impact of listener status as a heritage learner of Italian on acceptability ratings,

Table 11. Estimated marginal means of acceptability ratings by speaker accent and heritage status.

Context Speaker Accent	Non-Heritage			Heritage
	Emmean	SE	95% CI	Emmean	SE	95% CI
E-tandem
Northern	4.27	0.17	[3.93, 4.61]	3.6	0.203	[3.2, 4.01]
Central	4.4	0.17	[4.06, 4.74]	3.74	0.203	[3.33, 4.15]
Southern	3.95	0.171	[3.61, 4.29]	3.28	0.203	[2.87, 3.69]
Teacher
Northern	4.15	0.18	[3.78, 4.51]	3.62	0.216	[3.18, 4.05]
Central	4.31	0.18	[3.95, 4.67]	3.78	0.216	[3.35, 4.22]
Southern	3.82	0.181	[3.45, 4.18]	3.29	0.215	[2.86, 3.72]
Textbook
Northern	4.09	0.181	[3.72, 4.45]	3.61	0.216	[3.17, 4.04]
Central	4.28	0.181	[3.92, 4.65]	3.8	0.216	[3.37, 4.24]
Southern	3.86	0.182	[3.49, 4.22]	3.38	0.216	[2.95, 3.81]

reports the estimated marginal means for acceptability rating by educational context. For all three contexts, there is a statistically significant difference between responses, with non-heritage learners rating speakers as consistently more acceptable for these academic contexts than their heritage learner counterparts in this study. Interestingly, although the heritage learners found these speakers to be less acceptable for these educational contexts, the mean ratings were all above neutral (3.0) and should therefore be interpreted with caution.

4. Discussion

In this exploratory study, we investigated the impact of speaker accent and listener background factors on IFL learners’ perceptions of intelligibility, comprehensibility, and acceptability of regionally accented Italian speech. Across all three perceptual constructs, there was a statistically significant effect of speaker accent—namely, the Southern Italian accent was associated with the lowest intelligibility, comprehensibility, and acceptability ratings. Individual learners’ mean accuracy rates for transcribing Southern-accented utterances ranged from 33.5% to 95.3% (M = 64.81, SD = 15.15), demonstrating variability that was not explained through any interaction effects in this data set. For instance, further examination of the seven individual learners whose transcriptions were most accurate (>80% accuracy) showed that this group included both L1 English and Spanish, heritage and non-heritage, and varying levels of familiarity with Southern-accented Italian. Additionally, it is important to note that overall, the mean comprehensibility and acceptability ratings for Southern-accented speech remained at or above the “neutral” level, indicating that learners’ perceptions were generally positive. This suggests that introducing IFL learners to regionally accented varieties may therefore carry relatively limited risk of learners feeling that they cannot understand these accents or worrying about poor performance on listening tasks in their language classes. This in itself is a promising finding in terms of greenlighting the incorporation of regionally accented speech into the classroom (Major et al., 2005; Stollhans, 2020; Wuensch & Bolter, 2020).

This finding also offers insights into how intermediate-level learners responded to the three different regional accents, which can inform the logistics of including such speech in the IFL classroom. Given that learners found the Central and Northern accents to be more understandable and easier to understand, these accents may be a reasonable starting point to ease learners into their exposure to varieties beyond the “neo-standard” Italian typically presented in textbooks (Cutrì, 2016). For an accent that may seem more challenging to learners (i.e., the Southern Italian accent in this study), learners may need more scaffolding to develop their confidence in their ability to understand these speakers. Instructional approaches could include reading while listening to help with mapping the audio input to its familiar written form (Chang, 2009) or explicit instruction on segmental and suprasegmental features of Southern Italian that may have the most impact on listener comprehension (for an explicit instruction example with Spanish, see Schoonmaker-Gates, 2017).

Listener background factors also demonstrated some statistically significant associations with perceptual measures. Firstly, L1 background had a positive effect on intelligibility and comprehensibility, with listeners from a Romance L1 background (i.e., Spanish) displaying greater accuracy and ease of understanding than their non-Romance L1 peers, regardless of the speaker accent. This may reflect a sort of interlanguage speech intelligibility benefit (Bent & Bradlow, 2003), whereby learners from the same language family background are able to rely on cross-linguistic similarity in their processing of target language input due to a reduced linguistic distance (Carneglutti et al., 2022; Nelson et al., 2021). As one reviewer mentioned, it is also worth considering that this group may have other distinguishing characteristics beyond being L1 Romance speakers that account for this finding. An analysis of their other demographic characteristics shows that they primarily identified as intermediate-level IFL learners, they were enrolled in Italian classes taught by multiple teachers across two of the three universities we sampled from, and they all had FL learning experience. None of these characteristics are notably different from those of the L1 English speaking participants, who were also primarily intermediate-level IFL learners from eight different classes across three universities and who all had FL learning experience. In other words, of the demographic variables we measured, L1 background appears to be the only systematic difference between these two groups. Given the relatively small number (n = 8, 17%) of Romance L1 listeners in our sample, however, we acknowledge that this finding should be interpreted with some caution.

In terms of other measures of experience with Italian language, contact emerged as a statistically significant predictor of comprehensibility only, while accent familiarity did not show any statistically significant effects in this study. This contact finding is in line with research on the benefits of exposure for facilitating cognitive processing (e.g., Clarke & Garrett, 2004; Floccia et al., 2009; Witteman et al., 2014, 2015). Under the assumption that comprehensibility is a measure of processing ease/difficulty, then contact with Italian speakers at home and/or in the community provides opportunities for learners to adjust to a range of different accents and speech varieties. We anticipated that accent familiarity would have a similar effect; however, learners’ reported familiarity with the regional accents in this study was generally low (see Figure 1). Furthermore, familiarity was measured in the abstract (i.e., How familiar are you with Central/Northern/Southern Italian accents?) rather than in response to audio files. Thus, our measure may not have sufficiently captured listeners’ familiarity with the actual speech included in the study (Miao & Kang, 2023). It is also possible that learners may, in fact, have limited familiarity with regional variation in Italian, given that IFL textbooks do not typically present information about regional accents or include these varieties in audio materials (Cutrì, 2016).

For the final listener background factor, heritage learner status, our models indicated a statistically significant effect on acceptability ratings, with heritage learners reporting less positive views of the acceptability of regionally accented speakers as educational models, similar to Wheeler and Kang (2022). Notably, these ratings did not indicate that they found the speakers unacceptable but rather that these ratings were slightly less acceptable than those of the non-heritage language learners in the study. The reasons for this are not entirely clear. It may be that heritage learners have internalized views from Italian-speaking family members about “standard” Italian being appropriate for the educational domain. This finding may also indicate a degree of linguistic insecurity that heritage learners may hold in relation to the use of “home” varieties in more formal settings (Preston, 2013). This is an area that could benefit from further qualitative or mixed-methods research to explore Italian heritage learners’ beliefs and experiences in relation to regionally accented speech.

On the pedagogical side, these listener findings demonstrate that an awareness of learners’ backgrounds can help with student-centered and potentially differentiated instruction when it comes to including regionally accented speech as learning models. For instance, teachers may be able to draw on learners’ L1 backgrounds to spotlight similarities and differences between segmental and suprasegmental features in the target language and the students’ L1(s). The finding about the benefit of contact is particularly promising, as this is something that teachers can facilitate for their students. Activities such as language exchanges and guest speakers from different regions can expose students to different language varieties while providing an opportunity for meaningful communication in their target language. Technology can also help with providing audio models that learners can repeat, slow down, or listen to with captions.

Despite the pedagogical implications offered here, there are many opportunities for related future research. First, while we attempted to control for learner proficiency by recruiting from intermediate- and advanced-level IFL courses, two of the intermediate-level learners still reported that they could not understand the vocabulary in some stimuli (although they achieved 63.5% and 59.8% accuracy on their transcriptions). This, along with a couple of anecdotal comments from teachers, suggested that there might have been more variation in proficiency than expected. Future research could include a study-specific measure of proficiency as a covariate in the analysis, although we believe this variation may also reflect the nature of post-beginner LCTL classes, where there may be limited demand and thus a greater range of proficiency levels within the same course. Second, this study was exploratory and therefore did not purposefully sample for all listener background factors measured. Knowing now that L1 background, contact, and heritage status may have an impact, it would be valuable to explore these factors more systematically. We would also recommend adjustments to the measures of heritage status to specify which generation (e.g., second, third) of heritage learners they are, as this study did not include this level of detail. Third, there are opportunities to incorporate qualitative methodologies here, particularly when exploring heritage learners’ perceptions of acceptability. Finally, these findings can inform intervention-based research to examine how learners respond to the inclusion of regional accents in the FL classroom and the impact of such interventions on real-world communication in the target language.