L1–L2 Influence in Intonation: A Case of Russophone Immigrants in Brazil

Abstract

This paper is devoted to the features of sentence prosody (intonation) in Brazilian Portuguese spoken by immigrants whose first language is Russian, and explores the consequences that L1–L2 influence in intonation may have for communication. The study addressed four research questions: (1) Do Brazilian Portuguese L2 speakers with Russian L1 always succeed in producing the correct utterance type? (2) Can L1–L2 influence lead to misunderstanding of connotations? (3) Is it possible that sometimes L1–L2 influence leads to being perceived as too emotional or not emotional enough? (4) Can L1–L2 influence in intonation be a significant factor in the perception of accent? In a perceptual experiment, productions of four target utterances in Brazilian Portuguese by Russian L1 and Brazilian Portuguese L1 speakers were evaluated by 124 Brazilian listeners in terms of sentence type, possible connotations, accent and arousal. The target utterances included three questions of different types and an exclamation. The findings revealed that the speaker’s L1 influenced the perception of prosodic meanings by Brazilian listeners. In some cases, interference from Russian melodic contours caused the incorrect identification of the sentence type in Brazilian Portuguese. However, even when sentence type was perceived correctly, differences could be found regarding the perception of arousal or accent.

1. Introduction

1.1. Russophone Immigrants in Brazil

Migration of Russian-speaking people to other countries presents a special interest since the beginning of the 20th century when the wars and Bolshevik Revolution induced significant migrant waves (Ruseishvili 2016). Brazil was one of the destinations for these migrants, and this provides unique opportunities for studies of heritage Russian (Sekerina et al. 2023; Skorobogatova et al. 2021; Smirnova Henriques et al. 2022).

The collapse of the Soviet Union in 1991 induced a new wave of Russian-speaking migration (Aleshkovski et al. 2018), and even before the recent conflict between Russia and Ukraine, the Russian Federation had one of the largest populations of citizens living outside their country of origin, at over 10 million emigrants (IOM—International Organization for Migration 2018). However, the presence of Russian-speaking migrants from the post-Soviet wave in Brazil until recently was not so expressive. Following the SISMIGRA data tabulated by (Observatório das Migrações em São Paulo 2020; Smirnova Henriques and Ruseishvili 2019), between 2000 and 2020, 7031 Russian citizens received authorization of residence in Brazil. Considering other countries from the post-Soviet space where Russian is widely spoken, during the same period, authorization of residence in Brazil was received by 4261 immigrants from Ukraine, 548 from Kazakhstan, and 265 from Belarus. Following the SISMIGRA data, 28% of the Russian citizens residing in Brazil live in the São Paulo state. In 2018–2020, a unique database of Portuguese spoken as a second language by Russophone immigrants was created in São Paulo (Smirnova Henriques et al. 2020). Before the COVID-19 pandemic, the database, later named RusIm-Bra1, included about one hundred hours of audio and video recordings of 40 speakers of Russian as a first language.

The speech material contained in the RusIm-Bra1 database is suitable for research of how Brazilian Portuguese is acquired as a second language by Russian L1 speakers; see e.g., the publications devoted to acquisition of segmental features (Smirnova Henriques et al. 2019a, 2019b). Russian L1 speakers who contributed to the database took part in other related experiments such as verbal working memory assessment in Russian and Portuguese languages (Skorobogatova et al. 2021), and evaluation of bilingual profiles. In addition, the database samples have been used for perception experiments, e.g., perception of voice pleasantness of foreign-accented speech by Brazilians (Smirnova Henriques and Madureira 2020). A survey presented by Smirnova Henriques et al. (2020) reports on the perception by Russophone immigrants of their pronunciation difficulties in Brazilian Portuguese: 68% reflected on difficulties in producing some sounds, 61% in understanding how the sounds are produced, 32% in understanding Brazilian Portuguese grammar, and, importantly, 16% of learners mentioned having difficulties in using the correct intonation—the insight that is particularly relevant to the present paper.

The large-scale conflict between Russia and Ukraine in 2022 induced new significant migration waves of Russian speakers, for both Ukrainians frequently bilingual in Ukrainian and Russian and Russian citizens (Smirnova Henriques and Tesko 2022). While Ukrainians are welcomed by many countries and rarely choose Brazil, Russian citizens have fewer options. Brazil is one of a few countries distant from the post-Soviet space that receives Russian refugees crossing the border without visas and bureaucratic complications (Mantovani 2022). The number of Russian-speaking migrants who entered Brazil during the last two years is difficult to evaluate, but, in the current conditions, studies of Brazilian Portuguese acquisition by Russian L1 speakers seem to be especially relevant.

1.2. Theories of L2 Learning and Cross-Language Influence in Intonation

When a person speaks two languages, it is almost inevitable that the languages have some influence on each other. There is a range of publications on how segmental features are affected by cross-language influence (Lein et al. 2016; Ulbrich and Ordin 2014). In recent years, however, more and more attention has been drawn to prosody and intonation (see, e.g., the very recent special issue of Languages “The Effects of Cross-Language Differences on Bilingual Production and/or Perception of Sentence-Level Intonation” (Mennen and Colantoni 2023)). So far, this field remains under-studied, as most publications only cover certain aspects of L1–L2 influence, while no single language contact situation seems to be fully described in terms of all possible prosodic features involved.

When a person starts learning an L2, their first attempts to produce L2 utterances often reveal some of the features typical of their L1. This observation is explained in the two most influential theories of L2 learning, SLM (Flege and Bohn 2021) and PAM-L2 (Tyler 2019), which we will talk about further in this section. In papers dealing with L1–L2 influence in intonation and prosody, it is often claimed that one of the limitations of Flege’s Speech Learning Model (SLM) is that it deals with segmental features. However, its main propositions apply to prosodic changes as well with some adjustments. The revised version of Flege’s model, SLM-r (Flege and Bohn 2021), proposes that, initially, when one learns an L2, the sounds of the new language are automatically related to the phonetic system of L1. (This is what differentiates L1 acquisition from L2 learning: in the case of L2, there is already at least one language to compare with, as opposed to L1, which is acquired from scratch. Note that Flege’s model deals with sequential L2 learners.) With time, new categories arise, as well as some mixed categories between L1 and L2 (Flege and Bohn 2021, p. 23)—which also explains the changes in L1 caused by L2. In intonation learning, there is evidence of L1–L2 influence, such as using L1’s melodic patterns in L2 (Busà and Urbani 2011); the opposite direction of influence, that of L2 on L1, is also reported for several languages (Mennen et al. 2022).

At the same time, suprasegmental “categories” are claimed to be more complex than the segmental ones (Mennen 2015), involving a range of features—fundamental frequency, duration, intensity, voice quality and pausing; in addition, various prosodic features are interrelated in acquisition and perception (van Maastricht 2018). It is a methodological challenge to even establish those categories, e.g., for sentence prosody. However, with the development of ToBI, ToBI-like, and other systems for a significant number of languages (see e.g., Jun 2005), we may now have a tool to at least compare the sets of phonological units across languages.

Many papers, however, focus on cross-language differences in the phonetic realizations of intonation patterns. These differences may consist of peak alignment, melodic intervals, steepness of melodic change and others (Hirst and Di Cristo 1998). There is evidence that cross-linguistic influence not only concerns categories (such as pitch accents and boundary tones) but also the entire melodic curve of an utterance (Pešková 2023). But there is even more ground for language differences. In his paper on intonational typology (Ladd 2001), Ladd provides a list of the possible dissimilarities. Importantly, languages may differ in the mapping between melodic contours (“tunes”), i.e., the same tune may be used in questions in one language (or dialect), and in statements in another language (or dialect)—like the high terminal tone in dialects of English (Grabe 2004). This observation enables us to look at language typology from a different angle. While some comparisons rely primarily on the form (the contour shape and its phonetic realization), this other view is content based (relying on semantics, or meaning). (It is worth noting, though, that form-based approaches do not usually ignore semantics.) In cross-linguistic comparisons, the content-based approach is as viable as the form-based approach, as there have been several successful attempts to compare languages within this framework; see, e.g., the approaches suggested by Hirst and Di Cristo (1998) and Nikolaeva (1977).

Under Ladd’s typological approach (Ladd 2001), there are other sources of cross-language differences. Based on these ideas, Mennen developed a theoretical model that enables a cross-language comparison in terms of intonation and helps predict L2 intonation learning (Mennen 2015). The LILt (L2 Intonation Learning theory) recognizes four dimensions of L1–L2 differences/similarities: (1) the inventory and distribution of categorical phonological elements (“systemic dimension”); (2) the phonetic implementation of these categorical elements (“realizational dimension”); (3) the functionality of the categorical elements or tunes (“semantic dimension”); (4) the frequency of use of the categorical elements (“frequency dimension”). The first three dimensions have already been discussed above, except for variability in the distribution of the categorical elements. This is concerned with either combinatory possibilities of the elements, or “tune-text association”—that is, how melodic patterns align with the structural elements at different levels of linguistic analysis (e.g., syllables or words). An interesting example from Ladd (2001) concerns the placement of nuclear accent in non-emphatic yes/no questions: either the final word of the utterance (e.g., in English “Did he buy CHOCOLATES?”), or the finite verb, irrespective of its location (e.g., in Greek “AGORASE sokolates?”, and Russian).

Dimension 4 in the LILt is another curious aspect of how languages may differ. If we assume that a learner’s language experience is largely based on the input (what he or she hears or reads), the frequency of elements must have a significant influence on the process of learning. As reported in the literature, this is indeed the case, e.g., according to Willems (1982), Dutch learners of English use more rising pitch accents than falling ones in their English speech, as they do in their L1, while English L1 speakers use falls more frequently than rises. This difference can potentially be a source of intonational accent in L2.

It has been shown that L1–L2 language influence is observed at all of these dimensions, although not all of them are equal in terms of difficulty in L2 learning; see e.g., a recent series of experiments with Czech learners of Spanish and Italian, and German learners of Spanish (Pešková 2023). Apart from the four dimensions of LILt, Peškova suggests two more dimensions: syntactic and stylistic, as they are both connected with challenges in learning intonation.

Going back to Flege’s ideas as discussed in a recent publication (Flege and Bohn 2021), L2 learning of segmental features is somewhat in line with the LILt theory, despite the fact that the latter was resigned for intonation research. For segmental features, a contrastive analysis of two languages would involve comparing phoneme inventories, phonetic features of allophones’ realizations (e.g., voice onset time) and allophonic variation in context (Flege and Bohn 2021). These correspond to dimensions 1 and 2 of the LILt. Dimension 3 (“semantic”) is not applicable for segmental features, as there is no “meaning” in phonemes, while the existence of phonological contrasts (minimal pairs), and therefore a virtual connection with meaning, is implied by definition to all phonemes in the inventory. Dimension 4, though, is not included in the SLM, and this aspect has not been discussed much in the literature, to the best of our knowledge. Potentially, including this dimension into the research paradigm of L2, segmental features could be beneficial for predicting the L2 learning progress; if frequency matters for such categorical elements as “tunes”, why should it not matter for phonetic elements?

Flege’s Speech Learning Model is often compared with the PAM-L2 (Tyler 2019), an extension of the original Perception Assimilation Model (Best and Tyler 2007) which explained the perception of phones of non-familiar foreign languages, but not L2. In the new version of PAM, PAM-L2, the original principles were applied to L2, with the idea that at the beginning, L2 learners perceive the sounds of L2 in the same way as they would perceive the sounds of a completely unfamiliar language. PAM-L2 models the possible scenarios of the relationship between the phones of L1 and L2 in the process of learning and predicts the learning outcomes concerning segmental units.

Interestingly, PAM has proven suitable for prosody research. Its other extension, PAM-s (So and Best 2008), has been used in studies of suprasegmental aspects of foreign language perception. So and Best have shown that, as perceived by English L1 speakers, Mandarin tones are associated with similar melodic curves used in sentence prosody (So and Best 2008), e.g., Tone 2 is associated with questions. Although this application of PAM does not deal with L2 learning (as those English speakers did not learn Mandarin Chinese), there are no obstacles to using it with L2 learners.

The present paper explores how intonational accent in L2 influences the perception of the whole utterance by local residents. In terms of LILt, we are dealing here with the semantic dimension. Concerning semantics, there is a crucial difference between the segmental and intonational units: phonetic/phonological units do not have a meaning of their own, whereas intonational units do. This is why neither PAM nor SLM nor their extensions provide an approach capable of dealing with intonational units. Even more, models for learning lexical units would not fit here either, due to the different nature of intonational meaning.

One could speak of three types of discourse meaning signaled by prosody (Cole 2015): focus or information status, illocutionary force and affective meaning; prosody also signals information from the situational context. Intonational accent can be manifested in any of these aspects: using the wrong contour to signal focus may lead to misunderstanding of the logical connections; wrong use of the question contour may lead to misperception of the utterance as a statement or an exclamation; wrong use of means to signal modality may lead to a misunderstanding of the speaker’s attitude to what he/she is saying.

In world languages, different combinations of prosodic features are used for signalling different types of meaning. This is why it is so difficult to master foreign language prosody. In a similar way as was done in PAM (Best and Tyler 2007), we may try to make a list of all possible scenarios in L2 learning concerning intonation. Here, we will speak about Portuguese learners who encountered this language as adults; therefore, in line with both SLM-r and PAM-L2, we assume that L1 would influence L2 learning. First, L1 and L2 may have a contour that is similar both in form and meaning. In this case, we may expect successful productions in L2, which could be proved by correct identification of illocutionary force (sentence type) and focus placement, and correct perception of affective meaning. At the same time, even if the contours in L1 and L2 are identical, the L2 speaker’s speech may still be perceived as accented by local residents, as there are many other aspects of sentence prosody apart from the inventory of melodic contours, such as peak alignment or melodic movement in the pre-nuclear part of the utterance.

Another scenario could include those cases when there is a similar contour in L1 and L2 which differs in the intonational meaning. Here we could expect intonational errors at the early stages of L2 learning, that could go away with time as speakers gain more language experience—due to associations that arise between L1 and L2 categories, as described in PAM (see, e.g., the example of Mandarin tones mentioned above (So and Best 2008)).

A third possibility is that, for a particular meaning, the L2 contour is different from the one in L1. At the early stages of language learning, we may expect the use of L1’s contour in L2 speech. This may lead to significant misunderstanding of the intonational meaning by listeners and, potentially, to miscommunication. That is, these cases are most crucial for the successful use of language and integration into the new society. This is why this scenario lies at the center of the research presented in this article. The goal of this research is to find out which types of utterances require particular attention in L2 classrooms because using the wrong intonation could hinder successful communication.

1.3. Research Questions

The main research question of the present research could be formulated as “Concerning Russophone immigrants in Brazil, what consequences might L1–L2 influence in intonation have for communication?”.

As for our knowledge, there is only one publication on cross-language influence for this particular pair of languages and this particular direction of influence. Some intonational differences in Brazilian Portuguese spoken by speakers of Brazilian Portuguese as L1 and Russian L1 speakers are discussed by Lucente et al. (2023). According to this research, speakers of Brazilian Portuguese as L1 tend to start their utterances in a higher fundamental frequency, although the F₀ range is wider in the speech of Russian L1 speakers; there are also slight differences in peak alignment. In addition, some Russian L1 speakers’ recordings in Portuguese seemed to reveal intonational features typical for Russian intonation. One case of Russian L1 influence on Brazilian Portuguese speech concerned the division of relatively short complex sentences with an initial adverbial clause (e.g., “Quando chegar em casa, me liga” (When you come home, call me)) into intonational phrases: Russian L1 speakers often produced them as two IPs, while Brazilian Portuguese L1 speakers did so as one. Another case was observed in a yes/no question with a contrastive focus on the final word (target utterance “Vai pegar o bonde?” (Will you take the tram?) followed by “E eu, o ônibus”(And I’ll take the trolleybus)). In this example, the final word in speech in Portuguese produced by Russian L1 speakers gained much more prominence than in recordings of Brazilian Portuguese L1 speakers.

Within the language pair of Russian L1 and Portuguese L2, the other direction of influence, L2 on L1, has also been reported (Head and Semenova-Head 2010; Kachkovskaia et al. 2023).

There are also some publications on L1–L2 influence in intonation for Russian L1 speakers learning English. In an experiment performed by Crosby (2013), there was some evidence of Russian L1 influence on L2 English speech for yes/no questions. In another study presented by Poroshina (2022), the author observed a change in perception of affective meaning in English imperatives produced with Russian intonation. Another publication (Holden and Hogan 1993) also reports on the influence of Russian intonation in English L2 speech on the perception of negative emotions: Russian intonation was rated as more negative.

In a series of perception tests, Kublanova (2003) showed that using Russian intonation in English speech may lead to changes in the perception of emotions (both erroneous perception of emotional meaning and strength of emotions) and misinterpretation of illocutionary force (e.g., a polite request may be interpreted as an order); intonational influence also affects perception of accent. In the same thesis, there are data on particular sentence types. Thus, yes/no questions produced with Russian intonation are perceived as exclamations, often with the emotional meaning of surprise; echo questions requesting to repeat (e.g., “What was the last phrase you said?”) are perceived as surprised due to the higher tone. Another cross-linguistic study (Rathcke 2009) reports that Russian high boundary tone is associated by German listeners with friendliness, as opposed to incompleteness as judged by Russian L1 speakers.

Exploring the semantic dimension of LILt with recordings from Czech learners of Spanish and Italian, and German learners of Spanish, Peškova concluded that L2 learners could distinguish and reproduce the illocutionary force (e.g., statements vs. questions) but had more difficulties acquiring the pragmatic nuances and emotions (in such contexts, they tended to use unmarked patterns or patterns transferred from L1) (Pešková 2023). Potentially, such transfer could lead to misunderstanding, e.g., Czech learners of Spanish often produced L2 neutral vocatives with L%, which is typical for statements/wh-questions, instead of the required (H)!H%.

Based on this evidence from the literature, we hypothesized that in the Brazilian Portuguese speech of Russian L1 speakers, the influence of the Russian intonation system could lead miscommunication with respect to the following meanings of prosody (but not excluding the other):

• Illocutionary force;
• Affective meaning;
• Situational context (accent).

In particular, the paper will address the following questions:

• Do Brazilian Portuguese L2 speakers with Russian L1 always succeed in producing the correct utterance type, e.g., are all questions perceived as questions?—illocutionary force;
• Can L1–L2 influence lead to misunderstanding in terms of more distinct shades of meaning (connotations), e.g., if a question is perceived as a question, is it the correct type of question?—illocutionary force;
• Is it possible that sometimes L1–L2 influence leads to being perceived as too emotional or, on the contrary, not emotional enough?—affective meaning;
• Concerning Russophone immigrants in Brazil, can L1–L2 influence in intonation be a significant factor in the perception of accent?—situational context.

To answer these questions, we have taken the following steps. First, we compared the intonation systems of both languages in order to find those sentence types where L1–L2 influence could manifest itself very clearly (see Section 2.2). Then, we selected target utterances for the corresponding sentence types from the RusIm-Bra1 speech corpus (ibid.). After that, we chose five recordings for each target utterance so that two of the recordings were produced by Brazilian Portuguese L1 speakers and three by Russian L1 speakers with, supposedly, varying degrees of intonational accent in the target utterances (see Section 2.3). Finally, we ran a perception test where the Brazilian participants listened to these recordings and answered several questions related to our research questions 1–4 (see Section 2.4). The perception data were then subject to statistical analysis (see Section 3).

1.4. The DaTo System Framework

For comparison of intonation patterns produced by speakers of Brazilian Portuguese as L1 and L2, we are using the DaTo framework (Lucente 2012, 2017, 2022), which provides information on the height, movement and alignment of the melodic contours.

The DaTo (dynamic tones) system serves as an annotation system for intonation, focusing on dynamically describing the melodic contours specific for Brazilian Portuguese intonation (Lucente 2012, 2017, 2022; Lucente and Barbosa 2009). This system consists of manually assigned labels for describing melodic contours and an automatic boundary detection mechanism that relies on the presence of phrasal accents (Barbosa 2006).

The annotation rules of the DaTo system diverge from the linear structure seen in annotation systems grounded in autosegmental-metrical phonology (Goldsmith and Pierrehumbert 1992), including those developed in the aftermath of the ToBI system (Silverman et al. 1992; Pitrelli et al. 1994). The dynamic interpretation of melodic contours by the DaTo system arises from its phonetic interpretation of the sequence of tonal events within an utterance.

The DaTo framework incorporates the concept of a glottal mechanism (Lucente n.d.) that concurrently governs both voicing and tonal adjustments in speech. According to the hypothesis surrounding this glottal mechanism, the only feasible modifications involve an increase in F₀ values and the regulation of descending values. Elevating the F₀ value gives rise to three ascending contours: LH (rising), >LH (late rising), vLH (compressed rising), and HLH (falling–rising). The regulation of descending F₀ values results in falling contours: HL (falling), >HL (late falling), and LHL (rising–falling). The system denotes boundaries with the labels L and H, referred to as boundary levels (see Figure 1).

The phonological aspect of the annotation is evident through the use of the letters L and H, representing low and high values on the melodic curve. This choice aligns with proposals from intonational phonology (Ladd 1996; Pierrehumbert 1980) and the autosegmental-metrical theory. However, the characterization of each dynamic contour using these labels takes into account (i) the direction of the F₀ movement; (ii) the F₀ range; (iii) the rates of F₀ rising and falling; (iv) and the alignment with segmental components of speech. The following figures provide a visual representation of the dynamic aspects considered in determining the annotation labels for melodic contours in the DaTo system.

Figure 2 displays the interpolated and smoothed melodic curve (with bandwidths of 10 Hz and 5 Hz) for the spontaneous Brazilian Portuguese (BP) utterance: “Mesmo o Brasil ganhando, mesmo o Brasil perdendo, amanhã o sol estará a pino às seis e quarenta e cinco da manhã” (Even if Brazil wins, even if Brazil loses, tomorrow the sun will be high at six forty-five in the morning). In the annotation tiers, arranged from top to bottom, the following elements are depicted: first, the points indicating the maximum speed of F₀ change in the rising (R) and falling (F) movements, which serve as boundaries of intonation contours; second, the stress groups. Each of the six quadrants in the figure illustrates the melodic movement, delineated by the points of maximum speed of F₀ change, associated with the respective melodic contour indicated below in the quadrants.

When examining the rising contours in rectangles 1, 3, 4 and 5, differences in F₀ range and rates of rise and fall become apparent. While >LH exhibits a higher range and rate of rise, LH in rectangles 3 and 4 shows similar values between them. The rise observed in rectangle 5 is not associated with any prominence at the utterance level; hence, it is considered a preparatory dynamic contour, necessary for the subsequent occurrence of the HL contour (rectangle 6). Regarding falling contours, LHL in rectangle 2 has a greater F₀ range than HL but with a slower falling rate. These contours consistently precede the boundaries of intonational units.

2. Materials and Methods

2.1. Speakers That Produced the Stimuli

Ten Brazilian Portuguese L2/Russian L1 speakers, eight women, and two men, were selected among the participants of the oral corpus containing speech in Portuguese and Russian produced by Russophone migrants residing in São Paulo, recorded in 2018–2020 (Smirnova Henriques et al. 2020), later named RusIm-Bra1. Six control subjects, three women (Br1, Br3 and Br5) and three men (Br2, Br4 and Br6), Brazilians, who resided in São Paulo state for their whole life and speak Brazilian Portuguese as L1, were recorded in the same period. The tasks performed by the corpus participants and control subjects had already included a task on intonation production (Smirnova Henriques et al. 2022, section “Intonation task”).

The essential information about the Brazilian Portuguese L2/Russian L1 speakers whose recordings were selected for the further described perception experiment is given in

Table 1. Russian L1 speakers residing in São Paulo who recorded the stimuli in Brazilian Portuguese for the perception test (Smirnova Henriques et al. 2020).

Speaker ID	Sex	Country of Birth	Languages Spoken at Home	Age of Moving to Brazil	Time of Residence in Brazil	Education *	Way to Learn Portuguese
Ru1	F	Russia	Russian	26	5	Bachelor/specialist **	Self-taught
Ru2	F	Russia	Russian	28	2.5	Bachelor/specialist **	University course in Russia; private lessons
Ru3	M	Russia	Russian	36	6	Graduated	Six months university course in Brazil
Ru4	F	Russia	Russian	30	2	Bachelor/specialist **	Self-taught
Ru5	F	Russia	Russian, Ukrainian	36	7	Bachelor/specialist **	Self-taught
Ru6	M	Ukraine	Ukrainian	47	6	Secondary school	Four months of lessons
Ru7	F	Russia	Russian	24	6	Incomplete undergraduate course	Self-taught
Ru8	F	Belarus	Russian, Belarusian	36	17	PhD	Self-taught
Ru9	F	Russia	Russian	28	6.5	Incomplete undergraduate course	Self-taught
Ru10	F	Tajikistan	Russian, Tatar	28	1	Master’s degree	Some months of private lessons

* All the participants studied in their home countries, not in Brazil ** The term “specialist” is common in post-Soviet countries in five-year undergraduate programs.

. All the speakers acquired Brazilian Portuguese as adults at least at 24 years old and were sufficiently proficient to give an interview about their immigration experience in Brazil, in Portuguese, for 40 min. No formal evaluation of the proficiency was performed. The median age at the time of recording in this group was 33 years (the minimum and maximum were 30 and 53 years, respectively); the median time of residence in Brazil was six years (the minimum and maximum were 1 and 17 years, respectively, see Table 1). The median age in the control group of six speakers of Brazilian Portuguese as L1 was similar: 32 years (the minimum and maximum were 25 and 46 years, respectively).

For more clarity in explaining L1–L2 influence, we additionally analyzed the four target utterances produced by a native speaker of Russian who does not speak Brazilian Portuguese (female, 24, born in Russia, a resident of Saint Petersburg, Russia, bachelor/specialist).

2.2. Selecting the Target Utterances

As outlined in the introduction, our main goal was to estimate the influence of the Russian L1 intonational accent in Brazilian Portuguese L2 speech. To select the appropriate stimuli, we first compared the intonation systems of both languages. Since this paper explores the semantic dimension of intonation, we compared the melodic contours used for different sentence types.

Table 2. A comparison of Russian and Brazilian Portuguese contours for various sentence types using the DaTo system.

	Russian	Brazilian Portuguese
Non-emphatic statement	LHL L	>LH HL L
Non-emphatic yes/no question	LH L; >LH L	>LH L; LH L
Wh-question	HL L, LHL L, >LH H	LHL L; >LH LHL L; >LH HL L
Contrastive statement	LH L	>LH L, HLH L
Exclamation	LH L; LH H; >LH >HL L	>LH L
Echo y/n question	LH L	>LH H
Echo wh-question	LH L	>LH H; >LH L
Echo wh-question, asking to repeat	LH H; >LH H	>LH L; >LH H
Elliptical question	>LH H	>LH LHL L
Non-finality	LH L; LH H; >LH H	>LH H

presents the results of this analysis using the DaTo annotation system that has already been used for Brazilian Portuguese. (At the present time, there is no comprehensive description of Brazilian Portuguese intonation in terms of ToBI, while in some research the P-ToBI (Frota and Moraes 2016) system, which was developed for European Portuguese, is used).

For Russian, the DaTo labels were based on descriptions presented by Volskaya and Kachkovskaia (2016) and the annotation system ToRI (Odé 2008) as well as speech recordings from RusIm-Bra1. In assigning final boundary tones, we are using symbols L and H only (although the inventories of boundary tones for Russian vary across descriptions—see, e.g., (Paschen 2015) for an extended approach; note also that ToRI also uses M, which is controversial (Volskaya and Kachkovskaia 2016)). At this stage, we do not rely on phrase accents, although recent evidence (Duryagin 2023) shows that adding phrase accents could be helpful in describing post-nuclear part of the melodic curve in more detail. As DaTo is flexible in terms of timing, it is easy to make a difference between the curves used for echo wh-questions (labeled H*H H% in ToRI and LH H in DaTo) and elliptical questions (labeled L*H H% in ToRI and >LH H in DaTo).

Both languages have relatively free word order; as a result, practically any word within an utterance can bear prosodic prominence—depending on the context. There is a difference in producing non-emphatic yes/questions: in Brazilian Portuguese, the main word is the rightmost content word, while in Russian, it is the finite verb independent of its location. Grammatically, yes/no questions are usually not marked in both languages.

In Russian, there is a specific type of sentence that we refer to as an “elliptical question” (e.g., “And Anna?”, with a contrastive meaning); this type has a particular contour not used in other question types. Echo wh-questions, where the speaker is asking to repeat a statement or clarify what has been said, are formed with reduplication of the wh-word, and the main word is the second instance of the wh-word; that is, the main word is necessarily at the beginning of the utterance.

Based on the comparison presented in Table 2 and additional information discussed above, we hypothesized that the most notable L1–L2 influence could be observed in four sentence types:

• Non-emphatic yes/no questions (due to differences in the location of main accent);
• Exclamations (due to unusual contours used in Russian);
• Echo wh-questions, asking to repeat (this being a specific sentence type for Russian with a specific contour);
• Elliptical questions (also a specific sentence type for Russian with a specific contour).

Thus, four target utterances were selected for the perception experiment from the RusIm-Bra1 speech corpus which contained recordings of the intonational task (described by Smirnova Henriques et al. 2022). The task was specially designed for research of cross-linguistic influence on intonation between Russian and other languages. Briefly, it consisted of reading 17 utterances or short dialogues that contained target utterances where a particular Russian melodic contour is expected. The trials were originally prepared in Russian, translated into Portuguese and recorded in both languages by the RusIm-Bra1 corpus participants in 2018-2020 (Smirnova Henriques et al. 2020, also see Section 2.1). Three of them represented different types of questions, and the fourth was an exclamation (see

Table 3. Brazilian Portuguese target utterances selected for analysis (Smirnova Henriques et al. 2022).

Target Utterance Number	Type	Brazilian Portuguese	Translation into English
10	Elliptical question (contrastive)	— Me fale as estações do ano. — Inverno, primavera, verão. — E o outono?	— Name seasons of the year. — Winter, spring, summer. — And autumn?
12	Yes/no question (information-seeking)	— Você viu o apartamento novo do Pedro? — Sim. É pequeno, mas bem confortável.	— Have you seen Peter’s new apartment? — Yes. It is small but very comfortable.
15	Echo question	— A Ana e o Michel moram agora no Sri Lanka. — Onde é que eles moram?	— Anna and Misha now live in Sri Lanka. — Where do they live?
16	Exclamation	Que brincos lindos você tem!	What beautiful earrings you have!

).

2.2.1. Target Utterance 1: The Elliptical Question “E o outono?”

An elliptical question is a specific type of interrogative utterance, often very short, which has a separate melodic pattern in Russian (as well as some other Slavic languages, such as Ukrainian (Nikolaeva 1977)). In terms of meaning, such questions are often used in contrastive/comparative contexts. In many other languages, e.g., English, this type of question is not usually mentioned in intonational descriptions (see, e.g., (Wells 2006)), as it does not require any specific intonation.

In Russian, the standard and almost the only possible melodic pattern for elliptical questions, could be described as a low (fall-)rise (Volskaya and Kachkovskaia 2016) and L*H H% (Odé 2008) (also H+L* LH% (Rathcke 2009)), which could correspond to >LH H in DaTo.

In Brazilian Portuguese, the pattern for elliptical questions, according to the DaTo description, is an initial late rising (>LH) and a downstepping (LHL) preceding the low-level boundary (L) (see Figure 3).

2.2.2. Target Utterance 2: Information-Seeking Yes/No Question “Você viu o apartamento novo do Pedro?”

In most languages of the world, information-seeking yes/no questions are produced with a high final boundary tone (Hirst and Di Cristo 1998). However, Russian is one of the exceptions to this statement; such utterances are produced with a rising–falling tune, and as a result, the final boundary tone is low (Volskaya and Kachkovskaia 2016; Odé 2008). For non-emphatic yes/no questions, we use the DaTo label LH L (labeled H*L L% in ToRI but L*+H L% in, e.g., (Rathcke 2013)). In contexts with no post-nuclear syllables, this pattern is truncated (Odé 2008; Rathcke 2013); as a result, some utterances end in a high tone; however, our material does not contain such examples.

In Brazilian Portuguese, a rising–falling pattern is also possible. The pattern for information-seeking yes/no questions exhibited in Brazilian Portuguese is the initial late rising pattern (>LH), present in almost all initial contexts (Lucente 2015), followed by a slow falling movement interrupted by the falling–rising movement, described as >LH and a low-level boundary tone, as one can see in Figure 4.

2.2.3. Target Utterance 3: Echo Question “Onde é que eles moram?”

In Russian, some echo questions begin with a wh-word repeated twice, e.g., the parallel Russian utterance to context 15 would have the following word-by-word translation: “Where-where they now live?” (Rus. “Где-где oни теперь живут?”). Such questions have only one possible intonation pattern, which is a high rise on the second wh-word followed by a high plateau until the end of the utterance (Volskaya and Kachkovskaia 2016), or a high tone gradually declining up to the end of the utterance due to declination, especially when the utterance is long (Kachkovskaia et al. 2020). In ToRI (Odé 2008), this pattern is described as H*H H%, which corresponds to LH H in DaTo.

Brazilian Portuguese does not have this particular type of repeated question. The closest translation would be an utterance with only one wh-word at the beginning, i.e., “Onde é que eles moram?” (Where do they live?). For this type of question, the two types of boundary levels are possible, which results in a slight change of intention between them. In questions with the pattern >LH >LH H, the communicative intention is close to surprise (see Figure 5, left and middle panels).

2.2.4. Target Utterance 4: Exclamation “Que brincos lindos você tem!”

In Russian, exclamations can be produced using several different intonation patterns, ranging from simple falls to complex patterns consisting of two nuclear accents within the same tone unit (Volskaya and Kachkovskaia 2016). The most frequent melodic pattern is a rise–fall accompanied by higher intensity and often longer vowel duration; the melodic peak is located early in the stressed syllable, which makes this pattern different from that of a yes/no question (Rathcke 2006).

In Brazilian Portuguese, what is observed is a progressive fall in both the ceiling and floor, with the possibility of a local prominence. In the example presented in Figure 6, prominence is associated with the adjective “lindos” (beautiful).

2.3. Selecting the Stimuli

For each of the four selected target utterances (see previous section), we chose five productions from the RusIm-Bra1 speech corpus; that is, the resulting number of stimuli was 20. Given our experimental design, we decided that this is the maximum number of stimuli that we can have within one perceptual test, because for each stimulus, the listeners needed to answer 4 questions, which gave us 80 questions in total. Then, we estimated completion time to make sure it did not exceed 20 min.

The five productions for each target utterance were selected in the following way. In order to obtain balanced data and not attract listeners’ attention to foreign language accent, two of the five productions were selected from the recordings of Brazilian Portuguese L1 speakers; the other three were those of Russian L1 speakers (see Table 1). Then, we selected one recording with the melodic contour typical for Russian, and one recording with a Brazilian-Portuguese-like curve. The fifth recording contained a contour that was not present in our analysis of typical contours for either Russian or Brazilian Portuguese.

The speakers that recorded the stimuli are listed in

Table 4. Selection of recordings for the perception experiment.

Utterance	Russian L1 Speaker’s ID	Brazilian Portuguese L1 Speaker’s ID
10	Ru1, Ru2, Ru3	Br1, Br2
12	Ru3, Ru4, Ru5	Br3, Br4
15	Ru6, Ru7, Ru8	Br4, Br5
16	Ru3, Ru9, Ru10	Br1, Br6

below. The labeling of each stimulus consists of the utterance number (Utt) and speaker ID that includes the L1 identification (“Ru” for Russian; “Br” for Brazilian Portuguese) and an individual number. So, for example, the labeling Utt10_Ru1 means that this recording is composed of utterance 10 (as listed in Table 4) produced by the speaker identified as Ru1 whose first language is Russian (more details on each speaker are shown in Table 1).

2.4. The Perception Experiment

2.4.1. Survey Questions

For each of the stimuli, the listener needed to answer four questions.

Question 1 was used to find out whether the speaker had succeeded in producing the intended utterance type, i.e., whether a question was perceived as a question or an exclamation as an exclamation. The question was formulated as follows: Which punctuation mark would you put at the end of this utterance? (Qual pontuação você colocaria ao final dessa frase?) This was a multiple choice question, with four options: a full stop, a question mark, a comma, and an exclamation mark (see Figure 7).

Question 2 was added to determine whether the speaker had been successful in producing the intended connotations (apart from the intended sentence type). For questions, we tested whether the speaker produced the particular type of question that fit the original context. The listeners were asked to choose one of the most appropriate contexts from a list of three; one of those contexts was the original one (see

Table 5. Contexts used in Question 2.

Target	Alternatives	Brazilian Portuguese	Translation into English
E o outono?	1. Yes/no question (information-seeking)	— Da última vez eu te falei sobre minha estação do ano favorita. — É mesmo, você disse primavera. E o outono? — Sim, primavera e outono.	— Last time I told you about my favorite seasons. — Ok... Let me think... Spring and... And autumn? — Yes. Spring and autumn.
	2. Echo question (asking to repeat)	— Na prova tinha um texto sobre o inverno e o outono… — E o outono? Foi isso que você disse? Não ouvi direito.	— So in the test there was this text about winter and autumn... — And autumn? Sorry, I cannot hear well...
	3. Elliptical question, contrastive INTENDED	— Me fale as estações do ano. — Inverno, primavera, verão. — E o outono?	— Name seasons of the year. — Winter, spring, summer. — And autumn?
Você viu o apartamento novo do Pedro?	1. Yes/no question (information-seeking) INTENDED	— Você viu o apartamento novo do Pedro? — Sim. É pequeno, mas bem confortável.	— Have you seen Peter’s new apartment? — Yes. It is small, but very comfortable.
	2. Echo question (surprised)	— Eu vi o apartamento novo do Pedro? — Sério? Você viu o apartamento novo do Pedro? Quando? — Quarta-feira dessa semana.	— I’ve seen Peter’s new apartment. — Really? You’ve seen Peter’s new apartment? When? — On Wednesday this week.
	3. Yes/no question with a displaced focus	— Você viu o apartamento novo do Pedro? — Não, eu vi o apartamento do Miguel.	— Have you seen Peter’s new apartment? — No, I have seen Mi-chael’s apartment.
Onde é que eles moram?	1. Wh-question (information-seeking)	— Onde é que eles moram? — Acho que é no Sri Lanka.	— Where do they live? — In Sri Lanka, I suppose.
	2. Echo question (asking to repeat) INTENDED	— A Ana e o Michel moram agora no Sri Lanka. — Onde é que eles moram? O que é que você disse?	— Anna and Misha now live in Sri Lanka. — Where do they live? What did you say?
	3. Wh-question, rhetorical, wondering	O que eles estão fazendo? Onde é que eles moram? Eu gostaria de saber.	What are they doing?.. Where do they live?.. I wish I could know.

).

For the fourth target utterance, which was an exclamation, no contexts were provided for choice. Instead, the following question was presented: Does the speaker really like those earrings? (Essa pessoa realmente gostou dos brincos?) This was a multiple choice question with the options “Yes, definitely”, “Probably yes”, “I am not sure”, “Probably not” and “Definitely not” (Com certeza, Provavelmente, Não tenho certeza, Provavelmente não, Definitivamente não).

Question 3 was included based on the idea that in L2, speakers may not have full control of whether they sound too emotional or, on the contrary, not emotional enough. The question was formulated as “How strong is the speaker’s emotion/involvement? Rate from 1 (weak) to 5 (strong)” (Quanto forte é a emoção/envolvimento do falante? Escala de 1 (fraca) a 5 (forte)).

Question 4 was intended as an implicit way to ask whether the speaker’s recording sounded accented. As all the recordings were made by residents of São Paulo, and the São Paulo accent is familiar to most Brazilian listeners, the question was formulated as follows: Can you imagine people in São Paulo speaking this way? (Você imagina uma pessoa de São Paulo falando dessa forma?) This was a multiple choice question with the options “Definitely not”, “Probably not”, “I am not sure”, “Probably yes”, and “Yes, definitely” (Definitivamente não, Provavelmente não, Não tenho certeza, Provavelmente sim, Com certeza).

2.4.2. The Online Survey

The data were collected using the SurveyMonkey platform. The survey was structured in the following way.

Part 1. A brief introduction mentioning that the research was about intonation in Brazilian Portuguese. The listeners were not informed that some of the recordings had been made by speakers of Brazilian Portuguese as L2.

Part 2. Training pages included two questionnaire pages that were inserted to make the listeners familiar with the experimental procedure. Answers obtained from these pages were excluded from the analysis.

Part 3. The main part of the questionnaire consisted of 20 pages presented in random order. Each page included one sound file followed by 4 questions concerning this recording.

Part 4. A set of questions collected information about the listeners such as age, gender, education, profession, place of birth and place of residence and languages spoken.

In total, the survey consisted of 25 pages and 101 questions. The number of responses was 141 and the estimated completion time was 17 min. Of the 141 participants, only 124 completed all the listening tasks.

2.5. Participants of the Perception Test

Out of the 124 participants who completed the survey, 123 filled in the questionnaire containing personal data; 49 of them were male and 74—female. Within age groups, the distribution was as follows: 18 to 24: 10 participants; 25 to 34: 29 participants; 35 to 44: 22 participants; 45 to 54: 32 participants; 55 to 64: 22 participants; over 65: 8 participants. Concerning the level of education, 67 participants had graduated from higher education institutions, 49 had undergraduate degrees, and 7 only finished secondary school. There was a range of professions, from bank officers to university professors, including 6 students.

A total of 122 participants were born in Brazil and 53 of them in São Paulo. At the moment of the recording, 66 resided in São Paulo, 52 in other cities of Brazil and 5 abroad.

2.6. Statistical Analysis

First, we removed from the survey data the answers of those participants who had not completed all the listening tasks. Then, a preliminary analysis was performed using average values calculated across all participants for each stimulus and each question of the survey. The results of this analysis are presented in Section 3.1.

As the main step (Section 3.2), Multiple Factor Analysis (MFA) was used to compare the 124 judges’ assessments of the 20 stimuli. Multiple Factor Analysis, known as MFA (Husson et al. 2011), is a multivariate data analysis method. The data are described by groups of quantitative and qualitative variables. The contribution of each set of variables can be estimated by considering the distance between the variables. The more distant the variables are, the less similar they are, and the closer, the more similar. The MFA method analyzes variables simultaneously and allows for studying their relationships with the stimuli and the influences of the variables in the distribution of the stimuli. To run the MFA statistics, FactorMInerR in the R software (version 4.3.3) was used.

The listeners’ judgments from the perceptual test were considered quantitative variables. Median values for the quantitative variables were calculated and were normalized by z-score. The quantitative variables were as follows:

• Gc1: judgments of sentence type (interrogative, declarative, exclamatory or continuative), Question 1 of the survey;
• Gc2: type of the context chosen to explain the perceived connotations, Question 2 of the survey;
• Gc3: judgments of arousal (emotional involvement), Question 3 of the survey;
• Gc4: accent recognition (Brazilian from São Paulo or not), Question 4 of the survey.

The values of these variables were calculated in the following way. For Gc1 and Gc2, where the listeners were requested to choose one of the answers, the quantitative values were calculated as the percentage of “correct” answers, i.e., the answers that corresponded to what the speakers were reading during the recording of the speech material. For Gc3 and Gc4, listeners’ judgments were scaled from 1 to 5, which enabled us to average those values.

A qualitative variable (Gq1) was also used to refer to the utterance used in the stimuli. The categories were labeled as follows:

w: Utt10, “E o outono?”(And autumn?);

x: Utt12, “Você viu o apartamento novo do Pedro?” (Have you seen Peter’s new apartment?);

y: Utt15, “Onde é que eles moram?” (Where do they live?);

z: Utt16, “Que brincos lindos você tem!” (What beautiful earrings you have!).

In statistical analysis, we did not include the listeners’ personal data (e.g., age, gender or native city) because there were not enough data to take into account these factors.

3. Results and Discussion

3.1. Perception of Prosodic Features and Meanings

In this section, we will analyze data from the perception experiment according to the question that the participants answered and the target utterances that they heard. We will look for sources of cross-language influence by comparing the melodic curves of recordings produced by L1 and L2 Brazilian Portuguese speakers.

For more clarity in explaining L1–L2 influence, we will address the melodic curves for the same four target utterances in Russian produced by a Russian L1 speaker who does not speak Brazilian Portuguese that were presented in Section 2.2 (Figure 3, Figure 4, Figure 5 and Figure 6). All recordings of the target utterances and the corresponding Russian control recordings were transcribed using DaTo. The transcriptions are provided in

Table 6. DaTo labels (aligned with words) for all the stimuli and the four Russian control recordings.

Recording	DaTo Annotation
	E	o	outono?
Utt10_Br1			LHL L
Utt10_Br2			LHL L
Utt10_Ru1			LHL H
Utt10_Ru2			>LH L
Utt10_Ru3			LHL L
	А	oсень?
Utt10_Ru_control		>LH H
	Você	viu	o	apartamento	novo	do	Pedro?
Utt12_Br3		>LH					>LH L
Utt12_Br4		>LH					>LH L
Utt12_Ru3		>LH					L
Utt12_Ru4		>LH			LH		>LH L
Utt12_Ru5		>LH					>LH L
	Вы	видели	Петину	нoвую	квартиру?
Utt12_Ru_control		>LH			L
	Onde	é	que	eles	moram?
Utt15_Br4		>LH			>LH H
Utt15_Br5		>LH			>LH L
Utt15_Ru6					>LH L
Utt15_Ru7	>LH				>LH H
Utt15_Ru8	>LH				LHL L
	Где-	где	oни	теперь	живут?
Utt15_Ru_control		>LH			H
	Que	brincos	lindos	você	tem!
Utt16_Br1		>LH	>LH		>LH H
Utt16_Br6			>LH		>LH H
Utt16_Ru3		>LH	LH		LHL L
Utt16_Ru9			>LH		LHL L
Utt16_Ru10		>LH	>LH		LH L
	Какие	замечательные	у	Вас	серьги!
Utt16_Ru_control		>LH			>HL L

.

3.1.1. Sentence Type

To evaluate the correct identification of the intended sentence type, we analyzed answers to the question “Which punctuation mark would you use at the end of this sentence? (? . , !)”. Below, the results are presented separately for each target utterance, as they represent different types/subtypes of questions and, as a result, may reveal different patterns of cross-language influence.

The figures below present the speakers’ melodic curves temporally normalized to make it possible to compare the alignments of each speaker with the speech segments. Temporal normalization was carried out with the time-normalized_f0.praat script (Arantes 2015) and the figures were generated in the R software (R Core Team 2021).

Utterance 10 “E o outono?” (“And autumn?”) is mostly perceived as a question. The percentage of correct identification of the intended utterance type is close to 100% (see

Table 7. Correctness of question identification for the target utterance “E o outono?” (And autumn?).

Recording	Identification Rate
Utt10_Br1	0.98
Utt10_Br2	0.98
Utt10_Ru1	0.96
Utt10_Ru2	0.86
Utt10_Ru3	0.93

), except for one Russian L1 speaker (Utt10_Ru2, f). Her recording is confused with a statement (5%), exclamation (4%) and unfinished utterance (5%).

The melodic curve for Utt10_Ru2 (see Figure 8) is not typical for either Brazilian Portuguese or Russian speech. In this example, the speaker might have tried to adapt her pronunciation to L2, but she did not succeed.

In utterance 12 “Você viu o apartamento novo do Pedro?” (“Have you seen Peter’s new apartment?”), a Russian L1 speaker might be misunderstood. Utt12_Ru3 (male speaker) has only 32% of correct identification of the intended utterance type (see

Table 8. Correctness of question identification for the target utterance “Você viu o apartamento novo do Pedro?” (Have you seen Peter’s new apartment?).

Recording	Identification Rate
Utt12_Br3	0.97
Utt12_Br4	0.99
Utt12_Ru3	0.32
Utt12_Ru4	0.96
Utt12_Ru5	0.99

). This recording is confused with statements (52%), exclamation (14%) and unfinished utterances (2%).

Utt12_Ru3′s pattern is the one typical for Russian yes/no questions, with a significant melodic peak at the finite verb and no peak at the end (see Figure 8 and Table 6). It is clear, though, that since in Brazilian Portuguese, the final peak is necessary, omission of the final peak would lead to misperception. And, as the final tone is low, this utterance is perceived as a statement instead of a question. The other Russian speakers’ intention to ask a question was perceived well because these speakers produced the required utterance-final peak.

Utterance 15 “Onde é que eles moram?” (“Where do they live?”) is mostly perceived as a question independent of the speaker. For all speakers, the percentage of correct question identification ranged between 96 and 100 percent.

Utterance 16 “Que brincos lindos você tem!” (“What beautiful earrings you have!”) is confused with statements even when produced by Brazilian Portuguese L1 speakers (but more when produced by Russian L1 speakers). Even when the speaker speaks Brazilian Portuguese as L1, this utterance may be perceived as a statement or a question. But with a Russian L1 speaker, the error rate is higher (see

Table 9. Correctness of question identification for the target utterance “Que brincos lindos você tem!” (What beautiful earrings you have!) and rate of incorrect identification.

Recording	Correct Identification Rate (Exclamation)	As Statement	As Question
Utt16_Br6	0.88	0.07	0.04
Utt16_Br1	0.88	0.09	0.03
Utt16_Ru3	0.71	0.21	0.05
Utt16_Ru9	0.66	0.28	0.05
Utt16_Ru10	0.83	0.12	0.05

; for the melodic curves, see Figure 9).

3.1.2. Connotations

To observe finer differences in perception of the target utterances, we analyzed answers to the task of choosing the most appropriate context (for question target utterances—10, 12 and 15). For the exclamation (context 16, “Que brincos lindos você tem!”, “What beautiful earrings you have!”), we analyzed listeners’ ratings for the question “Essa pessoa realmente gostou dos brincos?” (Did the speaker really like those earrings?).

The rate of identifying the intended connotations varies across utterances. For context 12 “Você viu o apartamento novo do Pedro?” (Have you seen Peter’s new apartment?), the percentage of correct identification of the intended context goes up to 77%, while for context 10 “E o outono?” (And autumn?) the maximum is 57%, and for utterance 15 “Onde é que eles moram?” (Where do they live?), the maximum is only 47%.

A production of utterance 10 “E o outono?” (“And autumn?”) by a Russian L1 speaker might be perceived as fitting the wrong context. Quite often, the intended context for utterance 10 (elliptical question) is confused with a yes/no question (see

Table 10. Correctness of perceiving connotations for the target utterance “E o outono?” (“And autumn?”).

Recording	— Diga as quatro estações do ano. — Inverno, primavera, verão. — E o outono? Contrastive Question (Intended)	— Da última vez eu te falei sobre minha estação do ano favorita. — É mesmo, você disse primavera. E o outono? — Sim, primavera e outono. Yes/No Question	— Na prova tinha um texto sobre o inverno e o outono… — E o outono? Foi isso que você disse? Não ouvi direito. Echo Question (Ask to Repeat)	Nenhuma das opções acima
Utt10_Br1	0.48	0.37	0.10	0.05
Utt10_Br2	0.52	0.34	0.08	0.05
Utt10_Ru1	0.53	0.32	0.05	0.08
Utt10_Ru2	0.24	0.17	0.37	0.21
Utt10_Ru3	0.57	0.27	0.05	0.10

). This occurs for both types of speakers of Brazilian Portuguese, L1 and l2. The third option, which is an echo question (asking to repeat), is quite rare.

A different case is Utt10_Ru2. In most cases (37%), it is mistakenly interpreted as an echo question (asking to repeat). This recording differs from the other ones in its curve shape (see Figure 8).

Context 15′s connotations (“Onde é que eles moram?”, “Where do they live?”) are perceived poorly even if produced by speakers of Brazilian Portuguese as L1. Very often, it is confused with a standard information-seeking wh-question (see

Table 11. Correctness of perceiving connotations for the target utterance “Onde é que eles moram?” (Where do they live?).

Speaker	— A Ana e o Miguel moram no Sri Lanka agora. — Onde é que eles moram? O que é que você disse? Echo Question (Intended)	— Onde é que eles moram? — Acho que é no Sri Lanka. WH-Question, Information-Seeking	O que eles estão fazendo? Onde é que eles moram? Eu gostaria de saber. WH-Question, Wondering	Nenhuma das opções acima
Utt15_Br4	0.41	0.49	0.10	0.00
Utt15_Br5	0.47	0.39	0.11	0.03
Utt15_Ru6	0.36	0.49	0.14	0.01
Utt15_Ru7	0.41	0.49	0.06	0.03
Utt15_Ru8	0.07	0.77	0.12	0.04

). It is possible that the context was misinterpreted already at the stage of production. Utt15_Ru8 was misinterpreted in a different way, which is probably because her contour was different. We will go into more detail about this recording in the next section.

3.1.3. Accent

To obtain an impression of how natural the L2 productions sound, i.e., whether they sound accented or not, we analyzed answers for the question ”Can you imagine someone from São Paulo speaking this way?” with ratings from 1 (Definitely not) to 5 (Definitely yes). These ratings were treated as numbers reflecting the “degree of naturalness” of Brazilian Portuguese speech. For the overview of the general trends, we calculated the average values of these ratings for each of the stimuli. Thus, if the average value is close to 5, then the recording was perceived as really natural (such as that which you can hear in São Paulo). On the opposite end, if it is close to 1, then the recording is perceived as unusual for São Paulo.

Some Russophone immigrants may be perceived as speaking like São Paulo residents. Some recordings of Russian L1 speakers were rated as similar to those of Brazilian Portuguese L1 speakers; see e.g., Utt10_Ru1 “E o outono?” (And autumn?), as well as Utt15_Ru7 “Onde é que eles moram?” (“Where do they live?”)” (see

Table 12. Average rating (mean) of naturalness for the target utterance “E o outono?” (And autumn?), utterance 12 “Você viu o apartamento novo do Pedro?” (Have you seen Peter’s new apartment?), utterance 15 “Onde é que eles moram?” (Where do they live?) and utterance 16 “Que brincos lindos você tem!” (What beautiful earrings you have!).

Recording	Average Rating	Standard Deviation
Utt10_Br1	3.98	0.91
Utt10_Br2	3.77	1.03
Utt10_Ru1	3.81	0.95
Utt10_Ru2	3.06	1.15
Utt10_Ru3	2.78	1.26
Utt12_Br3	4.28	0.69
Utt12_Br4	4.20	0.87
Utt12_Ru3	1.92	1.24
Utt12_Ru4	2.31	1.35
Utt12_Ru5	2.75	1.30
Utt15_Br4	4.10	0.81
Utt15_Br5	4.05	0.82
Utt15_Ru6	2.28	1.28
Utt15_Ru7	3.70	1.07
Utt15_Ru8	1.66	1.09
Utt16_Br6	4.05	0.84
Utt16_Br1	4.10	0.93
Utt16_Ru3	2.50	1.28
Utt16_Ru9	3.14	1.24
Utt16_Ru10	2.62	1.36

).

If we go back to Figure 8, we will see that Utt10_Ru1′s curve is very similar to Utt10_Br1 and Utt10_Br2 (Brazilian Portuguese as L1), but Utt10_Ru2 and Utt10_Ru3 have very different patterns. At the same time, Utt10_Ru1 has a very typical Russian curve, but it does not lead to any misunderstanding—every measurement is very close to Brazilian Portuguese L1 speakers. This is a clear case when L1–L2 influence does not have any consequences for communication because L1 and L2 contours used in this type of sentence are similar.

Utt10_Ru3 is a very different example, namely, it has a contour not used in either Russian sentences of this type or Brazilian Portuguese (see Figure 8). And although his intention to ask a question was perceived quite well, the naturalness has suffered significantly. This might be the case for a speaker who tries to speak differently from his first language but has not yet fully mastered the Brazilian Portuguese intonation system.

Productions of some Russian L1 speakers received really low grades on this scale. While average ratings of Brazilian Portuguese L1 speakers ranged between 3.77 and 4.28, some of the Russian L1 speakers’ recordings received particularly low grades (see Table 12); see e.g., Utt15_Ru8 “Onde é que eles moram?” (Where do they live?) or Utt12_Ru3 “Você viu o apartamento novo do Pedro?” (Have you seen Peter’s new apartment?).

We already spoke about Utt12_Ru3 in the previous subsection, whereby the unnaturalness of his recording is due to the absence of the utterance-final melodic peak (see Figure 8 and Table 6). But Utt12_Ru4 also has a relatively low naturalness rating, while the final peak is there. A careful look at Figure 8 enables us to notice that in this recording, every content word received a rising–falling melodic movement, as opposed to other recordings. Such questions, with emphasis on each word, probably sound unusual or “weird” to a Brazilian Portuguese L1 speaker. On the other hand, due to the emphasis on each word, Utt12_Ru4 might have been perceived as produced by someone not fluent in Brazilian Portuguese. Utt12_Ru5, which received quite a low rating as well, used a higher F₀ range, which is typical for Russian L1 speakers, but not for Brazilian Portuguese L1 speakers (Lucente et al. 2023).

According to the naturalness rating, Utt15_Ru8 did not sound like a Brazilian Portuguese utterance. Looking at Figure 9, one may notice that Utt15_Ru8′s curve is very different from all other curves. This is a clear case of L1–L2 influence, as in Russian, such types of questions are produced with a high rise at the beginning followed by a slightly declining level high tone (Volskaya and Kachkovskaia 2016). This melodic pattern does not occur in Brazilian Portuguese at all, which leads to the lowest rating in terms of naturalness.

In utterance 16 “Que brincos lindos você tem!” (What beautiful earrings you have!), all Russian L1 speakers received lower ratings than Brazilian Portuguese L1 speakers. During production, all three Russian L1 speakers emphasized “que” (“what”), while Brazilian Portuguese L1 speakers did not. In Russian, though, emphasizing the wh-word in such sentences is optional. Another difference between L1 and L2 speakers is in the choice of the main accent: for Brazilian Portuguese L1 speakers, it is natural to put most emphasis on the word “lindos” (“beautiful”); all the Russian L1 speakers, on the contrary, put more emphasis on the word “brincos” (“earrings”).

3.1.4. Arousal

To obtain an impression of the L2 speakers’ arousal we analyzed answers to the question ”How strong is the speaker’s emotion/involvement? Rate from 1 (weak) to 5 (strong)”. These ratings were treated as numbers reflecting the “degree of arousal” of Brazilian Portuguese speech. For the overview of the general trends, we calculated the average values of these ratings for each of the stimuli. Thus, if the average value is close to 5, then the speaker was perceived as really involved in the conversation or emotional. On the opposite end, if it is close to 1, then the speakers are perceived as the least involved or emotional.

The involvement ratings differ between target utterances, being slightly higher or the exclamatory sentence (see recordings starting with “Utt16” in

Table 13. Average rating of arousal for all the target utterances 10 “E o outono?” (And autumn?), target utterance 12 “Você viu o apartamento novo do Pedro?” (Have you seen Peter’s new apartment?), target utterance 15 “Onde é que eles moram?” (Where do they live?) and target utterance 16 “Que brincos lindos você tem!” (What beautiful earrings you have!).

Recording	Average Rating	Standard Deviation
Utt10_Br1	3.24	1.08
Utt10_Br2	3.50	1.04
Utt10_Ru1	3.00	1.15
Utt10_Ru2	2.60	1.06
Utt10_Ru3	2.87	1.02
Utt12_Br3	3.14	1.21
Utt12_Br4	3.35	1.07
Utt12_Ru3	3.02	1.01
Utt12_Ru4	3.65	0.91
Utt12_Ru5	3.34	0.93
Utt15_Br4	3.41	1.06
Utt15_Br5	3.49	1.07
Utt15_Ru6	2.55	0.98
Utt15_Ru7	3.49	0.92
Utt15_Ru8	2.32	1.04
Utt16_Br6	3.82	0.87
Utt16_Br1	4.14	0.85
Utt16_Ru3	3.37	1.26
Utt16_Ru9	2.85	1.11
Utt16_Ru10	3.51	1.06

). There is a general trend showing that Russian L1 speakers’ productions in Brazilian Portuguese are perceived as less involved or emotional than those of Brazilian Portuguese L1 speakers. There are exceptions, though, such as Utt12_Ru4, Utt12_Ru5. We already discussed these examples in the previous subsection, and their melodic curves can explain the perceived high involvement or emotion. Thus, in Utt12_Ru4, the speaker emphasized each word of the utterance “Você viu o apartamento novo do Pedro?” (Have you seen Peter’s new apartment?), while in Utt12_Ru5 the speaker used higher F₀ range.

For speakers of many other languages, Brazilian Portuguese sounds more emotional. It is possible that in Utt12_Ru4 and Utt12_Ru5, the speakers were trying to imitate this involvement that they hear so often, and as a result, produced their sentences as too emotional.

3.2. Multi-Factor Analysis of the Data

The application of the MFA statistical method made it possible to analyze the similarity between the utterance productions. The resulting clusters revealed the proximity and distance between the stimuli. Furthermore, MFA analysis helped single out the tonal sequences in the intonation patterns of Brazilian Portuguese that were more difficult to reproduce by the Russian L1 speakers, and the influential factors (sentence type, context, tonal characteristics) that motivated them.

3.2.1. Prosodic Meanings

For the sake of visual clarity, the speakers’ L1s were indicated on the graphs by encircling the Russian L1 speakers’ productions in Brazilian Portuguese in red, and the Brazilian Portuguese L1 speakers’ productions in green.

The MFA analysis arranged the twenty stimuli into seven clusters, as shown in the dendrogram in Figure 10. Those stimuli that are presented within the same cluster share similarities. The inertia gain was very high in two dimensions of the vector space. This means that the distribution of the variables is relevant to be considered in these dimensions (Husson et al. 2011). Then, significant variables in these two dimensions were considered in the analysis.

In some cases, stimuli corresponding to the same L1 and the same utterance appear in the same cluster. This means that they bear similarities in terms of perception by Brazilian listeners, which is reasonable because either they have Portuguese as a first language or all have some features of a Russian accent. There is also one cluster containing only one stimulus, Utt12_Ru3, which is the one that differs significantly from all other stimuli in terms of both perception and melodic curve, as we already discussed in Section 3.1.

We can also see that Utt10_Ru1 appears in the same cluster with two Brazilian Portuguese as L1 stimuli, and Utt15_Ru7 is close to Utt15_Br4 and Utt15_Br5. The results concerning Russophones Ru1 and Ru7 can be interpreted as indicating that these speakers have a performance similar to that of Brazilian Portuguese L1 speakers.

Figure 11 shows the twenty stimuli in the two dimensions (labeled as Dim1 and Dim2) whose inertia gain is very high. The stimuli are distributed in four quadrants. The positions represented by the letters w, x, y and z correspond to the centroids for the three interrogative modalities (w, x and y) and the one exclamatory (z). The centroid of a set of points is the mean point position, that is, it corresponds to the geometrical center of gravity of an area.

In Figure 11, the stimuli recorded from speakers of Brazilian Portuguese as L1 (encircled in green) are all on the right quadrants, while those of Russian L1 speakers are on the left quadrants (encircled in red), except for stimuli Utt10_Ru1, Utt12_Ru4, Utt12_Ru5 and Utt15_Ru7. These stimuli might be considered successful productions as they are similar to those produced by Brazilian Portuguese L1 speakers.

Stimulus Utt12_Ru3 produced by a Russian L1 speaker corresponds to an interrogative utterance that was judged by Brazilian listeners in the perception test as an exclamatory instead of an interrogative. This is why it is located next to stimuli Utt16_Ru9, Utt16_Ru10 and Utt16_Ru3, which are exclamatory utterances produced by Russophones.

Interestingly, stimuli Utt10_Br1, Utt10_Br2, Utt15_Br4 and Utt15_Br5, produced by Brazilian Portuguese L1 speakers, are located very close to each other, even though they correspond to different utterances, namely, Utterance 10 “E o outono?”(And autumn?) and Utterance 15 “Onde é que eles moram?” (Where do they live?). If we look at the normalized curves (Figure 8 and Figure 9), we will see that they all have almost the same F₀ range and alignment, which is probably why they received similar ratings.

Also, utterances 10 and 15 can both be interpreted as echo questions by Brazilian Portuguese listeners; even though they are very different question types in Russian, the Brazilian Portuguese intonation system does not have this distinction.

The distribution of the stimuli in Figure 11 can be explained by the influence of the variables as shown in Figure 12. Variables “Gc1” (judgments of the sentence type), “Gc2” (type of context to explain the perceived connotations), “Gc3” (judgments of arousal) and “Gc4” (accent recognition) pushed stimuli to the right in the horizontal axis, and variable “Gq1” (pragmatic context) to the lower part of the vertical axis.

The strength of the influence of each variable is revealed in Figure 13 through the distance from the point where the variables (Gq1: target utterance; Gc1: sentence type; Gc2: type of context; C3: arousal; and Gc4: accent recognition) are placed in the quadrants to the point where the two 0.00 axes intersect. The more distant, the more influential the variable is, that is, the more representative of the vectorial space. Therefore, Gq1 (target utterance) and Gc2 (type of context to explain connotations) were the most influential variables. The influential variables are the ones with the higher MFA coefficients. (The Gq1 MFA coefficient was 1.67 and the Gc2 MFA coefficient was 0.90. The others had lower MFA coefficient values).

3.2.2. Intonation Patterns

In this section, our focus is on the influence of the initial tone, nuclear tone and final boundary tone. To consider the relation of the intonation patterns with the quantitative and qualitative context variables separately, the annotation of the dynamic tones in the four sentences was included as a variable at the next step of the analysis as described below. Three tones were considered for each of the sentences: initial tone (Gq1), nuclear tone (Gq2) and final boundary tone (Gq3).

The results are plotted in Figure 14, where the strength of the influence is revealed through the distance from the axes’ intersection. Figure 14 shows that the more influential variables were Gq2 (nuclear tone) for utterances 10 and 12; Gq2 (nuclear tone) and Gq3 (final boundary tone) for utterance 15; Gq1 (initial tone) Gq2 (nuclear tone) and Gq3 (final boundary tone) for context 16.

That is, for interrogatives (utterances 10, 12 and 15), the nuclear tone had the most influence on the listeners’ judgments. For utterance 15 “Onde é que eles moram?” (Where do they live?), which is an echo wh-question, final boundary tone was as influential as the nuclear tone; the initial tone also mattered, but to a lesser extent. The exclamation, i.e., utterance 16 “Que brincos lindos você tem!” (What beautiful earrings you have!), differed from interrogatives in this respect, as all three tones were equally influential.

4. Conclusions

In this paper, we have presented new data on cross-language influence in intonation for Russian as L1 and Brazilian Portuguese as L2. Based on the available data on Russian accent in English (Crosby 2013; Kublanova 2003; Poroshina 2022), we hypothesized that in the Brazilian Portuguese speech of Russian L1 speakers, the influence of the Russian intonation system could manifest itself in various meanings of prosody, including illocutionary force, affective meaning and situational context (accent).

To summarize the results of this research, we address the questions formulated in the Introduction.

Do Brazilian Portuguese L2 speakers with Russian L1 always succeed in producing the correct utterance type? In most cases, Russian L1 speakers produced the correct sentence type in their Brazilian Portuguese recordings, that is, questions were perceived as questions, and exclamations as exclamations. However, there were several cases of misinterpretation caused by differences in intonation patterns or pitch alignment. Thus, a yes/no question produced without the utterance-final rise–fall (characteristic of Brazilian Portuguese intonation patterns) was perceived as a statement. This is a clear example of L1 influence. These results are in line with the previous data on the intonation features of the Russian accent. Incorrect identification of sentence type for yes/no questions was also reported for English as L2 (Kublanova 2003; Crosby 2013).

Due to the limited number of research subjects in the present paper, this result is worth considering in further studies with a larger number of research subjects.

In some cases, sentence types were misinterpreted without clear evidence of L1 influence, as the speaker produced a curve not typical for either of the two languages.

Can L1–L2 influence lead to misunderstanding in terms of more distinct shades of meaning? Identification of the context in which the utterance was produced (even by Brazilian Portuguese L1 speakers) was not an easy task for listeners. Quite often, the contexts were not correctly identified when L2 speakers produced unusual melodic curves, and connotations were more often confused. We conclude that L1–L2 language situations may lead to misunderstanding in terms of more distinct shades of meaning in L2 speech, and these can be considered in practical applications for the acquisition of L2.

Is it possible that sometimes L1–L2 influence leads to being perceived as too emotional or, on the contrary, not emotional enough? In general, L2 speakers of Brazilian Portuguese were perceived as less emotionally involved in the conversations than L1 speakers. However, there were a few cases in which Russian L1 speakers were perceived as more emotional, leading to more frequent confusion in connotations. This might be the result of their attempt to sound as emotional as Brazilian Portuguese speakers.

In echo wh-questions where the speaker is asking the interlocutor to repeat what he/she said, sentence type was identified correctly even when the speaker used a Russian melodic contour that never occurs in Brazilian Portuguese. At the same time, this recording received a high rating of accentedness and was judged as less emotional. Interestingly, in the experiment performed by Kublanova (2003), this type of question produced in English with Russian intonation was perceived as sounding surprised, i.e., on the contrary, more emotional. This is not surprising, however, as languages differ regarding the prosodic features used for expressing emotions (Caldwell-Harris 2014). A crucial consequence of this analysis is that correct identification of sentence type does not necessarily guarantee the correct perception of the other components of prosodic meaning.

This phenomenon can be ascribed to distinct prosodic elements. The accurate classification of sentence types may be attributed to alterations in the melodic curve, whereas the perception of emotional engagement appears to be connected to other prosodic components, such as voice quality. An experiment detailed by Yanushevskaya et al. (2018) explored how voice quality and melodic contours are related to affect in Japanese, Spanish, Russian and Irish English. The findings suggested that stimuli incorporating non-modal voice qualities, with or without variation in fundamental frequency, generally prove more effective in conveying affective cues compared to stimuli solely varying in F₀.

Concerning Russophone immigrants in Brazil, can L1–L2 influence in intonation be a significant factor in the perception of accent? In general, the L1–L2 influence was a significant factor. L2 speech was perceived as more accented. In some cases, we found evidence for L1 influence of Russian intonation patterns. However, there were also cases not explicable in terms of L1 influence. Probably, the L2 speakers were trying to find the right pronunciation, but failed, because they had not mastered Brazilian Portuguese intonation patterns yet.

The main distinction in speech production between Brazilian Portuguese L1 and L2 speakers is attributed to pitch alignment, potentially influencing the perception of foreign accents. The DaTo system specifies consistent alignment between stressed vowels and the melodic curve in each melodic contour. Any alterations in this alignment may result in a varied perception of prominence and communicative intentions, as mentioned in Lucente and Barbosa (2010).

Brazilian Portuguese L1 speakers generally exhibit a tendency to delay the alignment between stressed vowels and the melodic peaks or valleys. In contrast, data from Russian L1 speakers indicate an advancement in these alignments at the initial and nuclear tones, as observed in the normalized curves.

In terms of accent, the MFA analysis indicates that the Gc4 variable (accent recognition) was not the most influential factor for prosodic meaning and melody. This outcome may be attributed to the question posed to the listeners, asking them to associate the speaker’s accent with a specific city accent rather than assessing the speaker’s “nativeness”. The selection of this type of question aimed to prevent the association of accent with segmental units but might have led to a more concentrated rating when considering the average ratings and standard deviations (see Table 12).

Another finding was that the overall ratings of some Russian L1 speakers’ recordings were very similar to those of Brazilian Portuguese L1 speakers. This has been clearly and originally demonstrated by Multi Factor Analysis. Given that the target utterances’ selection was based on differences between L1 and L2, these speakers have probably managed to master L2 prosody. Importantly, those speakers (Ru1, Ru7) had been residing in Brazil for no more than 6 years, used Russian at home, and did not attend any language courses.

Several limitations to this research need to be mentioned here. First, it is notable that listeners were often not able to identify the intended connotations even when the target utterances were produced by Brazilian Portuguese L1 speakers. As a result, these data need to be treated with caution, and the methodology should be improved in the future. Another limitation concerns the analysis of arousal. Due to high individual variation in emotion production and perception and given the small number of speakers in the experiment, we should bear in mind that our data on the acquisition of the emotional component only show tendencies that could be explored further in a large-scale experiment. Third, we should be careful in interpreting the results on accentedness because of how the survey question was formulated: “Can you imagine people in São Paulo speaking this way?” While this question has, hopefully, drawn the listeners’ attention away from the idea of foreign accent, we need to keep in mind that some recordings might have been perceived as non-accented in terms of Brazilian Portuguese in general, but still not sounding like the São Paulo accent; due to our experimental design, such recordings were counted as accented. However, because all of the speakers were residents of São Paulo, we might assume that their learning goal was to master this particular accent of Brazilian Portuguese.

Another limitation concerns individual variability in emotional productions due to the factor of a speaker’s gender. It is possible that listeners attribute higher levels of emotional engagement to female speakers due to their greater variability in fundamental frequency and the influence of gender. While the stereotype often associates female voices with heightened emotional expression, recent findings by Lausen and Schacht (2018) suggest that, in the vocal channel, the reliability of emotion judgments is not systematically influenced by a speaker’s gender. Unfortunately, in this research we were unable to support or reject this assertion, as most of the Russian L1 speakers in our dataset were females.

Practical implications of these experimental results include the following. First, our data have demonstrated that for a Russian L1 speaker, mastering the prosody of Brazilian Portuguese is possible, as several such recordings were successful not only in terms of illocutionary force, but also accentedness. Second, we have shown that accentedness on its own does not necessarily imply miscommunication (as also shown in a study presented by van Maastricht (2018)); see e.g., Utt_12, which was correctly perceived as a question despite higher accentedness (while connotations were perceived in a similar way to the stimuli produced by Brazilian Portuguese L1 speakers).

The current findings contribute to the characterization of the L1–L2 influence on intonation for a rarely studied language pair Russian L1–Brazilian Portuguese L2, and shed light to the question how the perception of the immigrants’ speech by local residents is affected by their foreign accent. In general, our results are in line with recently published data obtained in other language pairs: for Czech learners of Spanish and Italian, and German learners of Spanish (Pešková 2023). Most L2 learners could reproduce the illocutionary force but had more difficulties acquiring shades of meaning and emotions. There are also cases of L1 transfer which could lead to misunderstanding (Kublanova 2003; Pešková 2023).