Sage Journals: Discover world-class research

Abstract

This study investigates the role of structural similarity in third language acquisition by using two different semi-artificial languages (ALs, called Aliensk A and Aliensk N) as the targets and either Polish–Norwegian or Norwegian as the previously acquired languages (all speakers also know English). It is an extension of Mitrofanova et al., who did a similar study with one semi-artificial language with Norwegian, Russian–Norwegian, and Greek–Norwegian speakers. Mitrofanova et al. found that previous experience with a language (Russian) which is structurally similar to the target language is facilitative even in instances where the other previously acquired language (Norwegian) is lexically similar to the target language, but this was only the case when the property in question was both structurally and overtly similar (i.e. had similar overt morphological expression). The ALs in the current study are lexically based on Norwegian, a language which does not have morphological case marking. Aliensk A has case marking on articles (structurally but not overtly similar to Polish) and Aliensk N has case marking on nouns (structurally and overtly similar to Polish). Results provide support for the position that previous experience with a language with an overtly and structurally similar property (case) can be facilitative at early stages of acquisition. However, this process may be mediated by an SVO bias, the level of proficiency and activation of the lexically similar language. Overall, the findings support models of L3/Ln acquisition which consider crosslinguistic influence (CLI) to be the result of co-activation of all previously acquired languages and suggest that both lexical and structural similarity play a role at early stages of acquisition.

Keywords

artificial languages co-activation crosslinguistic influence Linguistic Proximity Model morphosyntax similarity in morphological expression structural similarity third language acquisition

I Introduction

In this study, we investigate the role of structural similarity in L3/Ln¹ crosslinguistic influence (CLI) at early stages of acquisition. Structural similarity here refers to the degree of similarity of previously acquired languages and the L3 in the levels of morphology, morphosyntax, and syntax. In the L3 acquisition literature, there are differing views regarding whether L3 learners acquire their new language property-by-property based on co-activation of their previously acquired languages, or whether they make a ‘wholesale’ transfer from one of their previously acquired languages into the new language. The first position is in alignment with the Linguistic Proximity Model (Westergaard, 2021a, 2021b; Westergaard et al., 2017, 2023) and the second with the interlanguage transfer hypothesis (Leung, 2003), later adopted by the Typological Primacy Model (TPM) (Rothman, 2011, 2013, 2015). The Typological Primacy Model proposes that the parser decides which previously acquired language to copy from, referred to as the ‘Big Decision’ (Schwartz and Sprouse, 2021), based on a hierarchy of cues, i.e. Lexicon, Phonology/Phonotactics, Functional Morphology, and Syntactic Structure (in order). Thus, the model predicts that lexical similarity between languages should override structural similarity in terms of the selection of a previously acquired language to transfer from in early L3 acquisition. Lexical similarity refers to the degree of similarity of previously acquired languages and the L3 at the level of the lexicon. For example, if a speaker of first language (L1) Polish and second language (L2) English learns third language (L3) Norwegian, the Typological Primacy Model would predict that CLI would occur only from English into Norwegian due to the closer lexical similarity between the two languages, even if there were structural similarities between Polish and Norwegian in some properties (e.g. reflexive possessives). On the other hand, the Linguistic Proximity Model proposes that both lexical similarity and structural similarity can play a role in CLI, as it argues that CLI can occur from both previously acquired languages. Specifically, it predicts that associative links would be established between similar (lexical as well as grammatical) representations and that they would be co-activated during processing. This can be empirically tested within the Subtractive Language Groups Methodology, which allows us to isolate the effect of individual languages on the target language by comparing otherwise matched participant groups with vs. without a specific previously acquired language (for a more detailed discussion of this method, see Westergaard et al., 2023).

We thus set out to investigate whether it is possible for structural similarity to play a role in CLI between one previously acquired language and an L3 in early stages of acquisition when there is lexical similarity between the other previously acquired language and the L3. The current study is an extension of Mitrofanova et al. (2023). The key questions in their study were:

Can the target language be influenced by both previously learned languages at the early stages of acquisition?

Does structural similarity play a role in CLI, or is this always overridden by lexical similarity?

Are abstract structural similarity (the same property is expressed in a different way, e.g. case marking in both grammatical systems) and overt structural similarity (the same property is expressed in the same way, e.g. case marked postnominally in both grammatical systems) different in terms of CLI?

Abstract structural similarity refers to cases where languages have the same property (e.g. case marking), but it may be expressed differently (e.g. marked on the noun vs. marked on the article). Overt structural similarity refers to languages which have the same property (e.g. case marking), and it is expressed in the same way (e.g. both languages mark case on the noun). Their participants were Russian–Norwegian and Greek–Norwegian speakers. A Norwegian-lexified artificial language (AL) with postnominal case marking, dubbed Aliensk, was used to assess facilitative CLI for speakers of languages with overt structural similarity (Russian – case on nouns, no articles) and structural similarity only (Greek – case on prenominal articles²). Aliensk has relatively free word order, with SVO and OVS constructions available, disambiguated by case marking. After a 10-sentence exposure phase in which participants were exposed to five correct SVO sentences and five correct OVS sentences in the artificial language, they engaged in a Sentence–Picture Matching Task (SPMT). All phrases in the experiment are short noun–verb–noun sentences which are either SVO or OVS, and nouns are marked with nominative (nom) or accusative (acc) case. In the SPMT, participants had to judge whether sentences in the AL matched a picture depicting an agent and a patient. CLI from Norwegian only would lead to an acceptance of the incorrect SVO condition and rejection of the correct OVS condition, as Norwegian is a SVO language without case marking. However, CLI from Russian or Greek would lead to rejection of incorrect SVO and acceptance of correct OVS, due to the presence of case marking in the previously acquired languages. Their findings indicate that overt structural similarity plays a role in CLI alongside lexical similarity, and thus that the target language can be influenced by both previously learned languages at early stages of acquisition. However, while the Russian–Norwegian group benefitted from overt morphological similarity between their L1 and the L3 (case marking on the noun only in both languages), the Greek–Norwegian group did not. The authors attribute this to the prominent role of determiners in disambiguating for case in Greek, especially in nom and acc, and the presence of determiners (which Russian and Aliensk do not have). They thus conclude that overt similarity matters in L3 acquisition.

The current study aims to further examine and test the above questions, with four key points of difference: the language pairing, the use of two ALs, and the length and type of exposure. A different language pairing allows us to assess whether the effect found is limited to Russian–Norwegian and Greek–Norwegian speakers, or whether it can be more broadly generalized. The use of two differing ALs allows us to test whether the same language group will behave in different ways based on abstract or overt similarity, i.e. how case is marked in the new language. Additionally, a slightly longer exposure time and the use of reversible sentences (correct SVO and OVS sentences with the same picture) allowed participants more learning time; this also indicated to participants that the new morphemes do not refer to an unchangeable property (e.g. gender).

We adopt the Subtracted Language Groups Design (Westergaard et al., 2017, 2023) to isolate the role of Polish in the acquisition of the new AL. The language groups assessed are Norwegian–English³ (hereafter: Norwegian) speakers and Polish–English–Norwegian (hereafter: Polish) speakers. Norwegian and English do not have any case marking on nouns or determiners, while Polish marks case by means of a postnominal suffix (as does Russian). Two Norwegian-lexifier ALs are used to investigate performance in the same groups: an AL with case on prenominal determiners (Aliensk A, only abstract structural similarity to Polish), and an AL with postnominal case marking (Aliensk N, overt structural similarity to Polish).

II Background

1 CLI in L3 acquisition

There are several models regarding CLI in L3 acquisition, particularly with regard to the source language(s) of CLI, the question whether one or both previously acquired languages can influence the target language and, indeed, the definition of CLI itself (for an overview, see Rothman et al., 2019). Some studies have posited that CLI in the L3 comes primarily from the L1 (Jin, 2009; Na Ranong and Leung, 2009), whilst others have suggested that it is the L2 which has the greatest influence on the L3, e.g. the L2 Status Factor model (Bardel and Falk, 2007, 2012). This article will focus specifically on models that prioritize linguistic similarity or proximity as the primary determinant of CLI (as opposed to order or manner of acquisition). These models include the Interlanguage Transfer Hypothesis (Leung, 2003) and the Typological Primacy Model (TPM; Rothman, 2011, 2013, 2015), on the one hand, and the Linguistic Proximity Model (LPM; Westergaard, 2021a, 2021b; Westergaard et al., 2017), and the Scalpel Model (Slabakova, 2017) on the other.

These models differ in terms of whether they predict that one or both of the previously acquired languages can influence the target language, which stems from the diverging ways in which CLI is defined. The LPM and the Scalpel model propose that in L3 acquisition, both previously acquired languages are activated during language processing, encompassing interpretation and production (Slabakova, 2017; Westergaard, 2021b; Westergaard et al., 2023). Thus, CLI can originate from either or both previously acquired languages. Furthermore, CLI occurs on a property-by-property basis, stemming from L3 learners’ ability to make fine morphosyntactic distinctions. The primary factor responsible for CLI is structural similarity, though L3 learners may often be influenced by lexical similarity in the early stages of acquisition (Westergaard et al., 2017; Westergaard, 2021a). During processing, co-activated structures from the previously acquired languages engage in competition, with the most strongly activated structure prevailing. This proposal is in line with contemporary processing theories which assume co-activation of similar (phonological, semantic, grammatical) representations in monolingual and bilingual language processing (for an overview, see Blanco-Elorrieta and Caramazza, 2021). As the speaker continues to learn the L3, the initially weaker L3 representations generated during processing gradually stabilize and the learner becomes more proficient in inhibiting CLI from the previously acquired languages. The LPM and the Scalpel Model support the concept of Full Transfer Potential (outlined by Westergaard et al., 2023), which refers to the idea that any structure from a language can potentially influence another language in the mind of a multilingual individual. Both models also suggest that other factors, including recency and frequency, may contribute to the strength of activation of competing structures in the previously acquired languages (see similar suggestions in Blanco-Elorrieta and Caramazza, 2021).

The Interlanguage Transfer Hypothesis and the TPM, in contrast, suggest that a speaker’s initial L3 grammar is a complete copy of one of the previously acquired languages. This is based on the Full Transfer / Full Access Hypothesis (FTFA) of L2 acquisition, which assumes that the L1 is ‘copied’ onto the L2 at the initial state, and that the L2 later undergoes restructuring due to parsing failures with L2 input (Schwartz and Sprouse, 1996). Note that the wholesale transfer hypothesis was couched within and based on the theorizing typical of the generative models of the 1990s and could not take into account rich empirical evidence of co-activation accumulated within processing research of the last three decades. Applying the assumptions of the FTFA hypothesis to L3 acquisition, where there are two possible sources of CLI, the wholesale transfer model suggests that the parser selects the typologically more similar language at the so-called ‘initial stages’, based on a cue hierarchy (lexicon > phonology/phonotactics > functional morphology > syntactic structure). Though the TPM does recognize the possible presence of features from the non-selected language coming through, these are given a different label (Cross Linguistic Effects or CLE) and refer to temporary instances of one language structure bleeding into another (Rothman et al., 2019). This is seen as distinct from ‘representational transfer’, which indicates wholesale copying from only one of the previously acquired languages onto the L3.

As similarity and linguistic proximity are recognized as important factors influencing CLI, it is important to clearly define them. In this article, structural similarity refers to cases where languages have the same abstract linguistic property (e.g. differently encoded structural similarity with different hosts – both languages have the means to express grammatical case), and overt structural similarity refers to instances where languages utilize the same grammatical strategy to mark a linguistic property (e.g. similarly encoded structural similarity with similar hosts – case marked by means of a postnominal suffix).

2 Artificial languages and implicit learning

In the linguistics literature, the broader term of artificial languages can be used to describe two categories in accordance with the characteristics of their lexical inventories: artificial languages (also mini-languages, miniature (artificial) languages) and semi-artificial languages (see an overview in Jensen, 2022). Artificial languages consist of nonce words only, e.g. BROCANTO (Friederici et al., 2002), BROCANTO 2 (Morgan-Short et al., 2010). Semi-artificial languages, on the other hand, combine features of one natural language with features of another, e.g. the lexical items of one language and the morphosyntax of another (González Alonso et al., 2020; Grey, 2020; Jensen and Westergaard, 2023; Mitrofanova et al., 2023). The languages in this study are semi-artificial, as the lexifier language is Norwegian, and the presence of a case-marking system is from Polish (though the actual case markings are nonce suffixes).

Artificial language learning has been found to reliably reflect natural language learning when the AL reflects complexity and meaning (Ettlinger et al., 2016; Friederici et al., 2002; Morgan-Short et al., 2010; Opitz and Friederici, 2003). Jensen (2022) suggests that semi-artificial languages with natural grammars (as opposed to artificial grammars) may be better suited to explore matters pertaining to the acquisition of natural language. Grey (2023) suggests that artificial language systems are useful in investigating questions related to L3 learning. The use of artificial languages allows researchers to control for relevant factors such as typological and structural similarity between the previously acquired languages and the new language, the extent of exposure to the new language, and the type of exposure to the new language (e.g. implicit or explicit) (Grey, 2023: 726).

According to González Alonso et al. (2020: 3), most of the studies using artificial languages prior to their use in L3 studies explored questions related to implicit vs. explicit Ln acquisition. Implicit language learning and implicit language training are defined by the absence of explanations of rules or instructions to attend to forms (DeKeyser, 1995; Norris and Ortega, 2000, 2001). Implicit language learning thus involves the recognition of patterns based on input alone. A good example of current implicit L2/Ln language learning is the popular language learning app Duolingo (including learning of artificial languages such as Esperanto, Klingon and High Valyrian), which uses gamified implicit statistical learning (Freeman et al., 2023). The literature using artificial languages to investigate the roles of structural and lexical similarity in L3 studies have used either implicit language learning (González Alonso et al., 2020; Jensen and Westergaard, 2023; Mitrofanova et al., 2023) or a combination of implicit and explicit language learning (Pereira Soares et al., 2022). In the current study, participants receive implicit language training only to assess whether structural similarity with a previously acquired language can be facilitative for participants who are not given explicit knowledge about the feature in question (for how this relates to models of L3 acquisition, see Section IV).

3 Expression of thematic roles in a transitive sentence

a Polish

Polish expresses thematic relations primarily through case marking. There are seven cases in Polish, marked as postnominal suffixes: nominative, genitive, dative, accusative, instrumental, locative, and vocative, shown for a masculine noun in (1). In addition to appearing postnominally, case is also marked on adjectives, pronouns, and numerals (Łockiewicz and Jaskulska, 2017). Realizations of case suffixes differ for gender and number (singular/plural), as well as whether the base is a noun, pronoun, or adjective.

(1) stół/stółu/stółowi/stół/stółem/stóle/stóle

table.nom/gen/dat/acc/instr/loc/voc

Polish does not have articles. It has relatively free word order, with SVO and subject-initial sentences being the dominant choice in neutral discourse style (2) (Marciszewska, 2021; Peeters-Podgaevskaja et al., 2020). The case-marking system allows for the creation of OVS sentences, in which word order does not match agency, and thus morphosyntactic information is crucial for understanding grammatical roles (3). Word order can also be used to express coreferentiality and uns appearing in positions other than sentence-finally (Szwedek, 1974, 1975).

(2) Kobiet-a widz-i mężczyzn-ę.

woman-NOM see-3SG man-ACC

‘The woman sees a man’ or ‘A woman sees a man’

(3) Mężczyzn-ę widz-i kobiet-a.

Man-ACC see-3SG woman-NOM

‘A woman sees the man.’

OVS word order and other object-initial sentences are used less frequently than SVO (Table 1; see OSF for calculations) (Martens, 2013; Patejuk and Przepiórkowski, 2018; Wróblewska, 2018). Some examples of SOV and OSV sentences from the corpus are provided in (4) and (5). Around half of the OSV sentences (47.8%) and 6.5% of the OVS sentences in the LFG corpus are interrogatives.

(4) To nie znaczy, że mężczy-źni kobiet-y wypychają.

It neg mean that men-nom.pl women-acc.pl push.out

‘This does not mean that men push women out.’

(5) Ten wiersz-Ø Dominika-Ø przykleiła sobie nad biurki-em, a Jadzia dziwiła się.

This poem-acc Dominika.nom stuck refl over desk-instr, and Jadzia surprised refl

‘Dominika stuck this poem over the desk, and Jadzia was surprised.’

Table 1.

Subject and object marking in Polish and Norwegian.

Notes. ^aUniversal Dependencies Polish Lexical Functional Grammar Treebank; ^bUniversal Dependencies Polish Dependency Bank Treebank; ^cUniversal Dependencies Norwegian–Bokmål; ^dUniversal Dependencies Norwegian–Nynorsk.

b Norwegian

With the exception of very few dialects (Eyþórsson et al., 2012), nouns in modern Norwegian do not have case marking. Word order and function words (typically prepositions) are used to express thematic relations. Modern Norwegian has both an indefinite and a definite article. The indefinite article is marked as a free prenominal morpheme, and the definite article is a bound postnominal morpheme (6).

(6) en stol stol-en

ART.INDF chair chair-ART.DEF

‘a chair’ ‘the chair’

Norwegian has basic SVO word order. It is also a verb second (V2) language, requiring that the finite verb appears in second position in the clause, regardless of whether the first position is occupied by the subject. XVSO word order is thus also not uncommon. However, SVO is generally used 70% of the time in both Bokmål and Nynorsk corpora; see Table 1 (Hagen et al., 2000; Kinn et al., 2014; Martens, 2013; Øvrelid and Hohle, 2016; Solberg et al., 2014; Velldal et al., 2017). The proportions in the spoken language are similar; see, for example, Westergaard (2009), Lohndal et al. (2020), and Westergaard et al. (2021).

III Research questions

This study aims to determine whether the findings from Mitrofanova et al. (2023) and ultimately the predictions of the LPM can be supported; in other words, to answer the following research questions:

Research question 1a: At early stages of L3 acquisition, does CLI come exclusively from one previously acquired language, or can the target language be influenced by both/all previously acquired languages? In other words, does CLI into Aliensk come exclusively from Norwegian, or can it come from both Norwegian and Polish?

Research question 1b: Is lexical similarity the key driver of CLI early in the acquisition process, or does structural similarity also play a role, i.e. does lexical similarity between Norwegian and Aliensk override structural similarity between Polish and Aliensk (presence of case marking) in terms of CLI?

Research question 2: In L3 acquisition, should overt similarity (e.g. postnominal case suffixes) and abstract structural similarity (e.g. presence of case marking) be differentiated?

IV Predictions

Research question 1

Before making predictions, it is important to understand one crucial difference between how CLI is viewed by the LPM and the TPM, respectively. As mentioned above, the TPM conceptualizes transfer as representational copying of one of the previously acquired grammars, as a shortcut to the L3 grammar. Meanwhile, the term CLE (crosslanguage effects) is used to indicate (possibly ephemeral) effects of processing, akin to ‘interference’ (Herdina and Jessner, 2002).⁴ In early studies supporting the TPM, it was argued that ‘complete transfer takes place at the earliest moment the parser is able to identify enough linguistic information from the L3 input stream to determine which of the two languages is . . . closer to the target L3’ (Rothman, 2015: 180). More recent studies using online methods in an attempt to find a P600 signature as evidence of a stable (transferred) representation have modified this claim, due to the lack of such findings (González Alonso et al., 2020; Pereira Soares et al., 2022). González Alonso et al. (2020) did find differences in the ERP signal between their participant groups, which the authors speculate may reflect domain-general, puzzle-solving mechanisms that are precursors to wholesale transfer. Given this, it is currently unclear how much time is adequate for transfer to occur. In González Alonso et al. (2020), participants engaged in a pre-training noun familiarization phase, a 45-minute follow-up adjectival training session, and a multiple-choice sentence–picture matching task, wherein they were excluded if they did not reach at least 80% accuracy in the second round. In Pereira Soares et al. (2022), participants were required to do two vocabulary learning phases divided by a 2–10-day consolidation period, a grammar learning phase introducing case morphology, and a multiple-choice sentence–picture matching task wherein they did not proceed if they did not reach at least 80% accuracy in the second round. In both of these studies, the possibility that participants have not had adequate time to make the ‘Big Decision’ is used as an explanation of the somewhat unexpected results.

In contrast, the LPM does not assume representational copying, but rather posits that learning is a result of co-activation in processing (Westergaard et al., 2023). When exposed to the L3, both previously acquired languages are co-activated in searching for potentially useful structures. Of these, the most strongly activated structure prevails. Thus, for the LPM, it is not necessary to copy one previously known grammar at the initial stages of L3 acquisition, and L3 acquisition is seen as a step-by-step process where initially weak representations gradually stabilize with exposure and use.

The predictions of the LPM are clear, since CLI occurs property-by-property from either or both of the previously acquired languages (Slabakova, 2017; Westergaard et al., 2017). Thus, structural similarities or differences between individual properties in the target and the previously acquired languages will also exert an influence and may override the effect of lexical similarity. In other words, learners who know Polish would have an advantage over the group lacking Polish in their repertoire in terms of recognizing case marking in the target language, and these learners are therefore expected to make more accurate judgements in the two critical conditions: incorrect SVO and correct OVS (Table 2; design explained in Section V). Please note that, with regards to the LPM, the terms Accept/Reject do not refer to categorical predictions, especially for the Polish group in the two critical conditions, as the LPM assumes that both previously acquired languages are active in the process of L3 acquisition. Thus, for the Polish group, for the LPM predictions, Accept means Accept more than the Norwegian group (which is predicted to reject in this case), while Reject means Accept less than the Norwegian group (which is predicted to accept in this case).

Table 2.

Experimental conditions and predictions of the Typological Primacy Model (TPM) and Linguistic Proximity Model (LPM).

Picture: A rabbit finding a carrot	Case	Word order	NOR-ENG	POL-ENG-NOR (TPM)	POL-ENG-NOR ALIENSK N (LPM)	POL-ENG-NOR ALIENSK A (LPM)
A. Rabbit-nom finds carrot-acc / nom rabbit finds acc carrot	correct	SVO	Accept	Accept	Accept
B. Rabbit-acc finds carrot-nom / acc rabbit finds nom carrot	incorrect	SVO	Accept	Accept	Reject	Reject?
C. Carrot-acc finds rabbit-nom / acc carrot finds nom rabbit	correct	OVS	Reject	Reject	Accept	Accept?
D. Carrot-nom finds rabbit-acc / nom carrot finds acc rabbit	incorrect	OVS	Reject	Reject	Reject

Assuming (with earlier conceptualizations of the TPM⁵) that the participants in our experiment have had adequate time to recognize the lexical similarity between the L3 and one of the previously acquired languages, the TPM would predict that bilinguals would engage in wholesale transfer from the language which matches the target language on the first level in the hierarchy, i.e. the lexicon (Rothman, 2015). Wholesale transfer would thus be from Norwegian, and there would be no difference between the Polish and Norwegian groups in the two critical conditions. If, however, the participants have not had enough time, the predictions of the TPM are unclear (see Table 2). Thus, without current guidelines as to the duration of the ‘initial stages’, the current study is unable to directly test whether the assumptions of the TPM are borne out in the data.

Mitrofanova et al. (2023) found that Russian–Norwegian speakers showed a clear advantage over Norwegian speakers when acquiring Aliensk, due to CLI from Russian (a language with clearly disambiguating postnominal case marking in nom and acc). The key differences between the groups were in conditions where case marking and word order do not align (rejection of SVO sentences with incorrect case marking, and acceptance of OVS sentences with correct case marking). Based on this finding, we thus predict that the Polish group may have an advantage over the Norwegian group in the two critical conditions, indicating an effect of structural similarity. It is important to note that the presence of an advantage does not indicate that structural similarity will completely override lexical similarity: the LPM does not predict that the Polish group will reject all incorrect SVO sentences and accept all correct OVS sentences, precisely because lexical similarity is predicted to also play a role.

Research question 2

To our knowledge, there is not much literature available on the crosslinguistic effect of abstract vs. overt similarities. Mitrofanova et al. (2023) tested this using one artificial language (Aliensk) and three language groups, Norwegian, Russian–Norwegian, and Greek–Norwegian, assessing whether the presence of abstract structural similarity between Greek (disambiguating case marking on the article) and the artificial language (case marking on the noun only) would be a sufficiently strong cue to facilitate the early acquisition of case by the Norwegian–Greek speakers. This was compared with the overt structural similarity between Russian and Aliensk, which both have case marking on the noun only and no articles. Mitrofanova et al. (2023) showed that speakers of a language with a structural property that is expressed in the same way as in the L3 (Russian) have an advantage over participants with languages without the property (Norwegian) as well as those with languages where the property is only structurally but not overtly similar (Greek).

In the current study, we investigate how speakers of a language with case on the noun only (Polish) behave when case marking in the artificial language is marked on the article (abstract structural similarity, Aliensk A), and compare this with how they behave when case marking is on the noun (overt structural similarity, Aliensk N). In Aliensk N, we predict that case marking from Polish will give the Polish speakers an advantage over Norwegian speakers⁶ in the two critical conditions.

In Aliensk A, however, we expect that the Polish group may not outperform the Norwegian group in the two critical conditions, for several reasons. First, there is abstract structural similarity only between the two languages (Polish and Aliensk A): case marking is present but marked in a different way. Second, Polish does not have articles, and thus the learning task involves the acquisition of a new functional host of case marking.⁷

Additionally, as SVO is a much more frequently occurring construction in all previously acquired languages (to different degrees), we predict that all groups in both Aliensk N and Aliensk A will have more difficulty in the two critical conditions (incorrect SVO and correct OVS).

V Methods

1 Participants

Participants were Norwegian–English bilinguals living in Tromsø, Norway, and Polish–English–Norwegian trilinguals⁸ in Poznań and Szczecin, Poland. These two groups were further split into four, as approximately half of each group took the version of the experiment in which case was marked by means of an article, and the other half took the version in which case was marked by a postnominal suffix. Data collection was conducted partly one-on-one in the lab (n = 53), and partly in classrooms at UiT The Arctic University of Norway (n = 8), Wyższa Szkoła Języków Obcych in Poznan (n = 24), and University of Stettin in Szczecin (n = 32). The experiment was approved by the Norwegian Social Science Data Service (NSD), and the data were collected in accordance with NSD’s ethical principles. Prior to testing, written informed consent was obtained from all participants. We recruited a total of 61 L1 Norwegian participants and 57 L1 Polish participants. Twenty-five Norwegian participants were excluded due to proficiency in languages with case marking (German, North Sámi, Bosnian, Russian, Greek). No exclusions were made on the basis of Norwegian proficiency: all Polish participants had B1 proficiency or above. Table 3 shows the participant metadata for all four groups after exclusions.

Table 3.

Participant metadata.

Participant group	Experiment version	n	Age, mean (range) (years)	Location	Norwegian test score (/28), mean (range)
Polish	Noun	33	21.3 (19–24)	Poznan and Szczeczin	25 (16–28)
Polish	Article	24	21.1 (18–25)	Poznan and Szczeczin	23 (18–28)
Norwegian	Noun	17	35.6 (19–56)	Tromsø	–
Norwegian	Article	22	25.9 (18–38)	Tromsø	–

2 Experiment design

To test the research questions, we used and further developed the sentence–picture verification task created in Mitrofanova et al. (2023). We used the same mini-artificial language, Aliensk, which is lexically based on Norwegian but has case marking on nouns (Aliensk N in this article). Lexical items are existing Norwegian words, with a nominative case marking -il or an accusative case marking -su. In order to test research question 2, we added a second artificial language lexically based on Norwegian but with case marking on articles (Aliensk A). Aliensk A is identical to Aliensk N, except that il and su are articles. Thus, while both artificial languages are lexically similar to Norwegian, Aliensk N is structurally and overtly similar to Polish, and Aliensk A is only structurally similar to Polish.

Participants were given general instructions that they were to first listen to 20 sentences which were correct in the new language they were learning, and then they would take a test where they would need to judge the sentences they heard as to whether they matched the pictures they saw by pressing 9 for YES and 1 for NO. In the online platform, they were then given specific written instructions, followed by a practice sentence, to indicate what the experimental phase would look and sound like. This was followed by an exposure phase, the testing phase, a placement test in Norwegian (adapted from Language Trainers, 2015), and a short background questionnaire. The L1 Norwegian group engaged in the same tasks, except for the placement test in Norwegian. The Norwegian version of the experiment was instructed in Norwegian and English, and the Polish version of the experiment was in Polish and English.

3 Experimental materials and procedure

a Exposure phase

Each participant was randomly assigned to do the experiment with either Aliensk N or Aliensk A. In the exposure phase, we presented the sentences in randomized order. There was an equal number of correct SVO and correct OVS sentences presented to each participant. This is different from the Mitrofanova et al. (2023) study, where participants only heard 10 sentences and did not see them reversed. We decided to include both possible versions of each sentence to indicate that the novel case marking was not some other grammatical category, e.g. gender or definiteness. Participants were informed that this is a learning phase, and that the sentences they heard described the pictures in the artificial language.

For the exposure phase, we created 20 correct sentences and 10 pictures (2 sentences per picture). Ten of the sentences were correct SVO, and 10 were correct OVS. Figures 1 and 2 illustrate four of the sentences presented in the learning phase: one SVO Aliensk N sentence (7), one OVS Aliensk N sentence (8), one SVO Aliensk A sentence (9), and one OVS Aliensk A sentence (10).

Figure 1.

Training item Aliensk N: Sentences (7) and (8).

Figure 2.

Training item Aliensk A: Sentences (9) and (10).

In alignment with Mitrofanova et al. (2023), we used high-frequency nouns and verbs to minimize the effort of memorization in terms of lexical learning and to maximize comprehension by the non-L1-Norwegian group. All sentences in the experiment were three-word NVN sentences, with either SVO or OVS word order.

(7) Skilpadde-il spiser egg-su

Turtle-NOM eats egg-ACC

‘The turtle is eating an egg’

(8) Egg-su spiser skilpadde-il

Egg-ACC eats turtle-NOM

‘The turtle is eating an egg’

(9) Il sebra tegner su sopp.

ART.NOM zebra draws ART.ACC mushroom

‘The zebra is drawing a mushroom’

(10) Su sopp tegner il sebra.

ART.ACC mushroom draws ART.NOM zebra

‘The zebra is drawing a mushroom’

b Test phase

The training session was followed by the test phase, wherein participants had to indicate whether the sentence they heard matched the picture they saw by pressing one of two keys. Each participant was randomly assigned to one of two lists and heard a total of 60 sentences in this phase. Sentences appeared in randomized order with an equal number of grammatical and ungrammatical sentences. Participants had 6 seconds per sentence to select an answer.

A total of 240 sentences were used in this study: 120 for Aliensk N taken directly from the Mitrofanova et al. (2023) and 120 created for Aliensk A, where the case marking occurred on the article; see examples (9) and (10). The sentences describe a total of 30 pictures illustrating simple transitive events (four sentences per picture). The four sentences for each language represent the four experimental conditions. In condition A, sentences are agent-first with SVO word order and correct case marking (nom–acc). In condition B, sentences are agent-first and SVO, but the case marking is incorrect (acc–nom). In condition C, sentences are patient-first and OVS, with correct case marking (acc–nom). In condition D, sentences are patient-first with incorrect case marking (nom–acc). In Conditions A and C (Table 4), the sentence matches the picture, and the correct answer would be to accept these sentences. In Conditions B and D, the correct answer would be to reject the sentences. For each AL, the 120 sentences were split into two 60-sentence lists with an equal distribution of conditions per picture.

Table 4.

Experimental conditions for Aliensk N and Aliensk A (adapted from Mitrofanova et al., 2023).

Language	Picture: A rabbit finding a carrot	Case	Word order
Aliensk N	A. Rabbit-nom finds carrot-acc	correct	SVO
	B. Rabbit-acc finds carrot-nom	incorrect	SVO
	C. Carrot-acc finds rabbit-nom	correct	OVS
	D. Carrot-nom finds rabbit-acc	incorrect	OVS
Aliensk A	A. nom rabbit finds acc carrot	correct	SVO
	B. acc rabbit finds nom rabbit	incorrect	SVO
	C. acc carrot finds nom rabbit	correct	OVS
	D. nom carrot finds acc rabbit	incorrect	OVS

Each of the five verbs were employed in six different scenarios, with various referents (e.g. holder ‘holds’: squirrel/nut, girl / teddy bear, woman/tray, baby/ice-cream, gorilla/apple). The sentences in the test phase had new semantic content for the nouns but retained the same verbs as in the exposure phase (high-frequency verbs spise ‘eat’, holde ‘hold’, oppdage ‘discover’, sparke ‘kick’ and tegne ‘draw’). The pictures from the Mitrofanova et al. (2023) experiment were used and are available on the project OSF. The sentences were recorded by a native speaker of Norwegian at a natural speech rate with neutral intonation. The experiment was conducted using Gorilla, an online psycholinguistics experiment creation platform (Anwyl-Irvine et al., 2020).

VI Results

The accuracy scores⁹ of the participants in the four experimental conditions (correct SVO, incorrect SVO, correct OVS, and incorrect OVS) in the two experimental tasks (case on the noun vs. case on the article) in two participant groups (Norwegian vs. Polish) are summarized in Table 5. The highest scores were obtained for the correct SVO condition, while the correct OVS condition and the incorrect SVO conditions turned out to be the most challenging, as expected. On average, the Norwegian group outperformed the Polish group in Aliensk A, whilst the Polish group outperformed the Norwegian group on the correct OVS condition in Aliensk N. As can also be seen from the individual performance scores (see Figure 3), the distributions of accuracy scores in each condition are bimodal, with some participants having relatively high accuracy and some having low accuracy.

Table 5.

Mean accuracy for each condition for both groups.

		Condition	Accuracy (L1 Polish)		Accuracy (L1 Norwegian)
			Mean	SD	Mean	SD
Artificial language	Aliensk N	SVO correct	0.833	0.374	0.862	0.345
		SVO incorrect	0.367	0.483	0.476	0.500
		OVS correct	0.552	0.498	0.323	0.468
		OVS incorrect	0.588	0.493	0.602	0.490
	Aliensk A	SVO correct	0.813	0.391	0.915	0.280
		SVO incorrect	0.326	0.469	0.489	0.501
		OVS correct	0.446	0.498	0.533	0.500
		OVS incorrect	0.620	0.486	0.790	0.408

Figure 3.

Individual accuracy across the four experimental conditions, two groups (Norwegian vs. Polish) and two experimental versions (case on nouns vs. case on articles).

To assess the differences in the performance of the two groups (Polish vs. Norwegian speakers), we fitted generalized binomial mixed effects logistic regression (glmer) models with the R package lme4 (Bates et al., 2015). In the surviving models, accuracy was predicted as an interaction of condition and group, with random by participant slopes. Including random effects of items rendered singular fit of the models. ANOVA model comparison revealed that the model fits did not improve significantly when random effects of items were included, which motivated omitting the random effects of items from the resulting models to avoid singular fit.

A total of four models were fitted:

one comparing Polish and Norwegian groups in Aliensk A;

one comparing the groups in Aliensk N;

one comparing Aliensk A and N within the Norwegian group; and

one comparing Aliensk A and N within the Polish group.

Correct SVO, Norwegian and Aliensk A were taken as reference levels. We then used the R package emmeans (Lenth et al., 2018) to conduct post-hoc pairwise comparisons with alpha levels adjusted for multiple comparisons. Finally, we ran a glmer model to analyse the outcome of the previously published Norwegian Aliensk study (case on noun) with our new Aliensk N dataset to check whether the additional 10 (reversible) sentences and a different set of speakers would affect the results.

In all five models, the Norwegian participants were significantly more likely to be accurate on correct SVO than on the other three conditions (see Appendices 1– 5 in supplemental material). There is a significant interaction between condition (correct SVO vs. correct OVS) and group (Polish and Norwegian) in the Aliensk N experiment (beta = 3.14, 95% CI [0.53, 5.74], p = .018), driven by the higher accuracy of the Polish group on the correct OVS condition. The model’s total explanatory power is substantial (conditional R² = 0.74), and the part related to the fixed effects alone (marginal R²) is 0.19. In linguistic data, such individual variation is not unusual and to be expected. 95% Confidence Intervals and p-values were computed using a Wald z-distribution approximation. Post-hoc pairwise comparisons confirmed that, in the noun model, the Polish group were significantly more likely than the Norwegian group to be correct in the correct OVS condition (p = .0188). No other effects were significant in Aliensk N.

In Aliensk A, the Polish group performed significantly lower compared to the Norwegian group in the correct SVO condition (beta = −1.38, 95% CI [−2.65, −0.11], p = .033). The model’s explanatory power related to the fixed effects alone (marginal R²) is 0.42. 95% Confidence Intervals and p-values were computed using a Wald z-distribution approximation. Post-hoc pairwise comparisons suggest that the Norwegian group performs significantly more accurately on the correct SVO (p = .0329) and incorrect OVS (p = .0324) conditions. No other effects were significant.

Turning now to the effect of Aliensk N vs. Aliensk A, there were no significant differences between the groups in individual conditions. The model that included data for the Polish participants had a substantial explanatory power (conditional R² = 0.60), and the part related to the fixed effects alone (marginal R²) was 0.17. The Norwegian model’s explanatory power was also substantial (conditional R² = 0.88) and the part related to fixed effects alone (marginal R²) was 0.22. For both models, the 95% Confidence Intervals and p-values were computed using a Wald z-distribution approximation.

Finally, the model examining Norwegian participants’ performance in the previous study (Mitrofanova et al., 2023) in comparison with the current study reveals no statistically significant contrasts between the two groups in post-hoc pairwise comparisons (Figure 4; see Appendix 5 in supplemental material). The model’s total explanatory power was substantial (conditional R² = 0.87) and the part related to the fixed effects alone (marginal R²) was 0.25. This comparison indicates that the additional exposure period (10 extra sentences) and the type of exposure (both SVO and OVS sentences with the same picture) did not contribute to a significantly higher accuracy in any condition. Participants’ behaviour across the two experiments is quite comparable.

Figure 4.

Accuracy scores across conditions and groups (original Norwegian Aliensk participants and current Norwegian Aliensk participants, noun condition).

VII Discussion

In this study, we aimed to expand on the original Aliensk study (Mitrofanova et al., 2023) with a new language combination, the use of two artificial languages, and the length and type of exposure. We investigated whether structural similarity with a previously acquired language (Polish) plays a role even when overall lexical similarity with the other previously acquired language (Norwegian) is present. We assess this in terms of both overt and abstract structural similarity.

1 Aliensk N: Overt and abstract structural similarity

In the current study, the Polish group performs significantly better than the Norwegian group in Aliensk N on one of the two critical conditions: a condition where case marking and word order did not align (correct OVS). This indicates that the participants in the Polish group are able to draw on their previous experience with a case-marking language to assist them in acquiring the same property in a new language. The fact that this occurs despite the strong lexical similarity to Norwegian suggests that speakers can experience facilitative CLI from both previously acquired languages, in alignment with the LPM and the Scalpel model.

However, it is important to recognize here that the Polish group did not perform significantly better than the Norwegian group in Aliensk N on the other critical condition: incorrect SVO. This is unlike the original Aliensk study, wherein there were significant differences between the Russian–Norwegian and Norwegian groups in both critical conditions, correct OVS and incorrect SVO, signifying that the Russian–Norwegian group were better able to recognize case cues in the AL. Although the following speculations pertaining to potential reasons behind the null results should be taken with a grain of salt, we nevertheless suggest some points that, in our view, may be valuable for future research:

There is a difference between the Russian–Norwegian group in the original Aliensk study and the Polish group in the current study regarding their proficiency in Norwegian: The Russian–Norwegian group were given a proficiency test which assessed their level of Norwegian to A2 level on the CEFR. The aim of this test was to make sure the participants had sufficient knowledge of Norwegian to understand the lexical items involved in the artificial language learning experiment. The task did not allow the authors to precisely estimate the participants’ proficiency; however, since some of the participants did not score at ceiling on this task, they roughly estimated the participants’ proficiency as A2. The Polish group, in contrast, were substantially more proficient in Norwegian and had learned it in a university classroom setting (mean score = C1 level, range B1–C2). It is possible that Norwegian is activated more strongly for this group, thus leading to Norwegian ‘winning out’ in the more difficult condition of incorrect SVO (that is, activating a stronger SVO preference).

Incorrect SVO may be more difficult to reject than correct OVS is to accept for several reasons. The first reason is the existence of an SVO bias, for Norwegian and Polish speakers alike, which can be difficult to overcome. Polish allows various word order permutations depending on the information status of structural elements (Mykhaylyk and Sopata, 2015). For all-new utterances, SVO is the neutral word order (Abels and Grabska, 2022). Additionally, SVO word order is much preferred, as indicated in Table 1. Presumably for these reasons, participants in the Polish group generally accepted SVO sentences more than they accepted OVS sentences (Table 6). This is also the case for the Russian participants in the original Aliensk study, though to a lesser extent.

With an expectance that the initial element should be the agent, participants would have to re-interpret their first impression upon hearing accusative case marking. Additionally, the agent and the patient in the majority of the scenarios in the experiment are semantically non-reversible. For correct OVS, the first noun participants hear already does not match the agent in the picture, perhaps triggering greater awareness and attention to the case marking. Additionally, it is possible that incorrect SVO is more difficult to reject than correct OVS is to accept for Polish speakers.¹⁰

One further possibility is the presence of a ‘yes’ bias, where sentences are easier to accept than they are to reject (which would lead participants to perform better on sentences with correct word order than on those with incorrect word order). However, this can be ruled out: Following Huang and Ferreira (2020), we applied Signal Detection Theory to assess the presence of biases and found a slight ‘no’ bias present in the Polish Aliensk N group (c = 0.06, analysis available on OSF).

Table 6.

Acceptance rates across conditions, Polish group Aliensk N.

	Accept	Reject	Total	% Accept
Correct SVO	398	80	478	83
Incorrect SVO	305	177	482	63.3
Correct OVS	267	217	484	55.2
Incorrect OVS	202	288	490	41

2 Aliensk A: Abstract structural similarity only

In Mitrofanova et al. (2023), it was found that overt similarity matters at very early stages of L3 acquisition, thus indicating that abstract structural similarity may not be enough for learners to discover the similarity and make a link between the relevant properties. In the Aliensk A experiment, there is no significant difference between performance on the two critical conditions (incorrect SVO and correct OVS) between the Polish and Norwegian groups (whilst there is a difference in Aliensk N in the correct OVS condition, as mentioned in the previous section). This indicates support for Mitrofanova et al.’s (2023) argumentation that similarity not only on the abstract level but also realized with a morphologically similar exponent is an important factor in predicting CLI at very early stages). We also see a similar result in terms of abstract similarity in Pereira Soares et al. (2022), a study where two groups of bilingual speakers (Italian–German heritage speakers and L1-German–L2-English adult learners) were investigated in terms of their L3/Ln acquisition of a mini-language based on Latin. The properties assessed were case marking and adjective–noun order. The authors did not find evidence of transfer from German into Latin in terms of case marking, despite the fact that both languages have this property (and Italian does not). We posit that this may be due to case marking occurring predominantly on the article in German, whereas it occurs on the noun in Latin – and as the learners are early learners of Mini-Latin, it is too early for abstract structural similarity to have an effect.

Interestingly, however, the Norwegians were significantly more accurate than the Polish group for correct SVO and incorrect OVS here. This can be interpreted as word order cues being more important for the Norwegian group than the Polish group. Specifically, this would predict strong reliance on word order cues for the Norwegian group due to the absence, or very weak representation, of the non-canonical OVS template. On the other hand, the availability of the OVS structure in Polish makes the reliance on word order cues weaker in this group. Thus, higher accuracy for the Norwegian group than the Polish group is expected on the conditions where just following word order cues would result in higher overall accuracy (correct SVO and incorrect OVS), but not on conditions where case cues also play a role (incorrect SVO and correct OVS).

3 Overarching patterns

Although nothing points to this effect in our data, and although wholesale ‘copying’ as the mechanism of additional language acquisition is highly unlikely, in our view, given robust processing evidence (see, for example, Sharwood Smith, 2021, amongst others), our results do not say anything regarding the possibility that wholesale transfer could in principle take place sometime later in the acquisition process (which would go against the LPM and co-activation accounts). In previous publications assessing CLI in artificial languages, unexpected or null results have been taken to indicate that wholesale transfer (creation of a full representational ‘copy’ of one of the previously acquired languages) has not yet occurred, and that the findings thus represent the ‘pre-transfer stages’ (González Alonso et al., 2020; Pereira Soares et al., 2022). If we follow this hypothetical logic, the possibility that wholesale transfer may occur at some undefined later stage cannot be ruled out in the current study either. This possibility would then entail that the Polish speakers in our study and the Russian speakers in the Mitrofanova et al. (2023) study would later become less sensitive to the case cues in the new language, and their accuracy would go down compared to early stages. However, our results (and those of Mitrofanova et al., 2023) indicate that learners are generally quite sensitive to morphosyntax where there is structural and overt similarity. Morphosyntax is at the bottom of the TPM hierarchy (Rothman, 2013, 2015), meaning that it is the last level that will be consulted by the parser for similarity between the L3 and the previously acquired languages. As argued in Jensen and Westergaard (2023), who find that syntax matters also at a very early stage of acquiring an AL, it is unclear how the hierarchy can account for a sensitivity to morphosyntax from one language, when lexical similarity should theoretically cause participants to select the other language as the only source of transfer. A more plausible explanation regarding our results relates to the co-activation approach combined with the strength of other factors, particularly that of word order and SVO bias.

The consideration of such factors is important to understand CLI in a more holistic way, not only in terms of whether it occurs property-by-property but also how this can happen and how it may be dependent on learner- and language-related variables (Slabakova, 2023). A key finding of this study is the strong influence of the language-related variable of word order and word-order based strategies for speakers of all languages involved. SVO is the preferred and most frequent word order in Norwegian and English, as well as Polish (see Section II.3 and Section VII.1). But for the Polish speakers, such a strategy competes with the morphosyntactic cues available from the Polish case system.

The strength of such cues (and their reliability) is measurable with a gradient approach to word order, as argued by Levshina et al. (2023). Despite differences in the word order preferences in Germanic and Slavic languages, Levshina et al. (2023) find in their corpus study that the subject still overwhelmingly precedes the object. For speakers of such subject–object languages, word order appears to be a commonly used processing strategy at the very early stages of L3 acquisition, progressively replaced by case marking (if the L3 has case marking) (Sanz et al., 2015; Stafford et al., 2010).

VIII Conclusions

This article examines the role of lexical and structural similarity between the previously acquired languages and the target language at very early stages of L3/Ln acquisition. It also assesses the difference between overt structural similarity and abstract structural similarity in this process. The target language is one of two versions of an artificial language with Norwegian as the lexifier language, a language without nominal case marking. The structural property to be acquired is case, which is present in Polish. The artificial languages, called Aliensk A and Aliensk N, have case expressed either on an article (unlike Polish) or on the noun itself (as in Polish). The current study involves two groups of participants, Polish–English–Norwegian multilinguals and Norwegian–English speakers (referred to as the Polish and Norwegian groups, respectively). The previously acquired language (Polish) appears to be facilitative in learning the overtly and structurally similar language Aliensk N: statistical analysis revealed a significant effect for one of the two critical conditions (the difference between the Polish and Norwegian groups did not reach significance on the other critical condition). To explain these results, we propose that factors such as proficiency and level of activation of the lexically similar language and a strong SVO bias may also play a role (in addition to the presence/absence of case cues in a previously acquired language), thus explaining the lower performance of the Polish group on the other critical condition. Furthermore, our results also indicate that the Polish participants perform better with the Aliensk N task (when case is morphologically marked on the noun itself, like in Polish). Finally, we find that Norwegian participants rely more on word order cues than Polish participants, who accept non-canonical word orders more readily. This research provides a springboard for further investigation of other factors involved in CLI, including proficiency level, SVO bias, and activation of the previously acquired languages. Future investigations should assess performance of a Polish group in cases where the lexifier language is Polish-like, to analyse whether difficulty rejecting the incorrect SVO condition would be comparable to the current study (with a language lexically based on Norwegian). Furthermore, eye-tracking experiments would be useful to explore the processing patterns associated with case marking and non-canonical word orders.

Supplemental Material

sj-docx-1-slr-10.1177_02676583251332128 – Supplemental material for Investigating crosslinguistic influence (CLI) in L3 morphosyntax through artificial languages

Supplemental material, sj-docx-1-slr-10.1177_02676583251332128 for Investigating crosslinguistic influence (CLI) in L3 morphosyntax through artificial languages by Chloe Michelle Castle, Isabel Nadine Jensen, Natalia Mitrofanova and Marit Westergaard in Second Language Research

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by Norway Funds/NCN project grant GRIEG-1 (UMO-2019/34/H/HS2/00495) ‘Across-domain investigations in multilingualism: Modeling L3 acquisition in diverse settings (ADIM)’.

ORCID iDs

Chloe Michelle Castle

Isabel Nadine Jensen

Natalia Mitrofanova

Marit Westergaard

Data availability

The data, experimental materials and analyses that support this study are openly available at:

Supplemental material

Supplemental material for this article is available online.

Notes

References

Abels

Grabska

(2022) On the distribution of scope ambiguities in Polish. Glossa: A Journal of General Linguistics 7: Article 8170.

Anwyl-Irvine

Massonnié

Flitton

Kirkham

Evershed

(2020) Gorilla in our midst: An online behavioral experiment builder. Behavior Research Methods 52: 388–407.

Bardel

Falk

(2007) The role of the second language in third language acquisition: The case of Germanic syntax. Second Language Research 23: 459–484.

Bardel

Falk

(2012) Procedural distinction. Third Language Acquisition in Adulthood 46: 61–78.

Bates

Maechler

Bolker

Walker

(2015) Fitting linear mixed-effects models using lme4. Journal of Statistical Software 67: 1–48.

Blanco-Elorrieta

Caramazza

(2021) A common selection mechanism at each linguistic level in bilingual and monolingual language production. Cognition 213: 104625.

DeKeyser

(1995) Learning second language grammar rules: An experiment with a miniature linguistic system. Studies in Second Language Acquisition 17: 379–410.

Ettlinger

Morgan-Short

Faretta-Stutenberg

Wong

(2016) The relationship between artificial and second language learning. Cognitive Science 40: 822–47.

Eyþórsson

Johannessen

Laake

Åfarli

(2012) Dative case in Norwegian, Icelandic and Faroese: Preservation and non-preservation. Nordic Journal of Linguistics 35: 219–49.

10.

Freeman

Kittredge

Wilson

Pajak

(2023) The duolingo method for app-based teaching and learning. Duolingo Research Report, 11 January 2023. Available at: https://duolingo-papers.s3.amazonaws.com/reports/duolingo-method-whitepaper.pdf (accessed March 2025).

11.

Friederici

Steinhauer

Pfeifer

(2002) Brain signatures of artificial language processing: Evidence challenging the critical period hypothesis. Proceedings of the National Academy of Sciences 99: 529–34.

12.

González Alonso

Alemán Bañón

DeLuca

, et al (2020) Event related potentials at initial exposure in third language acquisition: Implications from an artificial mini-grammar study. Journal of Neurolinguistics 56: Article 100939.

13.

Grey

(2020) What can artificial languages reveal about morphosyntactic processing in bilinguals? Bilingualism: Language and Cognition 23: 81–86.

14.

Grey

(2023) Using artificial languages to study third language learning and processing. In: Cabrelli

Chaouch-Orozco

González-Alonso

, et al. (eds) The Cambridge handbook of third language acquisition. Cambridge: Cambridge University Press, pp. 724–43.

15.

Hagen

Johannessen

Nøklestad

(2000) A constraint-based tagger for Norwegian. In: Lindberg

C-E

Nordahl Lund

(eds) Proceedings of the 17th Scandinavian Conference in Linguistics. Odense: University of Southern Denmark, pp. 31–48.

16.

Herdina

Jessner

(2022) A dynamic model of multilingualism: Perspectives of change in psycholinguistics. Bristol: Multilingual Matters.

17.

Huang

Ferreira

(2020) The application of signal detection theory to acceptability judgments. Frontiers in Psychology 11: Article 73.

18.

Jensen

(2022) Crosslinguistic influence in third language acquisition. Unpublished PhD thesis, UiT The Arctic University of Norway, Tromsø, Norway.

19.

Jensen

Westergaard

(2023) Syntax matters: Exploring the effect of linguistic similarity in third language acquisition. Language Learning 73: 374–402.

20.

Jin

(2009) Third language acquisition of Norwegian objects: Interlanguage transfer or L1 influence? In: Leung

(ed.) Third language acquisition and Universal Grammar. Bristol: Multilingual Matters, pp. 144–61.

21.

Kinn

Solberg

Eriksen

(2014) NDT guidelines for morphological annotation. Technical Report. Oslo: National Library of Norway.

22.

Language Trainers (2015) Online Norwegian level test. New York: Language Trainers. Available at: https://www.languagetrainers.com/norwegian-level-test.php (accessed March 2025).

23.

Lenth

Singman

Love

Buerkner

Herve

(2018) R package emmeans: Estimated marginal means, aka least-squares means. R Repository. Available at: https://cran.rproject.org/web/packages/emmeans/index.html (accessed March 2025).

24.

Leung

Y-KI

(2003) Failed features versus full transfer full access in the acquisition of a third language: Evidence from tense and agreement. In: J M Liceras (ed) Proceedings of the 6th Generative Approaches to Second Language Acquisition Conference. Somerville, MA: Citeseer, pp. 199–207.

25.

Levshina

Namboodiripad

Allassonnière-Tang

, et al. (2023). Why we need a gradient approach to word order. Linguistics 61: 825–83.

26.

Łockiewicz

Jaskulska

(2017) Polish as L1, English as L2: The linguistic transfer impact on Second language acquisition stemming from the interlingual differences: Implications for young learners’ education. Problemy Wczesnej Edukacji 37: 68–76.

27.

Lohndal

Westergaard

Vangsnes

(2020) Verb second in Norwegian: Variation and acquisition. In: Woods

Wolfe

(eds) Rethinking verb second. Oxford: Oxford University Press, pp. 770–89.

28.

Marciszewska

(2021) L3 morphosyntactic processing among Polish–English bilinguals: Considering learners’ level of bilingualism and language dominance. Open Linguistics 7: 353–71.

29.

Martens

(2013) TüNDRA: A Web application for treebank search and visualization. In: Kübler

Osenova

Volk

(eds) Proceedings of The Twelfth Workshop on Treebanks and Linguistic Theories (TLT12). Sofia: Bulgarian Academy of Sciences, pp. 133–44.

30.

Mitrofanova

Leivada

Westergaard

(2023) Crosslinguistic influence in L3 acquisition: Evidence from artificial language learning. Special Issue: Structural Similarity Across Domains in Third Language Acquisition 13: 717–42.

31.

Morgan-Short

Sanz

Steinhauer

Ullman

(2010) Second language acquisition of gender agreement in explicit and implicit training conditions: An event-related potential study. Language Learning 60: 154–93.

32.

Mykhaylyk

Sopata

(2015) Object pronouns, clitics, and omissions in child Polish and Ukrainian. Applied Psycholinguistics 37: 1051–82.

33.

Na Ranong

Leung

YKI

(2009) Null objects in L1-Thai–L2-English–L3-Chinese: An empiricist take on a theoretical problem. In: Leung

(ed.) Third language acquisition and Universal Grammar. Bristol: Multilingual Matters, pp. 162–91.

34.

Norris

Ortega

(2000) Effectiveness of L2 instruction: A research synthesis and quantitative meta-analysis. Language Learning 50: 417–528.

35.

Norris

Ortega

(2001) Does type of instruction make a difference? Substantive findings from a meta-analytic review. Language Learning 51 (pp. 157–213).

36.

Opitz

Friederici

(2003) Interactions of the hippocampal system and the prefrontal cortex in learning language-like rules. NeuroImage 19: 1730–37.

37.

Øvrelid

Hohle

(2016) Universal dependencies for Norwegian. In: N Calzolari, K Choukri, T Declerck, S Goggi, M Grobelnik, B Maegaard, J Mariani, H Mazo, A Moreno, J Odijk, & S Piperidis (eds) Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16). Portorož: LREC, pp. 1579–1585.

38.

Patejuk

Przepiórkowski

(2018) From lexical functional grammar to enhanced universal dependencies: Linguistically informed treebanks of Polish. Warsaw: Institute of Computer Science, Polish Academy of Sciences. Available at: http://nlp.ipipan.waw.pl/Bib/pat:prz:18:book.pdf (accessed March 2025).

39.

Peeters-Podgaevskaja

Janssen

Baker

(2020) The acquisition of relative clauses in Russian and Polish in monolingual and bilingual children. Linguistic Approaches to Bilingualism 10: 216–48.

40.

Pereira Soares

Kupisch

Rothman

(2022) Testing potential transfer effects in heritage and adult L2 bilinguals acquiring a mini grammar as an additional language: An ERP approach. Brain Sciences 12: Article 669.

41.

Rothman

(2011) L3 syntactic transfer selectivity and typological determinacy: The typological primacy model. Second Language Research 27: 107–27.

42.

Rothman

(2013) Cognitive economy, non-redundancy and typological primacy in L3 acquisition: Evidence from initial stages of L3 Romance. In Baauw

Dirjkonigen

Meroni

Pinto

(eds) Romance languages and linguistic theory 2011. Amsterdam: John Benjamins, pp. 217–47.

43.

Rothman

(2015) Linguistic and cognitive motivations for the typological primacy model (TPM) of third language (L3) transfer: Timing of acquisition and proficiency considered. Bilingualism: Language and Cognition 18: 179–90.

44.

Rothman

González Alonso

Puig-Mayenco

(2019) Third language acquisition and linguistic transfer. Cambridge: Cambridge University Press.

45.

Sanz

Park

Lado

(2015) A functional approach to cross-linguistic influence in ab initio L3 acquisition. Bilingualism: Language and Cognition 18: 236–51.

46.

Schwartz

Sprouse

(1996) L2 cognitive states and the full transfer/full access model. Second Language Research 12: 40–72.

47.

Schwartz

Sprouse

(2021) The full transfer/full access model and L3 cognitive states. Linguistic Approaches to Bilingualism 11: 40–72.

48.

Sharwood Smith

(2021) Language transfer: A useful or pernicious concept? Second Language Research 37: 409–14.

49.

Slabakova

(2017) The scalpel model of third language acquisition. International Journal of Bilingualism 21: 651–65.

50.

Slabakova

(2023) A shared linguistic system of multilingual representations. In: Brown-Bousfield

Flynn

Fernández-Berkes

(eds) Studies in bilingualism: Volume 65. Amsterdam: John Benjamins, pp. 29–48.

51.

Solberg

Skjærholt

Øvrelid

Hagen

Johannessen

(2014) The Norwegian dependency treebank. In: C Nicoletta, C Khalid, D Thierry, L Hrafn, M Bente, M Joseph, M Asuncion, O Jan & P Stelios (Ed.), Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC’14). European Language Resources Association: Paris: LREC, pp. 789–795.

52.

Stafford

Sanz

Bowden

(2010) An experimental study of early L3 development: Age, bilingualism and classroom exposure. International Journal of Multilingualism 7: 162–83.

53.

Szwedek

(1974) Some aspects of definiteness and indefiniteness of nouns in Polish and in English. In: Fisiak

(ed.) Warsaw studies in English language and literature. Lausanne: Peter Lang, pp. 203–11.

54.

Szwedek

(1975) A note on the relation between the article in English and word order in Polish. In: Fisiak

(ed.) Warsaw studies in English language and literature. Lausanne: Peter Lang, pp. 213–25.

55.

Velldal

Øvrelid

Hohle

(2017) Joint UD parsing of Norwegian Bokmål and Nynorsk. In: Tiedemann

(ed.) Proceedings of the 21st Nordic Conference on Computational Linguistics (NoDaLiDa). Linköping: Linköping University Electronic Press, pp. 1–10.

56.

Westergaard

(2009) The acquisition of word order: Micro-cues, information structure and economy. Amsterdam: John Benjamins.

57.

Westergaard

(2021a) Microvariation in multilingual situations: The importance of property-by-property acquisition. Second Language Research 37: 379–407.

58.

Westergaard

(2021b) L3 acquisition and crosslinguistic influence as co-activation. Response to commentaries on the keynote ‘Microvariation in multilingual situations: The importance of property-by-property acquisition.’ Second Language Research 37: 501–18.

59.

Westergaard

Mitrofanova

Mykhaylyk

Rodina

(2017) Crosslinguistic influence in the acquisition of a third language: The linguistic proximity model. International Journal of Bilingualism 21: 666–82.

60.

Westergaard

Mitrofanova

Rodina

Slabakova

(2023) Full Transfer Potential in L3/Ln acquisition: Crosslinguistic influence as a property-by-property process. In: Cabrelli

Chaouch-Orozco González Alonso

, et al. (eds) The Cambridge handbook of third language acquisition and processing. Cambridge: Cambridge University Press, pp. 219–42.

61.

Westergaard

Lohndal

Lundquist

(2021) Variable V2 in Norwegian heritage language: An effect of crosslinguistic influence? Linguistic Approaches to Bilingualism 13: 113–62.

62.

Wróblewska

(2018) Extended and enhanced polish dependency bank in universal dependencies format. In: de Marneffe

M-C

Lynn

Schuster

(eds) Proceedings of the Second Workshop on Universal Dependencies (UDW 2018). Kerrville, TX: Association for Computational Linguistics, pp. 173–82.

63.

Xanthos

Laaha

Gillis

Stephany

Aksu-Koç

Christofidou

Gagarina

Hrzica

Ketrez

Kilani-Schoch

Korecky-Kröll

Kovačević

Laalo

Palmović

B Pfeiler

Voeikova

M. D.

Dressler

W. U.

(2011). On the role of morphological richness in the early development of noun and verb inflection. First Language 31(4), 461–479.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.22 MB