Highlights
-
• Cognates facilitate the acquisition of L2 structures dissimilar to the L1.
-
• Cognates do not facilitate the acquisition of L2 structures similar to the L1.
-
• The cognate facilitation effect in L2 vocabulary acquisition is replicated.
1. Introduction
Bilinguals process cognate words (sharing form and meaning across languages, e.g., guitarra [Spanish] – guitar [English]) more easily than non-cognate words (sharing meaning but not form across languages, e.g., árbol [Spanish] – tree [English]). Specifically, cognates are recognised, read and produced faster and/or more accurately than non-cognates in lexical decision tasks (e.g., Dijkstra et al., Reference Dijkstra, Grainger and van Heuven1999; van Hell & Dijkstra, Reference van Hell and Dijkstra2002), reading tasks (e.g., Van Assche et al., Reference Van Assche, Duyck, Hartsuiker and Diependaele2009, Reference Van Assche, Drieghe, Duyck, Welvaert and Hartsuiker2011) and picture-naming tasks (e.g., Costa et al., Reference Costa, Caramazza and Sebastián-Gallés2000; Gollan et al., Reference Gollan, Fennema-Notestine, Montoya and Jernigan2007). This cognate facilitation effect occurs because cognates activate shared orthography and/or phonology across languages. That activation spreads to a shared meaning representation, which then feeds back its activation to the words’ orthographic and phonological forms. Consequently, the orthographic, phonological and semantic representations of cognates are more activated than those of non-cognates, leading to faster and more accurate processing (Dijkstra & van Heuven, Reference Dijkstra and van Heuven2002).
The stronger activation of cognates can also facilitate syntax processing across proficiency levels. Specifically, processing a sentence with a cognate verb in one of the bilingual’s two languages facilitates the production and comprehension of a sentence with a similar structure in the other language (e.g., Cai et al., Reference Cai, Pickering, Yan and Branigan2011; Chen et al., Reference Chen, Wang and Hartsuiker2023). This facilitation is attributed to the stronger activation of cognates spreading to the combinatorial node containing syntactic information in the bilingual mind, causing it to have a stronger residual activation after processing and to be more available for use in comprehension or production (Hartsuiker et al., Reference Hartsuiker, Pickering and Veltkamp2004; Hartsuiker & Pickering, Reference Hartsuiker and Pickering2008).
This study investigates whether the stronger activation of cognate verbs compared to non-cognate verbs additionally facilitates the acquisition of L2 syntactic structures by novice learners, and whether and how this facilitation is modulated by cross-linguistic syntactic similarity. To the best of our knowledge, only two studies have investigated whether lexical activation facilitates L2 syntax acquisition (Sanahuja & Erdocia, Reference Sanahuja and Erdocia2024, Reference Sanahuja and Erdocian.d.), possibly because lexical and syntactic processing have traditionally been studied separately. The facilitative role of cognates in syntax acquisition has been overlooked in adult and developmental literature. Research on this topic is limited to the study by Sanahuja and Erdocia (Reference Sanahuja and Erdocia2024), which showed that novice L2 adult learners who were explicitly taught an L1–L2 dissimilar grammatical rule were more accurate in applying it to L2 sentences containing a cognate vocabulary than a non-cognate vocabulary. Hence, the question remains regarding whether, apart from facilitating the application of an L2 grammatical rule, the stronger activation of cognates facilitates the acquisition of L2 syntactic structures.
Recent evidence by Sanahuja and Erdocia (Reference Sanahuja and Erdocian.d.) suggests that words differing in activation, such as high- versus low-frequency words, might influence adults’ acquisition of L2 structures. Novice L2 learners were exposed to L1–L2 similar and dissimilar structures with high- and low-frequency verbs. Then, learning was tested in a Grammaticality Judgement Task. Processing an L1–L2 dissimilar structure with high-frequency verbs facilitated its acquisition, as indicated by a greater ability to discriminate between the dissimilar structure and its ungrammatical counterpart when the L2 structure was learnt with high-frequency verbs than when it was learnt with low-frequency verbs.Footnote 1 By contrast, high-frequency words did not facilitate learning of an L1–L2 similar structure, which could be processed using L1 syntax. Discrimination between the similar structure and its ungrammatical equivalent was comparable when the target structure was learnt with high-frequency and low-frequency verbs.
These findings were explained within the Modular Online Growth and Use of Language (MOGUL) framework (Sharwood Smith & Truscott, Reference Sharwood Smith and Truscott2014). In MOGUL, L1 and L2 words and structures have a resting activation level (the baseline activation resulting from processing) and a current activation level (the sum of the resting activation level and the activation received during a processing event). High-frequency words have a higher resting activation level than low-frequency words because they are processed more often, which causes the former also to have a higher current activation level than the latter when activation from the ongoing processing is added.
The finding that differences in verb frequency facilitated the acquisition of an L1–L2 dissimilar structure was explained by claiming that, as the high- or low-frequency verb in a sentence was processed, its activation spread to the structure containing it, increasing its current activation level. When processing finished, the activation of the structure fell towards the resting activation level, landing above the original. The structure had a higher current activation level when processed with high-frequency verbs than with low-frequency verbs; therefore, in the first case, it landed at a higher resting activation level than in the second. The higher the resting activation level of a structure is, the greater its learning capacity is. Hence, the dissimilar structure was better learnt when processed with high-frequency verbs than with low-frequency verbs.
On the other hand, the finding that differences in lexical frequency did not facilitate the acquisition of an L1–L2 similar structure was explained by claiming that the L2 structure was processed using an L1 structure that had a high resting activation level, due to prior processing. If a structure has a high resting activation level, additional processing leads to small or no increases in this level (cf. Gordon & Caramazza, Reference Gordon and Caramazza1985). Thus, Sanahuja and Erdocia argued that the resting activation level of the cross-linguistically similar structure increased slightly more when processed with high-frequency verbs than with low-frequency verbs, but that, overall, the difference in the resting level resulting from processing with the two verbs was imperceptible. Consequently, learning of the structure was comparable with high- and low-frequency verbs.
We argue that the reasoning described should also hold for L2 syntax acquisition with other types of words differing in activation, such as cognates and non-cognates. In this study, we addressed two research questions (RQs):
-
• RQ1: Do cognates facilitate the initial acquisition of cross-linguistically dissimilar L2 structures?
-
• RQ2: If so, does this facilitation vary for the initial acquisition of cross-linguistically similar L2 structures?
Since cognates are more activated than non-cognates, processing L2 structures with cognates should facilitate their acquisition. Yet, this should only occur when the structures are cross-linguistically dissimilar and cannot be processed using similar L1 constructions. The Lexical Bottleneck Hypothesis (Hopp, Reference Hopp2014, Reference Hopp2018) could lead to the same predictions as the MOGUL framework. The hypothesis claims that costly lexical processing may exhaust the resources required to achieve target-like syntactic processing. Thus, it predicts that processing sentences with non-cognates may be so demanding that there are not enough resources left to conduct a target-like processing and acquisition of the cross-linguistically dissimilar structures, resulting in greater L2 syntax learning with cognates than with non-cognates. However, the hypothesis does not address real-time sentence processing or the influence of lexical processing on processing structures that differ in cross-linguistic similarity. In contrast, MOGUL allows for a detailed explanation of the influence of L1 syntax on L2 syntax acquisition, of L2 cognate/non-cognate processing and of the influence of this lexical processing on the real-time acquisition of cross-linguistically similar and dissimilar L2 structures.
In two experiments, we investigated whether cognates facilitated the acquisition of structures that were dissimilar in word order and agent–patient marking between the L1 (Spanish) and the L2 (mini-language based on Basque, Experiment 1), and whether this facilitation varied when the L2 structures were similar to the L1 (Experiment 2).
1.1. Similarities and dissimilarities between the Spanish and Basque lexicon and syntax
Basque has many cognate and non-cognate words with Spanish, with some pairs of cognates/non-cognates expressing the same meaning (e.g., pintatu/margotu, both translating into Spanish pintar, “to paint”). From a structural point of view, Spanish and Basque allow free word order. In spite of this, the canonical order is SVO for Spanish and SOV for Basque (De Rijk, Reference De Rijk1969; López, Reference López1997). Basque is a case-marking, ergative-absolutive language (Laka, Reference Laka1996). It does not morphologically mark the object of transitive verbs for the absolutive case, but the subject of transitive verbs receives the ergative case mark –k. By contrast, Spanish is a nominative–accusative language. It does not overtly mark the subject of transitive verbs for the nominative case, but sometimes marks the object of transitive verbs for the accusative case (Differential Object Marking). When direct objects are animate, specific and refer to particular individuals, they are preceded by the preposition a (“to”) (Fábregas, Reference Fábregas2013).
2. Experiment 1
2.1. Participants
Participants were 60 Spanish natives studying at the University of Barcelona. Thirty were exposed to sentences with cognate verbs (cognate learners), and the other 30 to sentences with non-cognate verbs (non-cognate learners). Cognate learners were aged 18–25 (M = 20, SD = 2.03). Non-cognate learners were aged 18–31 (M = 20, SD = 3.58). There was no significant age difference between groups (t(58) = −0.93, p = .36, d = −0.24). No participants reported knowing Basque or other case-marking, verb-final languages in a linguistic questionnaire. Almost all cognate learners (93%) and non-cognate learners (90%) knew Catalan. However, they all reported feeling more comfortable using Spanish and being spoken to only in Spanish by at least one of their parents before starting school. Participants had normal or corrected-to-normal vision and hearing. Before the experiment, they signed an informed consent. Experiment 1 (and Experiment 2) was approved by the Committee of Ethics for Research Involving Human Beings of the University of the Basque Country (Ref. M10_2019_167). Participants were paid 12€.
2.2. Materials
2.2.1. Cognate and non-cognate vocabulary
We designed a cognate version and a non-cognate version of a mini-language based on Basque. The vocabulary of the two versions consisted of five non-cognate nouns denoting professions (antzezle [Spanish actor] “actor”, epaile [Spanish árbitro] “referee”, sendagile [Spanish médico] “doctor”, margolari [Spanish pintor] “painter” and gidari [Spanish piloto] “pilot”) and, depending on the version, four Spanish–Basque cognate verbs (pintatu [Spanish pintar] “paint”, salutatu [Spanish saludar] “greet”, presentatu [Spanish presentar] “present” and kastigatu [Spanish castigar] “punish”) or the equivalent non-cognate verbs (margotu “paint”, agurtu “greet”, aurkeztu “present” and zigortu “punish”).
To objectively measure the cognateness of verbs, we calculated the orthographic and phonological Normalised Levenshtein Distance (NLD) between the Basque verbs and their Spanish translations (orthographic NLD, M = 0.30, SD = 0.08; phonological NLD, M = 0.27, SD = 0.08). The NLD ranges from 0 (identical words) to 1 (completely different words). The measures obtained confirmed that the verbs were very similar Spanish–Basque cognates. Additionally, we calculated the orthographic and phonological NLD between the non-cognate Basque verbs and their Spanish translations (both M = 0.87, SD = 0.02). The cognate verbs were more similar in form to their Spanish translations than the non-cognate verbs were. The two types of verbs were similar in length (in letters: cognate verbs, M = 8.5, SD = 1.29; non-cognate verbs, M = 7.0, SD = 0.82).
A male native speaker of Basque recorded each noun and verb. Then, these were associated with a picture, bought from 123RF Image Bank (https://www.123rf.com/) (Supplementary Appendix S1). These word–picture pairs were used to help participants learn the vocabulary of the mini-language and to generate sentence–picture pairs for the exposure phase, during which participants learnt the L2 structures (see Section 2.3).
2.2.2. Exposure materials
Cross-linguistically dissimilar structures. The structures of the mini-language were SOV and OSV constructions with postpositional agent–patient marking based on Basque grammar. Table 1 illustrates these structures. It contains four sentences expressing the same meaning and paired with the same picture. The four consist of the same non-cognate arguments (antzezleak “the actor” and gidaria “the pilot”) followed by a verb. Two sentences belong to the cognate language version and include a cognate verb (pintatu). The other two sentences belong to the non-cognate language version and include the equivalent non-cognate verb (margotu). In SOV sentences, the first noun is the agent (A), and the second noun is the patient (P) of the verb, as marked by –ak and –a, respectively. In OSV sentences, the agent and patient marking are the same, but the order of the nouns is reversed.
Table 1. Examples of SOV and OSV sentence–picture pairs for the cognate and non-cognate versions of the language in Experiment 1
Note: The four sentences have the same meaning, and thus, they are paired with the same picture.
These SOV and OSV structures differ from Spanish in word order and agent–patient marking. As mentioned, in Spanish, the canonical word order is SVO, and although SOV and OSV orders are possible, these are extremely infrequent (ADESSE corpus; García-Miguel et al., Reference García-Miguel, Vaamonde, González Domínguez, Calzolari, Choukri, Maegaard, Mariani, Odijk, Piperidis, Rosner and Tapias2010). Additionally, Spanish does not overtly mark the subject/agent of transitive verbs and marks animate, specific direct objects with the preposition a.
Sentence–picture pairs. We generated 80 SOV sentences with postpositional agent–patient marking and with a cognate verb. Then, we generated three additional versions of each sentence, manipulating word order (SOV vs. OSV) and verb type (cognate vs. non-cognate). The Basque speaker who recorded the vocabulary also recorded the sentences for use in the exposure phase. The four versions of each sentence were paired with the same picture representing their meaning (Supplementary Appendix S2). The 80 SOV and 80 OSV sentence–picture pairs containing a cognate verb were allocated to the cognate language version, and the 80 SOV and 80 OSV sentence–picture pairs containing a non-cognate verb were allocated to the non-cognate language version. The sentence–picture pairs for each version were divided into two lists to prevent participants from hearing and seeing an SOV sentence and its OSV equivalent. Each participant was exposed to 40 SOV and 40 OSV sentence–picture pairs. Each noun occurred twice as the agent and twice as the patient of a given verb in SOV and OSV sentences per list.
2.2.3. Testing materials
In the testing phase, the two groups of participants were tested with four new non-cognate verbs (aukeratu “choose”, aztertu “examine”, zelatatu “spy on” and gainditu “surpass”), each recorded and associated with a picture from 123RF Image Bank (Supplementary Appendix S1). These word–picture pairs were used in a second vocabulary-learning phase preceding the testing phase (see Section 2.3). As testing materials, we generated all possible sentences using the vocabulary of the test (the five nouns in the exposure materials and the four novel non-cognate verbs), resulting in 80 SOV sentences and 80 OSV sentences derived from the SOV ones. Then, we selected eight pairs of SOV and OSV sentences and associated them with a picture that represented their meaning. These pictures (Supplementary Appendix S2) were used in the test, which was a picture-description task. Each of the four verbs appeared in two pictures, and each of the five nouns fulfilled the role of agent and patient in at least one picture. Two characters never appeared together in more than one picture. The two groups of learners saw the same pictures.
2.3. Procedure
The experiment started with a vocabulary-learning phase, during which participants were told that they would learn some words in Basque and then do some sentence comprehension tasks. They were not informed that this was a syntax learning experiment. Cognate and non-cognate learners learnt the non-cognate nouns and the cognate or non-cognate verbs that were going to be used in the exposure phase associated with their pictures. Then, they were exposed to the SOV and OSV structures with postpositional agent–patient marking via sentence–picture pairs. Cognate learners were exposed to sentences with cognate verbs, and non-cognate learners to sentences with non-cognate verbs. Next, the two groups learnt the same novel non-cognate verbs and were tested on their learning of the structures in a written picture-description task with these verbs.
Cognate and non-cognate learners additionally performed the Spanish version of Unsworth et al.’s (Reference Unsworth, Heitz, Schrock and Engle2005) reading span task to control for working memory. The groups had comparable memory capacities (cognate learners’ mean partial reading span score: 48.03, SD = 10.35; non-cognate learners’ mean score: 48.30, SD = 10.08; t(58) = −0.10, p = .92, d = −0.03). Testing was done individually in a soundproof booth, monitored by the experimenter from an adjacent room. The experiment was run on the E-Prime 3.0 software (Psychology Software Tools, Inc., 2016) and lasted for around 90 minutes.
2.3.1. First vocabulary-learning phase
Each trial began with a picture representing a noun or a verb in the middle of the screen. To ensure that learners understood that the picture depicted the target word/action, the Basque word associated with the picture was written below, with its Spanish translation in brackets, and it was simultaneously played (Figure 1).
Figure 1. Example of a vocabulary learning trial. The picture represents the Basque noun gidari (“pilot”), which was presented visually and aurally. The Spanish translation “(piloto)” appeared below the noun.
Participants repeated the word aloud and pressed the space bar, by which a new trial began. Each word–picture pair was presented four times. The order of the pictures was randomised, but the nouns were always presented before the verbs. Learning was assessed in a picture–word matching task and a picture-naming task performed until participants reached 100% accuracy.
In the picture–word matching task, participants saw the picture of a noun or a verb and a list of the five nouns or the four verbs learnt written below. They had to select the word that described the picture as quickly as possible using the keyboard (keys 1–5). There was no time limit to respond. After selecting a word, feedback was provided (a green tick or a red cross). The list of nouns or verbs appeared in a different order in each trial. Pictures of nouns appeared interspersed with pictures of verbs. If participants made a mistake, they had to repeat the task. The presentation of the pictures was randomised so that, if participants performed the task more than once, they did not see the pictures in the same order.
In the picture-naming task, participants named pictures of nouns and verbs shown one at a time, as quickly and accurately as possible. Nouns had to be named before verbs. After naming a picture, participants pressed the space bar to see and hear the correct response. If no answer was given within 10 seconds, the correct response automatically appeared written below the picture. The computer recorded participants’ responses. If participants had difficulty naming the pictures, the experimenter reviewed the vocabulary with them, and they repeated the task.
2.3.2. Exposure phase
Participants were informed that they would see pictures involving some of the nouns and verbs learnt, each visually and aurally accompanied by a sentence describing it. They were instructed to look at each picture and listen to and silently read the sentence accompanying it. Cognate learners were exposed to SOV and OSV sentences with postpositional agent–patient marking and with cognate verbs; non-cognate learners were exposed to the same sentences but with non-cognate verbs. Learners were not informed that they would be exposed to two structures, nor were they instructed about the agent–patient marking in the mini-language. Each picture appeared in the middle of the screen. The sentence that described it was written below, and it was simultaneously played (Figure 2). Basque has a transparent written language and shares almost all phonology with Spanish, which facilitates that Spanish native speakers read sentences in Basque even if they do not know the language.
Figure 2. Example of an exposure trial for the non-cognate language version. Participants saw a picture while reading and listening to a sentence describing it. The figure displays an SOV sentence–picture pair representing “The actor is painting the pilot” with the non-cognate verb margotu (“to paint”).
Each sentence–picture pair remained on the screen for 500 milliseconds after the audio of the sentence ended. Then, a new pair was presented. In half of the pictures, the agent appeared on the right, and in the other half, on the left, with no contingency with the type of sentence structure. The presentation of sentence–picture pairs was randomised.
2.3.3. Second vocabulary-learning phase
Cognate and non-cognate learners learnt the same four novel non-cognate verbs that were going to be used in the testing phase to test the two groups’ syntax learning with the same materials (cf. Section 2.2). Verbs were learnt and tested as in the first vocabulary-learning phase.
2.3.4. Testing phase
Participants were told that they would see pictures similar to the ones in the exposure phase and that they had to write in a textbox a sentence that described each picture. The five possible (bare) nouns and the four possible verbs were listed to the right of the picture (Figure 3).
Figure 3. Example of a trial in the picture–description task. Participants saw a picture and had to describe it choosing the appropriate nouns and verb from the list and using one of the structures learnt. In this figure, possible descriptions were Antzezleak gidaria aztertu (SOV) or Gidaria antzezleak aztertu (OSV) (“The actor is examining the pilot”).
Cognate and non-cognate learners saw the same eight pictures randomised. The appearance of the agent on the right or the left of the picture was counterbalanced. When participants finished writing a sentence, or if they did not know how to describe a picture, they pressed ENTER, and a new trial began.
2.4. Predictions
Our main hypothesis for RQ1 was that cognates would facilitate the acquisition of the cross-linguistically dissimilar L2 structures. A secondary hypothesis was that cognate and non-cognate learners would not learn the SOV or the OSV structure with postpositional agent–patient marking significantly better than the other one, since the two structures were dissimilar to L1 grammar and had to be learnt from scratch. A prerequisite for the hypotheses to be met was that cognate and non-cognate learners learnt the SOV and the OSV structures. If the structures were learnt, in the picture-description task, learners would correctly write significantly more than 50% of SOV and OSV sentences with agent–patient marking.
If our main hypothesis was correct, we predicted (P1) that cognate learners would be significantly more accurate in their verb-final picture descriptions with agent–patient marking than non-cognate learners would. If our secondary hypothesis was correct, we predicted (P2) that the two groups of learners would not be significantly more accurate in their SOV picture descriptions with agent–patient marking than in their OSV picture descriptions with agent–patient marking.
2.5. Data pre-processing and analysis
Experiment 1 (and Experiment 2) was analysed using R (R Core Team, 2024, version 4.4.1). All analyses included the 30 cognate learners and the 30 non-cognate learners. We assumed p < .05 as alpha for significance testing.
2.5.1. First and second vocabulary-learning phases
Shapiro–Wilk tests calculated with the stats package (R Core Team, 2024) indicated that the number of attempts at picture–noun matching, picture–verb matching, noun picture-naming and verb picture-naming were not normally distributed, neither for cognate nor for non-cognate learners. Therefore, we compared cognate and non-cognate learners’ attempts at these tasks using Wilcoxon rank-sum tests computed with the stats package. We calculated the standardised measure of effect size r for these tests using the rstatix package (Kassambara, Reference Kassambara2021).
2.5.2. Testing phase
First, we evaluated whether the SOV and OSV structures with agent–patient marking were learnt. The testing phase included 480 trials, with 240 trials per group of learners (30 learners × 8 test items each). Two trials (one from a cognate learner and one from a non-cognate learner) were removed: one trial with no response and one trial using the same noun as the subject and object of the sentence. Next, we coded learners’ productions as SOV, SVO, OSV or OVS, looking at the order of the nouns and the verb in the sentence (irrespective of whether agent–patient marking was used). Only non-cognate learners wrote verb-medial sentences (31 SVO, 7 OVS out of 239). Taking the subset of verb-final picture descriptions, we calculated the proportion of SOV and OSV sentences by cognate and non-cognate learners, which were compared with a chi-square test fitted with the stats package.
Next, we coded agent–patient marking use in these picture descriptions (1 = sentence with marking, 0 = sentence without marking). To assess whether learners used agent–patient marking in significantly more than 50% of SOV and OSV sentences, we ran generalised linear mixed-effects models that set the intercept to zero for each group of learners separately. The models were fitted with the lme4 package (Bates et al., Reference Bates, Mächler, Bolker and Walker2015). The models with a by-item random intercept caused convergence issues, so the final models had the structure: glmer (Agent–patient marking ~ −1 + Word order + (1 | Participant)).
Finally, we analysed only the picture descriptions that included agent–patient marking, coding its use as correct (1) or incorrect (0). Generalised linear mixed-effects models assessed whether the two groups of learners used agent–patient marking as required in significantly more than 50% of SOV and OSV sentences. For cognate learners, the model with a by-item random intercept caused convergence issues, so the final model had the structure: glmer (Accuracy ~ −1 + Word order + (1 | Participant)). Footnote 2 For non-cognate learners, the model was as follows: glmer (Accuracy ~ −1 + Word order + (1 | Participant) + (1 | Item)). Word order was treatment-coded (OSV = 0, SOV = 1) in all models.
To conclude, we compared learning of the structures by cognate and non-cognate learners, that is, whether learners’ accuracy in writing SOV and OSV sentences with agent–patient marking differed significantly. A generalised linear mixed-effects model tested for an interaction between Group (Cognate = 0.5, Non-cognate = −0.5) and Word order of the picture description (SOV = 0.5, OSV = −0.5) on accuracy. The model only had random intercepts by participant and item, since including random slopes caused convergence issues.
2.6. Results and discussion
2.6.1. First vocabulary-learning phase
Picture–word matching task. The number of attempts at picture–noun matching was comparable for non-cognate learners (M = 1.77, SD = 0.97; Median = 2, Median Absolute Deviation (MAD) = 1) and for cognate learners (M = 1.43, SD = 0.77; Median = 1, MAD = 0; W = 349.5, p = .09, r = .22). Likewise, non-cognate learners’ number of attempts to match all non-cognate verbs with their pictures (M = 2.00, SD = 1.20; Median = 2, MAD = 1) was not statistically different from cognate learners’ number of attempts to match all cognate verbs with their pictures (M = 1.43, SD = 0.57; Median = 1, MAD = 0; W = 33.5, p = .07, r = .23).
Picture-naming task. The number of attempts to name all pictures of nouns correctly was not significantly larger for non-cognate learners (M = 2.07, SD = 0.74; Median = 2, MAD = 1) than for cognate learners (M = 1.97, SD = 0.81; Median = 2, MAD = 1; W = 410.5, p = .54, r = .08). In contrast, the mean and the median of the number of attempts to name all pictures of verbs correctly were significantly larger for non-cognate learners (M = 1.97, SD = 0.76; Median = 2, MAD = 1) than for cognate learners (M = 1.17, SD = 0.38; Median = 1, MAD = 0; W = 190, p < .001, r = .56).
In sum, both groups learned non-cognate nouns in a comparable number of attempts, but cognate verbs were learnt faster than non-cognate verbs, as evidenced in the picture-naming task. This cognate facilitation effect in L2 word learning supports prior findings with adults and children (e.g., De Groot & Keijzer, Reference De Groot and Keijzer2000; Tonzar et al., Reference Tonzar, Lotto and Job2009). L2 words were learnt with their L1 translations. We argue that due to the cross-linguistic activation of formally similar words (Dijkstra et al., Reference Dijkstra, Grainger and van Heuven1999), the L1 verbs were more strongly activated when they were cognate with their L2 equivalents than when they were non-cognate. We propose that this made it easier to link the L2 cognate verbs to their L1 representations, whereas the connection between the L2 non-cognate verbs and their L1 counterparts was weaker or more difficult to establish. However, by the end of the learning phase, cognate and non-cognate learners achieved 100% accuracy in the picture–word matching task and in the picture-naming task, which suggests that, ultimately, cognate and non-cognate verbs had similar links to their L1 representations.
2.6.2. Second vocabulary-learning phase
Picture–word matching task. Both groups attempted the task a comparable number of times (cognate learners: M = 1.37, SD = 0.72; Median = 1, MAD = 0; non-cognate learners: M = 1.37, SD = 0.61; Median = 1, MAD = 0; W = 431, p = .72, r = .05).
Picture-naming task. The number of attempts at picture naming by cognate learners (M = 1.23, SD = 0.43; Median = 1, MAD = 0) and non-cognate learners (M = 1.10, SD = 0.31; Median = 1, MAD = 0) were comparable (W = 510, p = .17, r = .18).
In sum, in the second vocabulary-learning phase, cognate and non-cognate learners learnt the same non-cognate verbs comparably fast. This would suggest that the learning difference observed in the first vocabulary-learning phase between cognate and non-cognate verbs was due to cognateness, not to individual differences between the groups of learners.
2.6.3. Testing phase
Cognate learners wrote 66.11% of SOV picture descriptions and 33.89% of OSV picture descriptions. Similarly, non-cognate learners wrote 60.20% of SOV picture descriptions and 39.80% of OSV picture descriptions. Hence, participants learnt that the mini-language had two word orders. The proportion of SOV and OSV sentences was similar across groups (X 2 (1) = 1.40, p = .24).
Additionally, cognate learners used agent–patient marking in 63.92% of SOV sentences (significance from 50%: β = 10.10, SE = 2.06, z = 4.91, p < .001) and in 80.25% of OSV sentences (significance from 50%: β = 11.50, SE = 2.77, z = 4.16, p < .001). Non-cognate learners used agent–patient marking in 62.81% of SOV sentences (significance from 50%: β = 11.04, SE = 2.37, z = 4.66, p < .001) and in 66.25% of OSV sentences (significance from 50%: β = 10.46, SE = 2.23, z = 4.69, p < .001). This shows that both groups learnt that the nouns in the sentence required marking. Although descriptively cognate learners wrote a much higher percentage of OSV sentences with agent–patient marking than non-cognate learners, the difference is not significant (X 2 (1) = 3.34, p = .07).
Finally, cognate learners used agent–patient marking correctly in 85.15% of SOV sentences (significance from 50%: β = 8.35, SE = 2.21, z = 3.79, p < .001) and in 86.15% of OSV sentences (significance from 50%: β = 8.45, SE = 2.26, z = 3.74, p < .001). Non-cognate learners used agent–patient marking correctly in 68.42% of SOV sentences (significance from 50%: β = 1.82, SE = 0.89, z = 2.05, p < .05) and in 62.26% of OSV sentences (significance from 50%: β = 2.21, SE = 1.08, z = 2.05, p < .05). This suggests that the target structures were learnt.
A mixed-effects model tested whether cognate and non-cognate learners were comparably accurate when writing the cross-linguistically dissimilar L2 structures (SOV and OSV sentences with agent–patient marking) (Table 2 and Figure 4). There was an effect of Group, no effect of Word order and no interaction between the two. The absence of an interaction indicates that there was no evidence for a difference in the accuracy with which cognate and non-cognate learners produced SOV versus OSV sentences with agent–patient marking. Additionally, the lack of an effect of Word order suggests that, overall, there was no evidence that learners were significantly more accurate when writing one or the other types of sentences. Therefore, as predicted (P2), cognate and non-cognate learners wrote a comparable number of SOV and OSV sentences with correct agent–patient marking. This suggests that the two cross-linguistically dissimilar structures were comparably learnt and that this was the case for the two groups of learners, supporting our secondary hypothesis for RQ1.
Table 2. Summary of the generalised linear mixed-effects regression model fitted for the test in Experiment 1
Note: Model structure: glmer (Accuracy ~ Group × Word_order + (1 | Participant) + (1 | Item), data = test_Exp1, family = binomial).
Figure 4. Percentage of correct picture descriptions with SOV or OSV word order and agent–patient marking produced by cognate and non-cognate learners in Experiment 1. Error bars represent 95% confidence intervals.
For cognate learners, the mean percentage of correct picture descriptions with verb-final word order (collapsing SOV and OSV) and agent–patient marking was 85.54% (SD = 35.27%). For non-cognate learners, the mean percentage was 65.89% (SD = 47.59%). The Group effect reveals that, as predicted (P1), cognate learners were significantly more accurate when writing verb-final picture descriptions with agent–patient marking than non-cognate learners. This supports the idea that the target structures were better learnt when processed with cognates than with non-cognates, in line with our main hypothesis for RQ1.
Based on MOGUL (Sharwood Smith & Truscott, Reference Sharwood Smith and Truscott2014), we argue that when learners were exposed to the L2 structures during the exposure phase, two novel syntactic representations were created and temporarily stored in the linguistic system with a low resting activation level. Repeated processing of the structures raised their resting activation level, and they were gradually acquired (they became firmly established in the linguistic system). Crucially, the stronger activation of cognates compared to non-cognates (Dijkstra & van Heuven, Reference Dijkstra and van Heuven2002) spread to the structures containing them. This caused the structures processed with cognates to be more strongly activated and to have a higher resting activation level after processing than the structures processed with non-cognates. The higher the resting activation level of a structure, the better it is learnt. Hence, learning of the dissimilar structures was greater when processed with cognates than with non-cognates. To investigate whether the facilitative role of cognates varied when the structures to be learnt were similar to the L1 (RQ2), we conducted Experiment 2.
3. Experiment 2
3.1. Participants
Sixty Spanish natives, different from participants in Experiment 1, were recruited from the University of Barcelona. Thirty were assigned to the cognate learning group (aged 19–33, M = 21, SD = 3.25) and 30 to the non-cognate learning group (aged 19–27, M = 21, SD = 1.98). The groups were matched in age (t(58) = 0.67, p = .50, d = 0.17) and had the same linguistic profile as in Experiment 1. All participants had normal or corrected-to-normal vision and hearing. Before the experiment, they signed an informed consent. Participation was rewarded with 12€.
3.2. Materials
The cognate and non-cognate materials of the mini-language in Experiment 1 were used in Experiment 2, with the exception of the cross-linguistically similar structures created for this experiment.
3.2.1. Exposure materials
Cross-linguistically similar structures. The structures were SVO and OVS constructions with prepositional patient marking, as illustrated in Table 3. The table displays four sentences expressing the same meaning with the same picture. Two sentences belong to the cognate language version and include a cognate verb (pintatu). The other two sentences belong to the non-cognate language version and include the corresponding non-cognate verb (margotu). The structures of the mini-language are similar to Spanish syntactically and morphologically. In both languages, the SVO structure starts with an animate agent (A) noun without overt agent marking, followed by a verb, the preposition a marking the patient (P) and an animate patient noun. In OVS sentences, the agent and patient order is reversed. Unlike Spanish, where left-dislocated objects use a co-referential clitic pronoun before the verb, the mini-language omitted the clitic to keep patient marking consistent regardless of word order.
Table 3. Examples of SVO and OVS sentence–picture pairs for the cognate and non-cognate versions of the language in Experiment 2
Note: The four sentences have the same meaning, and thus, they are paired with the same picture.
Sentence–picture pairs. We transformed Experiment 1’s SOV and OSV sentences with postpositional agent–patient marking into SVO and OVS sentences with prepositional patient marking. The Basque native speaker who recorded the stimuli in the first experiment recorded the stimuli in the second experiment. The sentence–picture pairs in Experiment 1 were used in Experiment 2 (Supplementary Appendix S3). The 80 SVO and 80 OVS sentence–picture pairs with a cognate verb were assigned to the cognate language version, and the 80 SVO and 80 OVS sentence–picture pairs with a non-cognate verb were assigned to the non-cognate language version. The sentence–picture pairs for the two versions were distributed into two lists, so that participants would not hear and see an SVO sentence and its OVS equivalent. Each participant was exposed to 40 SVO and 40 OVS sentence–picture pairs.
3.2.2. Testing materials
The pictures for the picture-description task were the same as in Experiment 1.
3.3. Procedure
The procedure was like that of Experiment 1. The reading span task indicated that cognate and non-cognate learners did not differ in working memory capacity (cognate learners’ mean partial reading span score: 44.45, SD = 11.76; non-cognate learners’ mean score: 48.90, SD = 10.47; t(57) = −1.54, p = .13, d = −0.40).
3.4. Predictions
Our main hypothesis for RQ2 was that cognates would not facilitate the acquisition of the cross-linguistically similar L2 structures. A secondary hypothesis was that there would be a learning advantage for the SVO structure over the OVS one, since the L2 structures would be processed using the equivalent L1 structures, and in Spanish, the SVO structure is much more frequent than the OVS structure (ADESSE corpus; García-Miguel et al., Reference García-Miguel, Vaamonde, González Domínguez, Calzolari, Choukri, Maegaard, Mariani, Odijk, Piperidis, Rosner and Tapias2010). This advantage would be comparable for cognate and non-cognate learners. A prerequisite of the hypotheses was that the structures were learnt. If this were the case, in the picture-description task, learners would correctly write significantly more than 50% of verb-medial sentences with a before the second noun (SVO structure with patient marking) and with a before the first noun (OVS structure with patient marking).
If our main hypothesis was correct, we predicted (P1) that cognate and non-cognate learners would be comparably accurate in their verb-medial picture descriptions with a before one of the nouns in the sentence (a-marking). If our secondary hypothesis was correct, we predicted (P2) that learners would be significantly more accurate when writing verb-medial picture descriptions with a before the second noun than before the first noun, and that this difference would be similar for the two groups of learners.
3.5. Data pre-processing and analysis
All analyses included the whole sample of participants. We assumed p < .05 as alpha for significance testing.
3.5.1. First and second vocabulary-learning phases
Data were coded and analysed as in Experiment 1.
3.5.2. Testing phase
First, we calculated the proportion of SVO and OVS sentences produced by cognate and non-cognate learners in the picture-description task (SVO if the subject/agent in the picture appeared in sentence-initial position and OVS if the object/patient in the picture appeared in sentence-initial position). A chi-square test compared the proportion of subject-initial and object-initial sentences across groups.
Next, we coded whether learners used a before a noun (1 = picture description with a, 0 = picture description without a). A generalised linear mixed-effects model with structure glmer (a-marking ~ −1 + Group + (1 | Participant) + (1 + Group | Item)) determined whether cognate and non-cognate learners used a in significantly more than 50% of sentences. Additionally, a generalised linear mixed-effects model with structure glmer (a-marking ~ Group + (1 | Participant) + (1 + Group | Item)) assessed whether the two groups produced a comparable amount of sentences with a.
Then, we calculated the percentage of sentences with a before the first or second noun, and we coded accuracy in the use of a (1 = correct use, 0 = incorrect use). To assess whether cognate and non-cognate learners used a-marking correctly in significantly more than 50% of verb-medial sentences with a before the first noun and before the second noun, we ran generalised linear mixed-effects models for each type of sentence separately. For sentences with a-marking before the first noun, the model fitted was glmer (Accuracy ~ −1 + Group + (1 | Participant) + (1 | Item)), as including a random slope of Group by item caused convergence issues. For sentences with a-marking before the second noun, the converging model was glmer (Accuracy ~ −1 + Group + (1 | Participant)). Group was treatment-coded (Cognate = 0, Non-cognate = 1) in all models.
To compare learning of the structures by cognate and non-cognate learners, we assessed whether the two groups’ accuracy when writing verb-medial picture descriptions with a before the first noun and before the second noun significantly differed. A generalised linear mixed-effects model tested for the interaction between Group (Cognate = 0.5, Non-cognate = −0.5) and Position of a (Before the first noun = 0.5, Before the second noun = −0.5) on accuracy. The model had random intercepts by participant and item, since including random slopes caused convergence issues.
3.6. Results and discussion
3.6.1. First vocabulary-learning phase
Picture–word matching task. The number of attempts at picture–noun matching was the same for the two groups (M = 1.13, SD = 0.35; Median = 1, MAD = 0). Similarly, non-cognate learners’ number of attempts to match all non-cognate verbs with their pictures (M = 1.40, SD = 0.56; Median = 1, MAD = 0) was comparable to cognate learners’ number of attempts to match all cognate verbs with their pictures (M = 1.27, SD = 0.45; Median = 1, MAD = 0; W = 401, p = .38, r = .12).
Picture-naming task. Non-cognate learners’ number of attempts to name all pictures of nouns accurately (M = 1.87, SD = 0.63; Median = 2, MAD = 0) was not statistically different from cognate learners’ number of attempts (M = 1.63, SD = 0.56; Median = 2, MAD = 0; W = 365, p = .15, r = .19). By contrast, the number of attempts to name all pictures of verbs correctly was significantly larger for non-cognate learners (M = 1.87, SD = 0.63; Median = 2, MAD = 0) than for cognate learners (M = 1.13, SD = 0.35; Median = 1, MAD = 0; W = 172, p < .001, r = .61). This suggests that cognate verbs were learnt faster than non-cognate verbs and adds to the studies supporting the cognate facilitation effect, including Experiment 1 (e.g., De Groot & Keijzer, Reference De Groot and Keijzer2000; Tonzar et al., Reference Tonzar, Lotto and Job2009).
In Section 2.6, we discussed how the cross-linguistic activation of formally similar words likely made learning cognates easier than learning non-cognates. The reasoning proposed would also explain why, in Experiment 2, cognate learners named all verb pictures appropriately in fewer attempts than non-cognate learners did.
3.6.2. Second vocabulary-learning phase
Picture–word matching task. The two groups attempted the task a comparable number of times (cognate learners: M = 1.27, SD = 0.58; Median = 1, MAD = 0; non-cognate learners: M = 1.27, SD = 0.52; Median = 1, MAD = 0; W = 439, p = .83, r = .03).
Picture-naming task. Non-cognate learners’ number of attempts to name all pictures of verbs correctly (M = 1.30, SD = 0.53; Median = 1, MAD = 0) was comparable to cognate learners’ number of attempts (M = 1.20, SD = 0.41; Median = 1, MAD = 0; W = 417, p = .51, r = .09).
In sum, the number of times that cognate and non-cognate learners attempted the picture–word matching task and the picture-naming task did not differ statistically. Since we considered that reaching 100% accuracy in these tasks indicated verb learning, we conclude that cognate and non-cognate learners learnt the same non-cognate verbs equally fast. This supports the claim that, in the first vocabulary-learning phase, cognate verbs were learnt faster than non-cognate verbs due to their cross-linguistic formal similarity, not due to participant-related factors.
3.6.3. Testing phase
Cognate learners wrote 70.42% of SVO picture descriptions and 29.58% of OVS picture descriptions. Similarly, non-cognate learners wrote 79.58% of SVO picture descriptions and 20.42% of OVS picture descriptions. Hence, the two groups learnt that the mini-language included two word orders. The difference in SVO versus OVS proportions between groups was significant (X 2 (1) = 4.9, p = .03). Additionally, cognate learners used a before a noun in 95% of verb-medial sentences (significance from 50%: β = 17.92, SE = 4.86, z = 3.69, p < .001). Non-cognate learners used a-marking in 90.42% of verb-medial sentences (significance from 50%: β = 15.65, SE = 4.07, z = 3.85, p < .001). This suggests that both groups were aware that the sentences of the mini-language contained the word a. The difference between groups was not significant (β = −2.27, SE = 5.69, z = −0.40, p = .69).
The two groups had a similar preference for writing sentences with a-marking before the second noun (cognate learners: 71.49% of sentences; non-cognate learners: 70.81% of sentences). Yet, they also used a-marking before the first noun (cognate learners: 28.51% of sentences; non-cognate learners: 29.19% of sentences), which indicates that both groups knew that a could appear in these two positions. Finally, cognate learners used patient marking correctly in 92.02% of verb-medial sentences with a before the second noun (significance from 50%: β = 9.14, SE = 2.14, z = 4.28, p < .001) and non-cognate learners did so in 98.65% of these sentences (significance from 50%: β = 10.15, SE = 2.66, z = 3.82, p < .001). This suggests that the SVO structure with prepositional patient marking was learnt. Likewise, both groups used patient marking correctly in significantly more than 50% of picture descriptions with a before the first noun (cognate learners in 89.23% of sentences, β = 11.66, SE = 3.96, z = 2.95, p = .003, and non-cognate learners in 75.41% of sentences, β = 10.49, SE = 3.11, z = 3.37, p < .001). This suggests that the OVS structure with prepositional patient marking was also learnt.
A mixed-effects model assessed whether cognate and non-cognate learners differed in the accuracy with which they wrote the cross-linguistically similar L2 structures (verb-medial sentences with a before the first noun or before the second noun). There was no effect of Group, an effect of Position of a and no interaction between the two (Table 4 and Figure 5). The lack of a Group effect indicates that there was no statistical evidence for a difference between cognate and non-cognate learners’ accuracy when writing verb-medial sentences with a, as predicted (P1). This suggests that the two groups learnt the structures to a comparable degree, in line with our main hypothesis for RQ2.
Table 4. Summary of the generalised linear mixed-effects regression model fitted for the test in Experiment 2
Note: Model structure: glmer (Accuracy ~ Group × Position_of_a + (1 | Participant) + (1 | Item), data = test_Exp2, family = binomial).
Figure 5. Percentage of correct picture descriptions with verb-medial word order and a-marking before the first or the second noun produced by cognate and non-cognate learners in Experiment 2. Error bars represent 95% confidence intervals.
Collapsing cognate and non-cognate learners’ productions, the mean percentage of correct picture descriptions with a-marking before the second noun was 95.18% (SD = 21.46%). The mean percentage of correct picture descriptions with a-marking before the first noun was 82.54% (SD = 38.11%). The effect of Position of a suggests that, as predicted (P2), learners were significantly more accurate in their verb-medial descriptions with a before the second noun than before the first noun. The absence of an interaction indicates that, contrary to what it may seem descriptively, this effect did not depend on the group of learners. This suggests that the SVO structure was better learnt than the OVS structure and that the learning difference was comparable for cognate and non-cognate learners, in line with our secondary hypothesis for RQ2.
Based on MOGUL, we claim that during the exposure phase, learners processed the SVO and OVS sentences using compatible L1 structures, which had a high resting activation level due to prior processing. The higher activation of cognate verbs compared to non-cognate verbs spread to the structures, which, as a result, had a higher resting activation level when containing a cognate than a non-cognate. Nevertheless, since the L1 structures used to process the L2 sentences already had a high resting activation level, the difference in this level resulting from processing with cognates versus non-cognates was negligible, and learning was comparable for cognate and non-cognate learners.
4. Comparing the results of the testing phase in Experiments 1 and 2
We conducted a joint analysis of the results of the picture-description task in the two experiments. A generalised linear mixed-effects model assessed whether there was an interaction between Group of learners (Cognate = 0.5, Non-cognate = −0.5) and Structure type (Dissimilar = 0.5, Similar = −0.5) on sentence accuracy (1 = sentence with correct word order and agent–patient marking, 0 = sentence with incorrect word order and agent–patient marking). The converging model had by-participant and by-item random intercepts (Table 5).
Table 5. Summary of the generalised linear mixed-effects regression model fitted to compare the performance in the test in Experiment 1 versus Experiment 2
Note: Model structure: glmer (Accuracy ~ Group × Structure_Type + (1 | Participant) + (1 | Item), glmerControl(optimiser = “bobyqa”), data = test_Exp1_vs_Exp2, family = binomial).
The test yielded an effect of Group, Structure type and an interaction between the two. This interaction confirms that there was a difference in the accuracy with which cognate and non-cognate learners wrote cross-linguistically dissimilar structures in Experiment 1 and cross-linguistically similar structures in Experiment 2.
5. General discussion
Experiment 1 showed that cognates facilitate adults’ initial acquisition of cross-linguistically dissimilar L2 structures, enriching the insights into this topic by Sanahuja and Erdocia (Reference Sanahuja and Erdocia2024). By contrast, the facilitative role of cognates was not observed for L1–L2 similar structures in Experiment 2. First, we will account for the findings of Experiment 1, detailing how learners could have processed the SOV and OSV sentences with cognate or non-cognate verbs during the exposure phase. We hypothesise that following the agent-first preference (Dryer, Reference Dryer, Dryer and Haspelmath2013), SOV and OSV sentences were initially interpreted as agent/subject-initial. Then, based on MOGUL (Sharwood Smith & Truscott, Reference Sharwood Smith and Truscott2014), we hypothesise that the mismatch between learners’ interpretation of OSV sentences and their accompanying picture led to associating the roles of agent and patient with –ak and –a, respectively. The target syntactic structures were created and, with each processing opportunity, their resting activation level (indicating how well the structures were learnt) increased.
Since cognates are more activated than non-cognates (Dijkstra & van Heuven, Reference Dijkstra and van Heuven2002), a stronger activation spread from cognate verbs than from non-cognate verbs to the structures containing them. Consequently, the activation of the structures was higher when including a cognate than a non-cognate. Within MOGUL, the result of this was that when activation decayed after processing, the structures processed with cognates landed at a higher resting activation level than that of the structures processed with non-cognates, and thus the former were better learnt than the latter. Since learning of the structures was greater for cognate than for non-cognate learners, the first group was more accurate than the second group when writing verb-final sentences with agent–patient marking in the picture-description task. These results and the reasoning behind them go in line with some of the findings in Sanahuja and Erdocia (Reference Sanahuja and Erdocian.d.), who argued that the stronger activation of high-frequency verbs compared to low-frequency verbs facilitated the acquisition of an L1–L2 dissimilar structure. Likewise, in Experiment 1, we argue that the stronger activation of cognate verbs compared to non-cognate verbs facilitated the acquisition of L1–L2 dissimilar structures.
The facilitative role of cognates in L2 syntax acquisition could also be tentatively explained by the Lexical Bottleneck Hypothesis (Hopp, Reference Hopp2014, Reference Hopp2018), which argues that costly lexical processing may deplete the resources needed to perform a target-like syntactic computation. Therefore, it predicts that, in Experiment 1, processing sentences with non-cognates may have been so costly that it exhausted the resources necessary to achieve a target-like processing and acquisition of the dissimilar structures. Consequently, as observed, learning of the dissimilar structures was greater for cognate than for non-cognate learners.
Finally, participants learnt that the subject/agent and the object/patient could appear sentence-initially, which would explain why cognate and non-cognate learners wrote SOV and OSV picture descriptions irrespective of whether agent–patient marking was used. Both groups favoured subject-initial sentences over object-initial sentences, and in a similar proportion. This could be explained by the agent-first preference, the tendency in most languages (including Spanish, learners’ L1) to put agents before patients. Additionally, there was no evidence that the SOV or the OSV structure with postpositional agent–patient marking was better learnt than the other one. That is, learners did not write significantly more SOV than OSV picture descriptions with correct agent–patient marking, arguably, because the two structures were dissimilar to L1 grammar and both had to be learnt from scratch.
Turning to Experiment 2, we discuss how learners could have processed the SVO and OVS sentences with prepositional patient marking in the exposure phase. Based on MOGUL, we assume that SVO sentences were processed using a compatible L1 structure having a high resting activation level due to previous processing in the L1. Following the agent-first preference, the first noun was correctly interpreted as the agent. Likewise, due to the agent-first preference, OVS sentences were initially processed as subject-initial. This led to an incorrect interpretation of these sentences, as indicated by the pictures representing their meaning. After misprocessing a few OVS sentences, these were reanalysed as object/patient-initial using an appropriate L1 construction, which also had a high resting activation level (although not as high as the L1 SVO structure; see below).
The stronger activation of cognate verbs, compared to non-cognate verbs, spread to the SVO or OVS structure, causing them to be more activated, and to have a higher resting activation level after processing, when including a cognate than a non-cognate. However, because the L1 SVO and OVS structures used to process the L2 sentences had a high resting activation level already at the beginning of L2 acquisition, any difference in their resting level resulting from processing with cognates versus non-cognates was negligible. Consequently, the structures were comparably learnt by cognate and non-cognate learners. This would explain why, in the picture-description task, we found no statistical evidence for a difference between cognate and non-cognate learners’ accuracy when producing verb-medial sentences with patient marking. Our result aligns with some of the findings in Sanahuja and Erdocia (Reference Sanahuja and Erdocian.d.). In that study, the stronger activation of high-frequency verbs compared to low-frequency verbs did not facilitate the acquisition of an L1–L2 similar structure. Similarly, in Experiment 2, the stronger activation of cognate verbs compared to non-cognate verbs did not facilitate the acquisition of L1–L2 similar structures.
Learners were significantly more accurate when writing verb-medial picture descriptions with a before the second noun (correct SVO sentences with patient marking) than before the first noun (correct OVS sentences with patient marking). The difference between the two types of sentences did not significantly vary by group of learners. This suggests that the SVO structure was better learnt than the OVS structure, arguably because the frequency of occurrence of the L1 SVO structure used to process L2 SVO sentences was higher than that of the L1 OVS structure used to process L2 OVS sentences (ADESSE corpus; García-Miguel et al., Reference García-Miguel, Vaamonde, González Domínguez, Calzolari, Choukri, Maegaard, Mariani, Odijk, Piperidis, Rosner and Tapias2010).
Finally, cognate and non-cognate learners wrote a larger number of subject-initial sentences than object-initial sentences, irrespective of patient marking. As in Experiment 1, this could be attributed to the agent-first preference. Non-cognate learners wrote a larger proportion of subject-initial sentences than cognate learners did. This could be because, during the exposure phase, the general preference to interpret the first animate noun in a sentence as the agent was more salient when lexical processing was hard (when the sentences included non-cognate verbs). The result of this was that for non-cognate learners, agent-first sentences were particularly easier to process than non-agent-first ones and, hence, that these were the preferred types of sentences in the picture-description task.
It might be possible to extract pedagogical implications from Experiments 1 and 2, suggesting that teachers exploit cross-linguistic syntactic similarity and use cognates to facilitate learning of L1–L2 dissimilar structures. However, classroom-based studies are needed to test if the facilitative role of cross-linguistic syntactic similarity and lexical processing holds in those learning environments.
6. Conclusion
Experiments 1 and 2 show that the stronger activation of cognates compared to non-cognates eased the acquisition of cross-linguistically dissimilar L2 structures. This facilitation was not found for cross-linguistically similar L2 structures, which could be processed using similar L1 constructions. These findings shed light on the interaction between lexical and syntactic processing during initial L2 syntax acquisition, extending the insights into this topic by previous research within the MOGUL framework.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S1366728925100692.
Data availability statement
The materials, data and scripts for analysis that support the findings of this study are openly available in the OSF repository at https://osf.io/6y24n/.
Acknowledgements
We would like to thank all the participants who took part in the study.
Competing interests
The authors declare none.
Funding statement
We thank the University of the Basque Country (Grant No. PIF19/08), the Basque Government (Grant No. IT1439-22) and the Ministry of Science, Innovation and Universities of the Spanish Government (Grant Nos. PID2021-124056NB-I00, PID2021-127146NB-I00 and PID2024-156359NB-100) for their financial support.
Open access
