“Les copains *dit au revoir”: On Subject–Verb Agreement in L2 French and Cross-Linguistic Influence

Abstract

This study focuses on the production of subject–verb (SV) agreement in number in L2 French and investigates the role of cross-linguistic influence (CLI) in this particular morphosyntactic domain. CLI is a well-known phenomenon in Second Language Acquisition (SLA) research but it has rarely been investigated systematically in relation to SV agreement in French. The participants of the study are 114 learners with Italian, German, Dutch and Swedish as L1. The source languages are all inflectional languages but they vary in terms of morphological richness in the verb paradigm, ranging from very poor (Swedish) to very rich (Italian). The participants performed an oral narrative task contrasting singular and plural contexts of SV agreement. Results indicate a significant difference between L1 groups in terms of correct SV agreement but they also show that the overall presence of rich verb morphology in the L1 does not, on its own, result in a more correct SV agreement. It is when comparing learners at two different proficiency levels that we observe differences in the rate of L2 development, which may be explained as an effect of CLI. Overall, results indicate a complex interplay of different factors, where the role of CLI must be further investigated in future studies in relation to L2 French.

1. Introduction

Previous literature on the acquisition1 of spoken French has shown that subject–verb (SV) agreement in number (third person singular vs. plural) is a difficult morphosyntactic phenomenon for L2 learners (Bartning and Schlyter 2004; Howard 2006; Michot 2014; Véronique 2009, among others). Even at advanced levels, learners continue to make agreement errors such as that presented in (1).

(1)	Les copains	*/di/	au revoir
	The–PL friend–PL	say–SG	goodbye
	“The friends say goodbye”

The use of singular verb forms in contexts where plural verb forms are required (e.g., /diz/ in example 1) is also observed in monolingual and bilingual children learning L1 French (Ågren and van de Weijer 2013b; Bassano et al. 2001; Kilani-Schoch 2003; Prévost 2009). Two factors are usually discussed that might explain these difficulties. The first factor is the complexity of the agreement system itself, involving many different agreement patterns and irregular forms. SV agreement in number is not audible in regular -er verbs from the first conjugation (e.g., discuter, “to discuss”) but clearly distinguished in verbs belonging to other verb classes (see Section 3). In addition, the frequency distribution of verb forms in native speaker discourse is biased towards the singular, meaning that the singular form of the verb is always much more frequent than the equivalent plural form (Ågren and van de Weijer 2013a). This fact makes the singular verb form more salient to language learners and therefore easier to memorize and to access than the plural form. The second factor is the learners’ exposure to the target language. Typically, learners who are exposed to more input will produce more correct SV agreement in number than learners who are exposed to less input (cf. young bilingual children and especially foreign language classroom L2 learners). A third factor in L2 learning, less often discussed in the literature, is cross-linguistic influence (henceforth CLI) from the learners’ L1 (see, for example, Stabarin and Gerolimich 2014). However, to the best of our knowledge, the influence of the learners’ L1 on their production of SV agreement in L2 French has rarely been investigated systematically, giving rise to the rationale for the focus on CLI in this paper. More precisely, in order to better understand the role of the learners’ L1 in the acquisition process, we compare the production of SV agreement in number in four groups of L2 learners with different L1s, namely Dutch, German, Italian and Swedish, performing the same oral narrative task. As will be further described below, these four languages differ in their relationship to the target language in this particular morphosyntactic domain (Ringbom 2007) and they vary strongly in morphological richness (Xanthos et al. 2011). Apart from these differences in their L1, the participants of the study are all adult classroom L2 learners of French at the A2–B1 levels (European Council 2001), studying French in a university setting in their home country.

2. Briefly on Cross-Linguistic Influence (CLI)

CLI, also known as transfer or interference, is a well-known phenomenon in the Second Language Acquisition (SLA) literature2 (Weinreich 1953; Sharwood Smith and Kellerman 1986a; Odlin 1989; Jarvis and Pavlenko 2008). In very general terms, CLI implies that the L2 learner will use prior linguistic knowledge from his or her L1, or from other previously acquired languages, when acquiring a new language. Previous knowledge of at least one other language is thereby a factor that distinguishes L2 from L1 acquisition. Initially, research on CLI focused on the influence of the L1 on the acquisition of an L2. However, in recent years, there has been a general awareness that many L2 learners have a multilingual repertoire, a fact that will influence their acquisition of a new target language. A growing body of empirical evidence shows that CLI is in fact multi-directional and that all languages known by the learner can influence each other (Aronin and Singleton 2012). Indeed, the growing interest in the study of L3 acquisition underlines that the impact of other L2s (previously acquired second languages) on the so-called L3 (the language currently being acquired) might be of great importance (see Falk and Bardel 2010 for an overview; Rothman et al. 2019 for a recent discussion). Furthermore, CLI might also work in the opposite direction since the L2 can influence the L1, the L3 can affect the L2, etc. (Jarvis and Pavlenko 2008; Tsang 2017).

In the early days of SLA research, transfer of knowledge from the L1 to the L2 was considered the key issue explaining L2 development and the errors produced in different groups of learners (Contrastive Analysis; Lado 1957). When comparing grammatical structures in the L1 with those of the L2, the assumption was that differences are difficult and similarities are easy to acquire in the L2. However, Selinker (1972) and others pointed out that L2 development, i.e., interlanguage development, was less straightforward than expected from the contrastive analysis approach. Empirical evidence showed that grammatical features that were different in the L2 compared to the L1 were not necessarily difficult for L2 learners to acquire and, on the other hand, L2 learners did not always easily acquire grammatical features that were similar to those of the L1. As exemplified in Ellis and Shintani (2014, p. 236), French learners of L2 English would not make transfer errors of word order such as “I them see”, even though the object pronoun is preverbal in French as opposed to English. Furthermore, French learners at beginner levels would struggle with the inverted question forms in yes/no questions in L2 English, even though French has the same word order. These and similar results spoke in favor of universal patterns in L2 development rather than of L1 transfer across the board.

Over the last few decades, the role and the effects of CLI on the production and comprehension of the L2 have been extensively discussed (Jarvis and Pavlenko 2008). Most researchers in this field agree that CLI plays a role and that its effects are evident in both classroom and naturalistic settings (Ellis and Shintani 2014). However, the importance, the limitations and the mechanisms involved in the process are still under debate (see Rothman et al. 2019). Generally, transfer effects from the L1 are considered stronger in low-proficiency learners (Hermas 2014) but the relation between CLI and proficiency is not straightforward. According to Ellis and Shintani (2014, p. 237), some transfer errors seem to appear only when learners have reached a certain proficiency threshold. They suggest that, in some respects, it is fruitful to think of the effects of CLI in terms of rate of acquisition, as the L1 might help L2 learners to overcome typical interlanguage errors faster and make them shift to more target-like structures early on. In general, CLI is considered one internal factor among others (cf. age and motivation), interacting with external factors (quantity and quality of input, teaching, etc.) and linguistic factors (regularity, saliency, frequency of linguistic structures) in shaping L2 development (a.o. Long 1990; Jarvis and Odlin 2000; DeKeyser 2005; Odlin 2005; Treffers-Daller and Sakel 2012; von Stutterheim et al. 2013; Tsang 2017; Tang et al. 2020).

The effects of CLI can be positive or negative (Odlin 1989). Positive CLI from one language to the other will facilitate and accelerate the learning of specific linguistic phenomena. On the other hand, negative CLI will slow down or hinder the acquisition process of specific structures. The identification of negative CLI is simple and straightforward since it leads to systematic errors that are easily observable in learner data. In fact, early research has mainly focused on negative CLI as a source of interference in the L2 learning process. However, Jarvis and Pavlenko (2008) claim that negative transfer only accounts for a small proportion of all transfer effects. They observe that it does not cover more subtle effects such as overgeneralizations or avoidance strategies. In addition, positive effects of CLI can indeed be more difficult to observe in learner data since they result in target-like language use. In this respect, Foote (2009) underlines that positive effects of CLI seem to increase when the languages involved are typologically similar. Typological proximity is often discussed as an important factor in the CLI literature, meaning “the distance that the linguist can objectively and formally define and identify between languages and language families” (De Angelis 2007, p. 22). This is what Ringbom and Jarvis (2009, p. 107) refer to as actual similarities between languages. Put simply, two typologically related languages—for example, French and Italian—are more likely to influence each other than two languages that are typologically distant—for instance, French and Mandarin Chinese. The importance of the typological factor is, for instance, discussed in detail by Rothman (2011) in relation to his Typological Primacy Model (TPM) in L3 acquisition. A closely related phenomenon put forward by Kellerman (1983) is psychotypology, namely the similarities and differences among languages as perceived by learners. Ringbom and Jarvis (2009, p. 107) refer to this phenomenon as assumed similarities and they claim that this kind of similarity has the strongest and most direct impact on language learning and performance. They underline that while actual similarities are constant over time, assumed similarities change with learners’ increasing experience with the target language. However, according to Rothman (2011, p. 112), psychotypological and actual typological similarities are in many cases the same, as revealed in his study of the L2 and L3 acquisition of different Romance languages.

In relation to typological and psychotypological proximity, it is interesting to consider Ringbom’s (2007) suggestion that there are three different types of relationships between source and target languages concerning specific linguistic structures (see also Ringbom and Jarvis 2009). First, there might be a similarity relation involving a one-to-one relation between form and function in the two languages (for instance, the same morpheme is used to express agreement in both languages). Full-scale cross-linguistic similarity in both form and function is, according to Ringbom and Jarvis, a rare phenomenon. Second, a contrast relation involves an underlying similarity in function alone (e.g., SV agreement exists in both languages but is realized differently). Finally, in a (near) zero relation, there is an absence of similarity in both function and form between the languages involved or a very abstract relation that a typical learner will not be able to grasp. In our study of SV agreement in L2 French, we will compare groups of learners with typologically different SV agreement systems in their source languages in order to study the effects of both positive and negative CLI. From a linguistic point of view (i.e., actual similarity), we note that three of the source languages in our study have contrast relations to French as far as SV agreement is concerned, even though they differ from each other in terms of morphological richness (Dutch, German and Italian). One of them has a zero relation to French in this respect (Swedish). The different relationships between source and target languages will be further described in Section 3.

3. Subject–Verb Agreement in Number

In this section, we briefly introduce the SV agreement system of spoken French, followed by a description and a comparison of the L1s involved in the study. Based on this description, we formulate hypotheses for the possible influence of the different L1s on the production of SV agreement in L2 French.

3.1. Subject–Verb Agreement in Number in Spoken French

Like many other languages, French encodes a distinction between singular and plural reference. This distinction results in grammatical number agreement. However, compared to written French, where plural verb forms are orthographically clearly distinct from singular ones, SV agreement in number in spoken French is best described as partial and heterogeneous. As underlined by Dubois (1967), the study of number agreement in the verb phrase (VP) should concentrate on the third person, since this is the most frequent and unambiguous number alternation in French3. In spoken French, this agreement is expressed by an alternation of the verb stem. This singular vs. plural alternation is involved in less than 12% of French verbs, since verbs from the first conjugation (-er verbs) are invariable in number, as exemplified in (2) and (3) below (see New et al. 2004, based on the corpus Lexique, New and Pallier 2020).

(2)	Le copain	/paʁl/	français
	The-SG friend-SG	speak-SG	French
	“The friend speaks French”
(3)	Les copains	/paʁl/	français
	The-PL friend-PL	speak-PL	French
	“The friends speak French”

Thus, in the present tense, only a small proportion of French verbs have a distinct verb stem in the plural as compared to the singular. However, verbs with an audible stem alternation in number are still essential to the use of spoken French, since they belong to the most frequent verbs. According to the Gougenheim et al. corpus4 (1964), the eleven most frequent verbs in French belong to this category (être “to be”, avoir “to have”, faire “to do”, dire “to say”, pouvoir “to be able to”, aller “to go”, voir “to see/understand”, savoir “to know”, vouloir “to want”, venir “to come”, devoir “to have to”). In addition, among the 50 most frequently used verbs in French, 29 involve an audible SV agreement in number, as exemplified in (4) and (5).

(4)	Le copain	/di/	bonjour
	The-SG friend-SG	say-SG	hello
	“The friend says hello”
(5)	Les copains	/diz/	bonjour
	The-PL friend-PL	say-PL	hello
	“The friends say hello”

In addition, the morphophonological phenomenon called liaison must be mentioned. In cases where a plural subject pronoun (ils/elles “they”) meets a vowel-initial verb (ils appellent /ilzapɛl/ “they call”), the final -s of the pronoun will surface in connected speech through the liaison consonant /z/ (see Howard and Ågren 2019, for details). For this to happen, the verb must start with a vowel. The liaison consonant is never realized if the two words are pronounced in isolation or when the verb starts with a consonant (ils parlent /ilpaʁl/ “they speak”).

In what follows, we distinguish four different types of agreement patterns in number (3rd SG vs. PL) in spoken French (following Michot 2014; Granget et al. forthcoming).

• The first pattern (Vont) includes four very frequent verbs, namely être “to be”, avoir “to have”, faire “to do” and aller “to go”. In these verbs, the singular/plural alternation is expressed by totally (est/sont “is/are”) or partially (fait/font “does/do”) different morphemes, e.g., suppletive forms, with no or little connection to the base form (Prévost 2009). In this group of verbs, the singular/plural alternation is based on a vowel shift on /ɛ/ or /a/ in singular versus /$\widetilde{ɔ}$ // in plural and it does not have plural forms ending in a consonant (see below, patterns 2 and 3). These verbs are used both as lexical and auxiliary verbs (avoir “to have” and être “to be” + past participle) and modal verbs (aller “to go” and faire “to do” + infinitive), which obviously increase their frequency in both spoken and written French (see Ågren and van de Weijer 2013a).
• The second pattern (Vrad) is that of verbs like prendre “to take” or savoir “to know”, based on a stem alternation including a vowel shift in combination with the adjunction of a consonant in the plural (elle sait /ɛlsɛ/ “she knows” vs. elles savent /ɛlsav/ “they know”). The verb final consonant in the plural varies from verb to verb, which makes this agreement pattern very irregular in spoken French.
• The third pattern (Vcons) does not include a vowel alternation in the plural. Instead, the plural is marked via the adjunction of a consonant in verb-final position, which varies from verb to verb (see, for example, the verb dire “to say” in examples 4 and 5, but also finir “to finish”, vendre “to sell”, devoir “to have to”, etc.). In these verbs, as in pattern 2 above, the final consonant in coda position is not always clearly articulated and can indeed be difficult to perceive in the spoken input, especially when the verb is followed by a consonant-initial word. As in pattern 2, this pattern is based on the use of an irregular stem alternation that will have to be acquired verb by verb, even though some sub-groups can be distinguished—for example, many verbs on -ir, like finir “to finish” with a plural form on /is/, e.g., ils finissent “they finish”.
• The fourth pattern (Vuni) includes verbs that lack an audible singular/plural distinction on the verb in the third person, such as regular -er verbs like discuter “to discuss” /ildiskut/ “he/they discuss(es)” or some irregular verbs like courir “to run” /ilkuʁ/ “he/they run(s)”. In this study, we will take into consideration liaison contexts in vowel-initial verbs from the Vuni pattern. As mentioned above, liaison is a clear discriminative number marker in VP (i.e., plural).

To conclude, SV agreement in number in spoken French is unmarked in the singular and expressed morphologically in the plural by a combination of liaison, suppletive forms, vowel alternations and adjunctions of various verb-final consonants. Therefore, number agreement is very heterogeneous in spoken French and expressed mainly in a small group of (sometimes) very frequent verbs.

3.2. Subject–Verb Agreement in Number in the Learners’ L1

Swedish is a North Germanic language characterized by the absence of SV agreement in person and number. In the present tense, in both spoken and written Swedish, there is only one finite verb form, ending in /r/, used in all persons in both singular and plural, as in the verb att spela “to play”: jag/du/han/hon/vi/ni/dom /spelar/ “I/you/he/she/we/you/they play” (Dahl and Koptjevskaja-Tamm 2010). It should be mentioned that the final consonant /r/ of the finite forms is not always clearly articulated, which means that the audible difference between the non-finite form, spela “to play” and the finite verb form, spelar “play(s)” is subtle. There are only a few exceptions to this pattern—for example, some modal verbs where the finite form is måste “must”, kan “can” and vill “want” (cf. German below). However, Swedish speakers are not totally unfamiliar with morphosyntactic agreement since Swedish is characterized by rich agreement in NP (gender, number and definiteness).

Dutch is a Germanic language that distinguishes regular and irregular verbs and its agreement system marks singular agreement whereas the plural forms coincide with the default infinitive form in -en (werk-en, “to work/work”) (Haeseryn et al. 1997). In spoken language, the -en form is not always clearly articulated but sometimes pronounced /ə/. In regular verbs, SV agreement is expressed by means of suffixation: +Ø for first person singular (werk-) and +t for second and third person singular (werk-t). Irregular verbs typically agree by stem alternation—in some cases, also in combination with the +t suffixation—resulting in a small variety of agreement paradigms. The verbs hebben “to have” and zijn “to be” show the richest verb paradigms, as they distinguish first person heb “have” and ben “am”, second person heb-t “have” and ben-t “are” and third person heeft “has” and is “is”. The irregular verbs mogen “may”, zullen “will” and kunnen “can” have one or two singular forms depending on the register (formal/informal). The verb gaan “to go” represents an intermediate irregular paradigm, which distinguishes ga- Ø “go” for first person and gaat “go/goes” for second and third person. To summarize, Dutch SV agreement is marked by suffixation in regular verbs, resulting in three distinctive forms, and by stem alternation in irregular verbs, with a range of two to four distinctive singular and plural forms (cf.

Table 1. Cross-linguistic similarity and morphological richness of the source languages.

L1	Cross-Linguistic Similarity Relation with French SV Agreement in Number (Ringbom 2007)	Morphological Richness (Xanthos et al. 2011)
Swedish	Zero	1
Dutch	Contrast	2-3-4
German	Contrast	4
Italian	Contrast	6

).

In German, which is also a Germanic language, all verbs have distinct forms in singular and plural. There is an alternation of suffixes on /t/ in third person singular and on /ən/ or /n/ in third person plural, as in kauft “buys” vs. kaufen “buy” (cf. Dutch). This alternation is very systematic, with only a few exceptions—for example, some modal verbs which form their singular form without the /t/ morpheme: soll (must), kann (can) and will (want) (Bittner 1996). In addition to suffixation, certain verbs in German also include stem alternation, where the vowel in the first syllable is modified. For instance, the verb nehmen (to take) has the third singular form nimmt and the third plural form nehmen (homophone to the infinitive form). However, this vowel alternation is not a regular phenomenon.

Italian is a Romance language characterized by a rich morphological agreement system in both NP and VP, largely based on suffixation. In all types of verbs, person and number are phonologically marked on the verb form itself, while the presence of a subject pronoun is optional (pro-drop). Hence, as far as the number distinction in third person is concerned, the singular form on /a/ or /e/ is always clearly distinguishable from the plural form on /anɔ/ or /ɔnɔ/. Verbs from the first conjugation forming their third person singular on /a/, such as parla (“speaks”), are the most frequent in spoken Italian according to corpus data (De Mauro et al. 1993; Bellini and Schneider 2019). Even though suffixation is the main form of SV agreement in Italian, many irregular verbs also involve variation in the verb stem, which even further distinguishes the different persons from each other, such as the verb andare (to go): vado (1SG), vai (2SG), va (3SG), andiamo (1PL), andate (2PL) and vanno (3PL).

When examining the possible influence of the four source languages on the production of SV agreement in L2 French, we were inspired by the work of Xanthos et al. (2011) on morphological richness and language development. We used their model to describe the morphological richness of the verb paradigm in the different source languages of our corpus. Xanthos et al. (2011, p. 461) define paradigmatic morphological richness as “the tendency of a language to have a large number of formally distinct inflected word-forms per lemma”. This model was originally used to examine the role of morphological richness in the parental input on children’s early development of noun and verb morphology. Xanthos et al. found a strong positive correlation between morphological richness and the rate of morphological development in child speech. As illustrated in Table 1, we use a simplified version of the model and consider the number of distinct verb forms in the present tense (first to third person singular and plural) in the respective source languages as a measure of morphological richness. The source languages range from one verb form in Swedish to six distinct verb forms in Italian, with Dutch and German placed at intermediate positions. In Ringbom’s (2007) terms, Dutch, German and Italian all have a contrast relation to French, involving some degree of similarity in function (number agreement), while Swedish and French exhibit a zero relation in this respect.

In order to investigate the role of L1 influence on the production of SV agreement in number in L2 French, the following research question is addressed in the present study:

• To what extent do the morphosyntactic properties of the learners’ L1—more precisely, its morphological richness in VP—influence their production of SV agreement in number in spoken L2 French?

The null hypothesis (H0) is that the linguistic properties of the target language and the complexity of the agreement system in spoken French will override the importance of the learners’ L1 in this particular domain. According to the null hypothesis, we would thus expect L2 learner groups with different L1s to perform in a similar way when producing SV agreement in number. However, an alternative hypothesis (H1) would be that the morphosyntactic properties of the L1 will play a prominent role in the acquisition of SV agreement in number in spoken French. If this is the case, we expect learners with an L1 characterized by a rich SV agreement in person and number (Italian) to perform significantly better than learners with an L1 characterized by a partial SV agreement system (Dutch and German) or a L1 that lacks SV agreement altogether (Swedish).

4. Materials and Methods

4.1. Participants

The study includes 114 participants learning French in a university setting (

Table 2. General description of the different learner groups.

L1 Groups	Number of Learners at Level A2	Number of Learners at Level B1	Number of Participants (Total)	Mean Age at Testing	Mean Age of Onset (French)
ITA	15	15	30	21.3	16.4
GER	15	10	25	27.1	15.8
NLD	7	22	29	19.4	10.5
SUE	15	15	30	27.2	13.6
TOTAL	52	62	114	23.7	14.1

). The learners belong to four different subgroups according to their L1, henceforth labeled ITA (Italian), GER (German), NLD (Dutch) and SUE (Swedish). Each L1 group includes between 25 and 30 participants. All four groups include learners at two distinct developmental stages: the post-initial A2 level and the intermediate B1 level of the Common European Framework of Reference for Languages (European Council 2001). We evaluated the proficiency level of each participant using the vocabulary test included in the DIALANG test battery available online. DIALANG is a digital platform for self-evaluation of general second language proficiency aligned to the six proficiency levels of the Common European Framework of Reference for Languages (CEFR, European Council 2001). Previous research (see, for example, De Jong et al. 2012) has shown that vocabulary knowledge is a good predictor of general language proficiency. We used the vocabulary test as an independent measure of the learners’ general proficiency level of L2 French. Only learners at the A2 and the B1 levels were included in the study.

Each participant filled in a language background questionnaire including personal data and information on first and second languages, learning contexts, etc. The participants in the NLD group, coming from the Dutch-speaking part of Belgium, generally started learning French at school earlier than the other three groups. Furthermore, the mean age at testing is lower in the NLD and the ITA groups. This is because these students generally start their studies at the university at a younger age and because these groups do not include any learners over 30. Both the GER and the SUE groups include a couple of participants over 50, which raises their mean age at testing. The common denominator for the participants is their L2 French proficiency level (see Table 2) and the fact that they are speakers of L2 English. In addition, some of them have other languages in their linguistic repertoire5.

4.2. Tasks

In order to test if the learners’ use of SV agreement in spoken French varies in singular and plural contexts, we used a narrative task called “Paul and Pauline are having a party” (Paul et Pauline font la fête). This task has been used in previous research on SV agreement in other groups of learners (see, for example, Ågren and van de Weijer 2013a, 2013b; Ågren 2014). The task is a picture-story including 30 colored images of two children preparing and attending the birthday party of their friend. It includes an alternation of pictures where one child (singular) versus several children (plural) are involved in the action (see examples in Appendix A). In this study, the task was further elaborated in order to introduce a more varied use of different agreement patterns. Thus, the pictures elicit the use of specific verbs with an audible number agreement in spoken French (recevoir “to receive”, aller “to go”, dire “to say”, mettre “to put on”, etc.). The learners watched the picture-story on a computer screen and scrolled from one picture to the next at their own pace. They were asked to retell the story in the present tense, in as much detail as possible, and were recorded with the computer tool Audacity.

4.3. Data Analysis

We transcribed the narratives according to the CHAT format and analyzed the corpus by means of the CLAN tools (MacWhinney 2000). In our annotation system, a dependent tier called %ver was added to indicate which verb forms were used in each specific subject–verb context. We annotated the following information:

• Number context of the sentence (SG vs. PL)
• Verb type (Vuni, Vont, Vrad or Vcons; see Section 3.1)
• Type of subject:

  ○

  NP (la fille ‘the girl”)

  ○

  proper noun (Anne)

  ○

  pronoun (elle “she”)

  ○

  NP+pronoun (la fille elle…)

  ○

  coordinated subject (la fille et le garçon “the girl and the boy”)

  ○

  relative pronoun (qui, “who”)

• Verb form used in each specific agreement context:

  ○

  (i) target-like according to the context

  ○

  (ii) non-target-like but present in the paradigm of the target verb, such as the 3sg form (prend “takes”) instead of the plural form (prennent “take”), the 3pl form instead of the 3sg form or the infinitive form (prendre “to take”)

  ○

  (iii) unexpected forms, which means all other verb forms produced outside the verb paradigm of a specific verb (ils */ʁəsyv/, instead of ils reçoivent /ilʁəswav/ (“they receive”)).

Example 6 from the narrative of the Swedish learner SUEA2076 illustrates a transcribed utterance with its annotation tier.

(6)	SUEA207
	STU:	ils */ʁəsyv/ [*] une invitation pour une fête à le maison de leur ami
	%ver: recyv&ContPlur&Vcons&Spron&Finatt7

In the case of a repetition or a reformulation of the verb form used, we maintained the last produced form in our analyses. This annotation system allows us to calculate the number and the type of verb forms produced per learner and per group, in relation to the target context (singular or plural). In order to compare the results in different learner groups, number contexts and verb patterns, we used a statistical analysis based on the Pearson’s X² test. The total number of correct and incorrect verb forms produced in the corpus is summarized in Appendix B. In the following section, we present the results, taking into consideration differences between number categories (singular vs. plural), L1-groups, proficiency levels and verb patterns.

5. Results

Table 3. Overview of the analyzed verb forms in third person singular and plural (tokens).

	SINGULAR					PLURAL
L1	Vont	Vrad	Vcons	Vuni	Vindef	Vont	Vrad	Vcons	Vuni	Vindef	Total
ITA	164	75	91	257	15	198	31	181	471	12	1495
GER	129	46	51	169	5	150	25	114	212	15	916
NLD	277	85	124	327	0	229	33	196	290	3	1564
SUE	388	118	151	390	0	299	47	224	353	0	1970
Total	958	324	417	1143	20	876	136	715	1326	30	5945

gives an overview of the total number of verb forms (tokens) in all agreement patterns. Overall, the corpus includes 5945 verb forms produced with third person singular and plural subjects.

In the four L1 groups and in both singular and plural contexts, verbs from the Vuni and from the Vont patterns are the most frequently used. However, note that number agreement is silent in Vuni and therefore not analyzed further in this study, unless the verb is vowel-initial and preceded by a plural subject pronoun (see Table 7). Verb forms belonging to the Vcons and especially the Vrad patterns are clearly less frequent in the corpus. Fifty forms have been categorized as Vindef, meaning that they are interlanguage forms, which are not identifiable as belonging to any of the analyzed agreement patterns (see example 7). These forms will not be included in the subsequent analyses.

(7)	GERA206
	ils	/pʁaktis/ [*]	danser
	“they practice dancing”

We also note that the different L1 groups differ in productivity. The Swedish learners produce the largest amount of SV agreement contexts (1970 verb forms—on average, 65.7 verbs/learner)—which is more than the Italian- and Dutch-speaking learners (1495 forms, 49.8 verbs/learner vs. 1564 forms, 53.9 verbs/learner) and more than the double those of the German-speaking group, which is the least productive (916 forms, 36.6 verbs/learner).

The frequency of verb forms that agree correctly in number with their subject (irrespective of agreement pattern) is shown in

Table 4. Number of correctly agreeing verb forms in singular and plural contexts (proportion correct agreement within brackets).

	SINGULAR			PLURAL
L1	Total	A2	B1	Total	A2	B1
ITA	337 (94.4)	147 (94.2)	190 (94.5)	302 (58.3)	108 (45.7)	194 (68.8)
GER	211 (92.1)	110 (91.7)	101 (92.7)	222 (62.2)	109 (51.4)	113 (77.9)
NLD	496 (95.8)	135 (94.4)	361 (96.3)	390 (71.8)	70 (59.3)	320 (75.3)
SUE	661 (94.7)	350 (92.1)	311 (97.8)	392 (57.0)	187 (54.5)	205 (59.4)
Total	1705 (94.6)	742 (92.9)	963 (96.0)	1306 (62.0)	474 (52.1)	832 (69.5)

. Here, we indicate the number of correctly agreeing verb forms as well as the proportion of correct SV agreement per L1 group and per proficiency level (see Appendix B for details). As expected, the results clearly indicate that the participants in all L1 groups struggle more with agreement in plural than in singular contexts. In plural contexts, 62% of the verbs produced agree in number. In contrast, 94.6% of all verb forms agree in singular contexts. The overall level of SV agreement in singular and plural contexts differs significantly (X² = 583.72, df = 1, p < 0.001).

Furthermore, Table 4 illustrates that the difference in overall SV agreement between L1 groups is small. In the singular, results range from 92.1% correct agreement (GER) to 95.8% (NLD) and the differences between L1 groups are non-significant. In plural contexts, differences between groups are larger and range from 57% (SUE) to 71.8% (NLD), with statistical analysis revealing a significant difference between L1 groups (X² = 32.626, df = 3, p < 0.001). In order to investigate this finding, we made a pairwise comparison of all L1 groups involved. This comparison reveals that the NLD group differs significantly from the other groups. However, there was no significant difference between the ITA, GER and SUE groups (NLD vs. ITA X2 = 21.37, df = 1, p < 0.001; NLD vs. GER X2 = 9.1954, df = 1, p < 0.01; NLD vs. SUE X2 = 28.869, df = 1, p < 0.001). Thus, the differences in SV agreement in plural contexts observed between L1 groups are due to the relatively high performance of the NLD group in relation to the other three groups. In addition, we note that the L1 groups at the two end-points of our morphological richness scale, ITA and SUE (cf. Table 1, Section 3.2), perform almost identically. It is between these two groups that we hypothesized that we would find a difference in SV agreement due to their variation in morphological richness. However, the data analysis of our corpus does not confirm this hypothesis. These results will be further discussed in Section 6.

Finally, we consider the level of SV agreement in the two proficiency levels studied: A2 vs. B1. Overall, results are stable at the A2 and B1 levels in singular contexts. The L1 groups range from 92.9% correct agreement at the A2 level to 96.0% correct agreement at the B1 level. These differences are non-significant. However, an interesting observation is that in plural contexts, the difference between the average agreement at the A2 and the B1 levels is large, 52.1% vs. 69.5%, and overall significant (X² = 66.112, df = 1, p < 0.001). If we look at the data in each individual L1 group, we note that the proportion of correct agreement in plural contexts differs significantly between proficiency levels in the ITA, GER and NLD groups (ITA, X2 = 28.034, df = 1, p < 0.001; GER, X2 = 25.745, df = 1, p < 0.001, NLD, X2 = 6.9432, df = 1, p < 0.01) but not in the Swedish group. These results indicate an interaction of proficiency level and L1 group. It is only in the Swedish-speaking learners that there is a lack of development between the A2 and B1 levels in plural contexts. It is striking that, even at the intermediate B1 level, many Swedish learners have trouble producing plural verb forms in plural contexts. This observation could indicate a developmental delay in this morphosyntactic domain for Swedish learners, to which we will come back in Section 6. The examples in (8) and (9) are typical interlanguage forms produced by the Swedish learners in plural contexts.

(8)	SUEB103	FdBsing
	Paul et Pauline euh ## /ɛ/ [*] parti pour la  fête    [target form: /s$\widetilde{ɔ}$/]
	Paul and Pauline euh  is left  for  the party
	“Paul and Pauline left for the party”
(9)	SUEB102	FdBsing
	et  donc Pauline et Paul  /il/  /ekut/ [*] la musique  [target form: /ilzekut/]
	and then Pauline and Paul they listen the music
	“and then Pauline and Paul listen to the music”     [omission of liaison]

Looking now at the production of SV agreement in number in the different agreement patterns, respectively,

Table 5. Total number of correctly agreeing verb forms in the three different agreement patterns Vont, Vrad and Vcons (proportions within brackets).

	ALL CONTEXTS (SG + PL)
	Vont	Vrad	Vcons
TOTAL	2007 (90.8)	328 (78.7)	627 (61.4)

presents the total results in the Vont, Vrad and Vcons patterns, whereas

Table 6. Number of correctly agreeing verb forms in singular contexts in the three different agreement patterns Vont, Vrad and Vcons (proportions within brackets).

SINGULAR CONTEXTS
	Vont			Vrad			Vcons
L1	Total	A2	B1	Total	A2	B1	Total	A2	B1
ITA	208 (100.0)	91 (100.0)	117 (100.0)	63 (90.0)	34 (89.5)	29 (90.6)	66 (83.5)	22 (81.5)	44 (84.6)
GER	137 (97.9)	72 (96.0)	65 (100.0)	36 (83.7)	18 (90.0)	18 (78.3)	38 (82.6)	20 (80.0)	18 (85.7)
NLD	332 (98.2)	86 (97.7)	246 (98.4)	68 (93.2)	19 (100.0)	49 (90.7)	96 (89.7)	30 (83.3)	66 (93.0)
SUE	441 (97.6)	245 (96.5)	196 (99.0)	104 (92.9)	49 (87.5)	55 (98.2)	116 (86.6)	56 (80.0)	60 (93.8)
Total	1118 (98.2)	494 (97.2)	624 (99.0)	271 (90.9)	120 (90.2)	151 (91.5)	316 (86.3)	128 (81.0)	188 (90.4)

and

Table 7. Number of correctly agreeing verb forms in plural contexts in the different agreement patterns Vont, Vrad, Vcons and Vuni (liaison) (proportions within brackets).

PLURAL CONTEXTS
	Vont			Vrad			Vcons			Vuni (Liaison)
L1	Total	A2	B1	Total	A2	B1	Total	A2	B1	Total	A2	B1
ITA	211 (88.3)	81 (79.4)	130 (94.9)	10 (34.5)	3 (27.2)	7 (38.8)	65 (38.7)	22 (28.9)	43 (46.7)	16 (19.5)	2 (4.2)	14 (40.0)
GER	138 (80.7)	76 (77.5)	62 (84.9)	14 (58.3)	8 (47.1)	6 (85.7)	61 (56.0)	20 (37.0)	41 (74.5)	9 (16.9)	5 (11.6)	4 (40.0)
NLD	258 (89.2)	55 (83.3)	203 (91.0)	16 (61.5)	1 (25.0)	15 (68.2)	108 (60.7)	13 (36.1)	95 (66.9)	8 (16.0)	1 (8.3)	7 (18.4)
SUE	282 (75.6)	140 (70.3)	142 (81.6)	17 (42.5)	8 (44.4)	9 (40.9)	77 (38.3)	29 (31.5)	48 (44.0)	16 (21.6)	10 (29.4)	6 (15.0)
Total	889 (82.9)	352 (75.7)	537 (88.5)	57 (47.9)	20 (40.0)	37 (53.6)	311 (47.4)	84 (32.6)	227 (57.0)	49 (18.9)	18 (13.2)	31 (25.2)

present the singular and plural contexts separately. The tables show the total number of correctly produced verb forms and the proportion of correct SV agreement in a given pattern. Recall that number agreement is silent in Vuni and therefore not included in Table 6 (singular). However, verbs from the Vuni pattern are included in Table 7 (plural) only when the verb is vowel-initial and preceded by a plural subject pronoun, which creates a context for preverbal liaison.

When we compare results overall in Table 5, there is a significant difference in the level of correct number agreement produced by the learners in different agreement patterns (X² = 757.9, df = 2, p < 0.001). We note that the Vont pattern (90.8%) stands out from the other two patterns as being most frequently produced correctly.

Table 6 displays the result in singular contexts, where we note that SV agreement in Vont verbs is close to 100% and very high for the other agreement patterns as well. The statistical analysis reveals a significant difference between verb patterns (X²-squared = 86.528, df = 2, p < 0.001) and again it is the Vont pattern that stands out as being more correctly produced than the other two patterns. Results are homogeneous across L1 groups. When we compare the more detailed A2–B1 results at L1 group level, we note an unexpected decrease in correct agreement of Vrad forms in the GER and the NLD groups.

Overall, the results in Table 7 show that SV agreement in plural contexts is most correct in the very frequent Vont verbs in all L1 groups. Within this pattern, there is an overall significant difference between L1 groups (X²-squared = 27.798, df = 3, p < 0.001), with specific differences between GER and ITA (X2-squared = 4.5256, df = 1, p < 0.05) and between SUE and NLD (X²-squared = 20.242, df = 1, p < 0.001). Moreover, Table 7 also indicates that the total level of agreement is comparable for the Vrad and Vcons patterns in plural contexts (47.9% vs. 47.4). This similarity is confirmed within L1 groups, even though the proportion of correctly agreeing forms differs between groups: 34.5–38.7% for ITA; 58.3–56% for GER; 61.5–60.7% for NLD and 42.5–38.3% for SUE. Investigating these differences between L1 groups further, we found a significant difference between groups in the Vcons pattern (X-squared = 27.56, df = 3, p < 0.001) but not in the Vrad pattern. This result could possibly be due to the very low total number of occurrences produced in this particular agreement pattern, which might affect the statistical analysis. Overall, this result seems to indicate that the verb patterns Vrad and Vcons are equally difficult for the L2 learners, even if they are not mastered to a similar level in the different L1 groups. Again, we see that the ITA and the SUE learners perform in a similar way, as do the GER and the NLD learners.

Finally, the general results for the production of liaison in Vuni verbs are low in all groups. However, we consider the total number of liaison contexts produced in this corpus too low to make a statistical analysis worthwhile. We note nevertheless that the comparison of the A2 and B1 levels confirms a lack of positive development within the SUE group for the production of liaison. In fact, we observe a decrease in the production of liaison from the A2 (29.4%) to the B1 level (15%) in the Swedish learners. This tendency is not reflected in the other groups, where we see an increase in the use of liaison between the A2 and the B1 levels. We believe that this agreement pattern will have to be studied in a more experimental setting in a future study in order to elicit a sufficient number of vowel-initial verbs.

6. Discussion and Conclusions

Our study has focused on the role of cross-linguistic influence (CLI) on L2 learners’ ability to use subject–verb agreement in number (third person singular vs. plural) in spoken French. This phenomenon is a well-known difficulty for L2 learners and previous studies have shown that they need a long time to integrate and automatize this morphosyntactic agreement (Ågren 2014; Bartning and Schlyter 2004; Howard 2006; Michot 2014; Véronique 2009). We wanted to compare the production of SV agreement in learner groups with different L1s in order to pinpoint a rarely mentioned factor in previous literature, namely CLI. More precisely, we wanted to know if the morphological richness of the learners’ L1, Italian, German, Dutch and Swedish, in this morphosyntactic domain, influences the acquisition process (Ringbom 2007; Xanthos et al. 2011). During the data collection, several factors were controlled for in order to evaluate a possible CLI. All participants are L2 learners of French in a university setting in their home country, they are matched for proficiency level (A2 vs. B1) and they perform the same oral narrative task (Paul and Pauline are having a party), which involves an alternation of singular and plural contexts in different agreement patterns.

Our comparison of the production of SV agreement in the four L2 groups reveals a significant difference between L1 groups for SV agreement in plural contexts—however, not in the way that we predicted. According to our analyses, the NLD group performed significantly better than the other three groups overall. Note that Dutch has a contrast relation to spoken French in terms of SV agreement (Ringbom 2007; Ringbom and Jarvis 2009) but its morphological richness in the verb paradigm is not particularly strong (cf. Table 1). According to predictions based on the importance of the morphological richness of the L1, we would have expected the Italian group, if any, to outperform the other groups. However, this was clearly not the case. On the contrary, the Italian learners performed at the level of the Swedish learners, who have an L1 that lacks SV agreement in person and number altogether and therefore has a zero relation to spoken French in this domain (Ringbom 2007). Here, it is important to mention that the learners in the NLD group live in the Dutch-speaking region of Belgium. Since French and Dutch are both national languages in this country, the learning of “the other language”, their first L2, starts early, already in primary school. Therefore, these learners started learning French at a younger age compared to the other groups (see Table 2, mean age 10.5 years) and they have been exposed to French in an educational setting for a longer period. Note, however, that the Dutch-speaking learners do not come from the bilingual regions close to Brussels and their extracurricular contact with French was controlled for when they were recruited. They reported very little informal exposure to French through television, newspapers, music, etc. Their earlier age of onset of acquisition of French has not resulted in a higher proficiency level at the time of testing, at least not according to the vocabulary test used as an independent measure of proficiency in this study (DIALANG n.d.). Nevertheless, we cannot exclude the possibility that the difference in age of onset and therefore also possibly in type of exposure to the target language inside and outside school may explain the differences observed between L1 groups in our empirical study rather than the morphological richness of the source languages themselves.

However, in relation to this finding, it is interesting to mention another difference observed between L1 groups in the data, which concerns the comparison of different proficiency levels (A2 vs. B1). In this respect, our analysis revealed a significant difference between A2 and B1 levels in L1 groups with a contrast relation to spoken French (ITA, GER and NLD) and a lack thereof in the Swedish group (zero relation to French). This result could be interpreted as a developmental delay in the Swedish group, which might, in turn, be an effect of the absence of SV agreement in person and number in the source language. The lack of SV agreement in Swedish could prolong the acquisition process of SV agreement in L2 French for these learners. In other words, as underlined by Ellis and Shintani (2014), a negative influence from a source language on the target language could result in slower L2 development. On the contrary, a positive effect of a contrast relation between source and target language, irrespective of the type of relation, could underlie an accelerated speed of L2 development, as observed in the data from our ITA, GER and NLD groups. At the same time, when it comes to the production of SV agreement in different verb patterns, including liaison, the performance of the different L1 groups is mainly similar, even if the development is slower in the Swedish learners. To this end, it is important to remember that the differences observed in developmental speed between groups in our corpus was only found when we compared the results of different proficiency levels within each L1 group. This finding could be an important indication for future studies. When looking for possible CLI effects, it is crucial to take into consideration the proficiency level of the learners within and across L1 groups in relation to the typological differences between source languages.

To conclude, this study highlights the complex interplay of different variables involved in the L2 acquisition process of SV agreement in spoken French. We have shown that CLI is one factor among others that needs to be taken into consideration, at least as far as the rate of L2 acquisition is concerned. Even so, as underlined already by Selinker (1972), the effects of CLI on L2 production are not always straightforward, as seen in the results of our Italian learners. The expected facilitating effect of a morphologically rich L1 could not be confirmed in this study, where the Italian group did not outperform their NLD and GER peers. However, during the detailed analysis of our rich corpus, we have noticed that there might in fact be more subtle effects of the learners’ L1 on their production of SV agreement in French than those presented in this quantitative analysis. This observation is being explored in a separate qualitative study of different types of learner errors, where we will continue the discussion of the possible effects of CLI on the acquisition process of morphosyntactic agreement patterns.