2.1 Current second language acquisition theory

White (Reference White2003a) argues convincingly that L2A is guided by UG, a position that we adopt here. However, even amongst researchers who believe that L2A is guided by UG, there is an ongoing debate as to whether or not parameters can be reset in L2A. Parametric variation, often related to uninterpretable features of functional categories, shows reflexes in both syntax and morphology. Consistently correct morphological inflection should be diagnostic of correct feature specification of the functional category in question. However, L2 learners frequently make morphological errors (cf. Prévost, Reference Prévost2007; Nadasdi, Reference Nadasdi2001; White, Reference White2003), something which must be accounted for in any L2 theory, as it forms part of the L2 developmental process. Two views of L2 parameter resetting are examined in this section.

2.2 Parametric differences between English and French TP

English and French differ not only in the morphology of their verbs, but also in the syntactic configurations in which the verbs are situated. Lasnik (Reference Lasnik and Lasnik1999) proposes that the lexical entry for main verbs in English is their bare form, with the bare form and T joined at Phonetic Form (PF) by ‘affix hopping.’ In contrast to bare English forms (3), French verbs start out fully inflected, with Tense specified as a feature (4).Footnote ⁶

1. (3) Weall/often/never read the newspaper.

2. (4) Nous (ne) lisonstous/souvent/jamais le journal.

As we see in (3), there is no overt agreement marker in English, whereas French displays the –ons ending for first person plural in present (4), as well as in other tenses. With the exception of present tense, which has no overt marker in French, tense is usually indicated by an additional morpheme such as [j] <i> in past lisions. These distinctions between the two languages reflect the well-accepted difference in morphology that is significant in determining the verb raising parameter (Emonds, Reference Emonds1978; Koeneman & Neeleman, Reference Koeneman and Neeleman2001; Pollock, Reference Pollock1989). French, with ‘richer’ morphology, requires raising of all finite verbs to T to check agreement, whereas in English only auxiliaries and modals raise.Footnote ⁷ A diagnostic for locating the position of the tensed verb is the placement of what Conradie (Reference Conradie2005) refers to as ‘left-edge markers’—negation, adverbs and quantifiers—assumed to be in a left adjoined position to VP in both English and French. In English they precede the main verb (3), but in French they follow the inflected verb (4), indicating that all verbs and auxiliaries move upward and leftward to T. In contrast, French thematic verbs without tense generally do not raise, as evidenced by the fact that they usually follow negation and adverbs (5).Footnote ⁸

1. (5) French unraised infinitives

   Ne pas/Souvent lire le journal serait un supplice

Another diagnostic of verb raising in French is the presence of nominative pronouns which must be checked for case by the inflectional projection, such as nous in (4). Thráinsson (Reference Thráinsson and Hendrick2003: 166) defines inflection and raising in terms of IP projections: morphological richness shows ‘clearly separable tense and agreement morphemes in the verbal inflection,’ his Split IP Parameter. [+SIP] languages have separate AgrSP and TP, while [-SIP] ones have unsplit IP. [+SIP] French requires verb raising since VP is not in the checking domain of AgrSP (due to intervening TP), whereas [-SIP] English can check the verb in situ since VP is in the checking domain of AgrSP (IP). In both English and French nominative case on the raised subject is checked by I/T.

In the bare phrase structure framework (Chomsky, Reference Chomsky and Kenstowicz2001, Reference Chomsky2002), two kinds of features are significant for a syntactic derivation, interpretable and uninterpretable. The former are necessary for semantic interpretation and persist at LF; the latter must be valued and deleted before Spell-Out. Uninterpretable features include, for example, nominative Case [uNom] on subject DPs (checked by a finite verb) or the uninterpretable person feature on AgrS that requires verb agreement, [upers]. In this framework, the verb raises in [+SIP] languages to check the uninterpretable feature on T, which is subsequently deleted by the interpretable [+/- past] of the inflected verb; the inflected verb then raises to AgrS to check that uninterpretable feature against interpretable person features. The raising of the subject results in deletion of its uninterpretable nominative Case feature in the match (Pesetsky & Torrego, Reference Pesetsky, Torrego and Kenstowicz2001).

We adopt a Distributed Morphology approach (Halle & Marantz, Reference Halle, Maranz, Hale and Keyser1993), whereby syntactic terminal nodes are separated from phonological realisation. Underspecified forms may enter agreement relations as long as they are non-distinct from the agreeing item (Lumsden, Reference Lumsden1992), and there is no other candidate that is more explicitly inflected (cf. Carstens, Reference Carstens2000). For example, in French, the first, second and third person singular forms of most verbs are non-distinct in having an identical spoken form (je/tu/il parle(s) [parl]), whereas second person plural always has a distinct spoken inflection [e] (vous parlez [parle]).

Summarising, main verbs in French are fully inflected, raised and checked for tense/person inflection in TP and AgrSP, while English main verbs do not raise (only auxiliaries are checked in TP). Uninterpretable features are crucial to both tense and person agreement in French, with interpretable features overtly marked on the verb valuing uninterpretable ones in T and AgrS.

2.3 L2 studies of verb raising

A solid body of L2 research has been devoted to the investigation of verb raising parameters, SIP, and V2 in French, German and Afrikaans (For overviews, see Hawkins, Reference Hawkins2001; Herschensohn, Reference Herschensohn2000; White, Reference White2003. See also Ayoun, Reference Ayoun1999, Reference Ayoun2003, Reference Ayoun and Ayoun2006; Conradie, Reference Conradie2005, Reference Conradie2006; Håkansson et al., Reference Håkansson, Pienemann and Sayehli2002; Hawkins et al., Reference Hawkins, Towell and Bazergui1993; Hulk, Reference Hulk1991; Myles, Reference Myles and Dewaele2005; Prévost, Reference Prévost, Prévost and Paradis2004; Schwartz & Sprouse, Reference Schwartz and Sprouse1996; Vainikka & Young-Scholten, Reference Vainikka and Young-Scholten1996, 2002; Walsh, Reference Walsh2005 inter alia). Many studies of verb raising (VR) in L2 French conclude that L2 learners eventually reset the VR parameter, although individual variation exists. For example, Hulk (Reference Hulk1991), in a cross sectional study of L1 Dutch-speaking learners of L2 French, finds that learners gradually reset VR, going through interlanguage stages disallowed in both L1 and L2 (cf. Håkansson et al., Reference Håkansson, Pienemann and Sayehli2002). In cross sectional studies, Hawkins et al. (Reference Hawkins, Towell and Bazergui1993), Herschensohn (Reference Herschensohn2000) and Walsh (Reference Walsh2005) observe graduated mastery of verb raising in L2 French, with differences noted in terms of the type of left edge marker involved, verb raising across negation and adverbs being mastered by advanced learners and verb raising across floating quantifiers not being mastered. Hawkins et al. (Reference Hawkins, Towell and Bazergui1993: 221) suggest a misanalysis account, but note that ‘subjects appeared to have made considerable progress in resetting the [+/–opacity] parameter.’ Herschensohn (Reference Herschensohn2001) examines temporal morphosyntax of two anglophone adolescent learners of L2 French. The intermediate level learners produce nonfinite verbs and other morphological errors, although their verbal syntax (e.g. postverbal negation and nominative subjects) is accurate by the last of three interviews. She concludes that the subjects’ morphological errors are therefore a result of missing inflection instead of a syntactic deficit. Conradie (Reference Conradie2005, Reference Conradie2006) looks at L2 acquisition of [+SIP] [+V2] in Afrikaans by speakers of English ([–SIP] [–V2]) and German ([+SIP] [+V2]); and of French ([+SIP] [–V2]) by speakers of Afrikaans. She concludes that the Afrikaans learners of L2 French, who had reset the V2 parameter, provide evidence for full access.

Overall, research in L2 acquisition of verb raising in French indicates eventual mastery of core syntax by Anglophones, but with some variation in morphological accuracy, and even proficient end-state speakers seem to have residual indeterminacy in certain properties such as mastery of quantifier placement in verb raising constructions. Previous L2 research leaves open the question of whether inflectional morphology is a crucial diagnostic of syntactic deficit (RD/I) or parameter resetting (FA). We now turn to our own data.

3.1 Subjects

The three subjects were interviewed before, during and after their stay abroad. Chloe, interviewed at age 22–23, had studied French for nine years, including two years in high school, before she spent six months immersion at age 16 in France (Herschensohn, Reference Herschensohn2001, Reference Herschensohn2003). Subsequently, she studied French at university level, spent four months in France at age 20, and then became an assistante d'anglais in the French overseas department of Réunion for nine months. Max began his French studies at age 48, independently completing the French in Action video program of first year French. By the time of the interview, he had studied French for twelve years on his own and with help from a native French tutor who met with him and his wife Eleanor one hour weekly for conversational exchanges over a period of 11 years prior to the interviews. Eleanor had studied for two years in high school, had a university minor in French and had spent two months with a family in France at age 28; moreover, she and Max spend two to three week vacations annually in France. Both Eleanor and Max do extensive reading, independent vocabulary/grammar study, audio listening and television viewing in French for 16–18 hours per week when at home. They also speak French to each other at dinner three nights a week. At the time of the interviews (August–September, December, April-May), Max and Eleanor were spending four months in Paris and three months in Lyon, where they had daily contact with French in a variety of contexts. Table 1 summarises these points.

Table 1. Subject information

Max and Eleanor's L2 learning experience outside the target environment consists of informal but very structured explicit training—independent grammar and vocabulary study, weekly tutorials, audio input through tapes and video, and crafted conversations—supplemented by annual stays in France. Strictly speaking, only their weekly native tutorial hour could be considered primary linguistic data (PLD), although for listening comprehension and speaking practice, their crafted acquisitional input options seem to serve the same function as PLD. Their seven months in France represented exposure to substantially more oral French spoken by a range of speakers in various kinds of situations; it represented more PLD than they had ever been exposed to in the same period of time. Earlier vacation sojourns did not entail the everyday commercial interactions required by life in a rented apartment during the described stay, and both Max and Eleanor independently established social networks with French speakers as part of their daily routines. For example, as an academic, Max did research (interacting with French scholars), while Eleanor had numerous organised activities with French speakers (e.g. neighbourhood tours, exercise classes). It is difficult to compare the quality of their social interactions to those of Chloe, whose age and social networking are quite different.

Chloe's experience with PLD was more extensive than Max and Eleanor's during the year under consideration and also in earlier periods when she lived in the target environment. While in Réunion, she shared a rental house with a French roommate and spent her workday dealing with French students. The group of friends with whom she socialised included many French speakers. Her written French is weaker (e.g. spelling errors, failure to write final consonants as in verb paradigms) than that of Max and Eleanor, but her spoken French is more rapid and more colloquial in its use of fillers and more appropriate in terms of register. Furthermore, she uses a broader range of verb tenses than Max and Eleanor.

The oral interviews, conducted by one of the authors, began with generalities about the here and now, moved to past and future narration, raised abstract topics (often cultural or political), engaged debate, and included a role play, such as introducing a keynote speaker or returning shoes to a shoe store. The interviews were transcribed and verified by a second linguist fluent in French.

3.2 TP morphology

We begin our presentation of the data by reviewing these advanced L2 learners’ morphology, specifically, examining their productive use of nonfinite and finite inflection. While only finite verbs in French undergo verb raising, it is important to look at both nonfinite and finite forms to verify that the learners understand the distinction. The three subjects use a range of tenses of regular and irregular verbs in their interviews (see Appendix 3). For example, in the third interview Max uses 89 different thematic verbs, Chloe 81 and Eleanor 72; they all use present, passé composé, imperfect, conditional and varying additional tenses. To compare these learners to less advanced ones who frequently use infinitival forms, we first consider finite inflection in opposition to infinitival forms (cf. Herschensohn, Reference Herschensohn2001, Prévost & White, Reference Prévost and White2000). We are not concerned with choice of tense or aspect, which is a problematic area for L2 learners (e.g. Howard, Reference Howard2006; Labeau & Larrivée, Reference Labeau and Larrivée2002) that we consider in a separate study. Therefore, we do not consider, for example, non-target uses of imperfect or pluperfect for passé composé, errors of lexical form, auxiliary choice or pragmatic errors. Instead, we focus on the accurate use of finiteness, since earlier studies of intermediate learners demonstrate the use of nonfinite forms in finite environments (cf. Herschensohn, Reference Herschensohn2001, Reference Herschensohn2003; Prévost, 2007; Prévost & White, Reference Prévost, White, Friedemann and Rizzi2000a, Reference Prévost and Whiteb), and since RD/I proposes that person/number (p/n) agreement (hence [upers]) should be unattainable by anglophone learners. Herschensohn (Reference Herschensohn2001, Reference Herschensohn2003) provides data from Chloe's verb production showing clear uses of nonfinite for finite forms in her early intermediate interlanguage. The subjects’ use of nonfinite verbs is presented in Table 2, where roman numerals indicate interview number and “TOC” represents the total number of contexts of obligatory nonfinite.

Table 2. Production, nonfinite verbs, Total Obligatory Contexts (TOC), Suppliance in Obligatory Context (SOC)

As can be seen in Table 2, the subjects make virtually no errors with nonfinite forms.

Table 3 indicates the L2 learners’ production of finite verb forms in the three interviews. TOC represents total number of contexts of obligatory finiteness (i.e. tensed forms as opposed to nonfinite forms), and p/n indicates accuracy of person-number marking on the verb. The learners’ errors (see Appendix 1 for a complete list) include use of default forms, incorrect suppletive forms and null subjects.

Table 3. Production, finite verbs and person/number (p/n) accuracy by interview

Unlike less advanced L2 learners, these subjects make almost no substitutions of nonfinite forms in finite contexts. One example of this rare error type is provided in (6) (an utterance from Interview III with Max). (The accurate target form follows the utterance in parentheses.)

1. (6) *pas problème, revenir mardi prochain (= revenez)

The learners make few errors in person-number agreement, the main type of error involving substitution of default third singular (3 sg) for third plural (3 pl) especially with suppletive verb forms as in (7).

1. (7) Examples of p/n mistakes

   1. a. Max, III

      *Deux profs m'a rencontré à la gare (= m'ont rencontré)

   2. b. Chloe, III

      * Ils prend [prã] un examen (= ils prennent/passent)

The errors of (6)–(7) are not ‘grammatical with respect to the learners’ systems,’ as one reviewer suggests; rather, they are clearly performance problems that stand out as mistakes among hundreds of tokens of correct inflection. These errors pattern similarly (but not identically) to what was found in Howard (Reference Howard2006), in that most (20/30) involve substitution of 3sg for 3pl in suppletive verbs. These irregular verbs have distinct spoken forms for third person singular and plural and are always distinguished by native speakers. It is important to note that the vast majority of our L2 subjects’ errors involve production of analogical forms based on the dominant morphological pattern in French. In Modern spoken French, verb forms for all persons except the first and second plural are homophonous in -er verbs, the largest morphological verb class.

Four of Chloe's errors involve relative qui, which we believe she has either analysed as singular in all contexts (e.g., *les gens qui devient for les gens qui deviennent) or has difficulty with because of long distance agreement (cf. Howard, Reference Howard2006). Two other error types involve null subjects and overregularisation, error types also observed in L1A (Clahsen et al., Reference Clahsen, Aveledo and Roca2002). Three of the four null subjects accompany impersonal verbs (e.g. [ce n’] était pas la peine, Eleanor III). The regularisation of irregular verb forms is in line with the dominant conjugation patterns in French as, for example, Chloe's use of ils prendent [prãd] (cf. ils rendent) for ils prennent. The overall accuracy rate with respect to p/n inflection is almost at ceiling, nearly 99% for all three. We next consider another aspect of French verbal morphosyntax, namely verb raising.

3.3 Verb raising

In Table 4, using the placement of the verb relative to left edge markers (negation, adverbs and quantifiers) as a diagnostic for verb raising, we see that the three subjects have a high rate of accuracy for verb raising (VR), oral production data that is corroborated by their grammaticality judgments and a written production task.Footnote ⁹

Table 4. Verb Raising (VR) oral production with neg/adv/quant diagnostics, by interview; Grammaticality Judgment (GJ) accuracy

The GJ task is adapted from Hawkins et al. (Reference Hawkins, Towell and Bazergui1993) and includes 40 sentences related to verb raising (plus distracters)—half grammatical, half ungrammatical—with negations (pas, jamais, rien), adverbs (e.g. rapidement, prudemment, souvent) and tout floated from the subject (8). This adapted GJ task (and the original one) had been tested with native French speakers who judged the target interpretations as expected at 100% (Hawkins et al., Reference Hawkins, Towell and Bazergui1993; Herschensohn, Reference Herschensohn2000). The test included only finite verbs.

1. (8) Sample GJ sentences

   1. a. *Marie souvent va au cinéma.

   2. b. Elles ont traversé rapidement la rue.

The learners were asked to correct all ungrammatical sentences. Table 4 presents the number of verbs preceding left edge markers (and, hence, taken to have raised to TP/AgrS) in the oral interviews (cf. Appendix 1, 2).

As shown in Table 4, in virtually all contexts where diagnostics of negation, adverbs and quantifiers appear in the interview, Max and Chloe raised verbs with near perfect accuracy. Indeed, all subjects produced several tokens of negation, adverbs and quantifier placement postverbally, which serve as clear diagnostics for VR as exemplified by (9)-(10).

1. (9) (9) Max, III

   quelqu'un qui parle bien anglais

2. (10) Eleanor, III

   Il me reste encore le piano

The subjects made only five errors with respect to verb raising in oral production (four of them in the first interviews). Eleanor made three of the errors. An example of one of her errors is provided in (11), where she failed to raise the verb over the modifying adverb. Max and Chloe each made one error only. Max, for example, raised the infinitive over the adverb (see (12), while Chloe failed to raise the verb across aussi (13).

1. (11) *Sa fille maintenant travaille (= travaille maintenant)

2. (12) *La possibilité de manger bien (= de bien manger)

3. (13) *qui aussi traversaient (= traversaient aussi)

With respect to grammaticality judgment, GJ aggregate numbers reported in Table 4 mask the fact that all but one of the GJ errors involves the quantifier tout. In Table 5 we see scores on the GJ tasks by diagnostic.

Table 5. Accuracy of GJs, Max, Eleanor, Chloe, left-edge markers

Over three interviews, 16 of the 17 errors involve the following grammatical sentence type presented in (14a), which is grammatical only under the reading that tous modifies ils, but for which all three subjects repeatedly corrected the sentence so that tout modified (with correct gender) the direct object following the verb (14b):

1. (14) tous modifying subject pronoun ils [tus]

   1. a. Ils relisent tous [_DP leurs notes] [tus]

      tous modifying NP [tu]

   2. b. Elles relisent [_DP tous leurs livres] [tu]

In other words, the learners’ first interpretation was sentence type (14b); they rejected (14a). Since they correctly placed/judged tout in other contexts, at issue here seems to be that their preferred interpretation of this sentence type was not native-like, instead of a VR error per se. The fact that these errors almost exclusively involve quantifiers, mirrors a pattern previously observed by Hawkins et al. (Reference Hawkins, Towell and Bazergui1993) and Herschensohn (Reference Herschensohn2000), as noted above. The three subjects do not show native-like interpretation of the lexical item tout, but their placement of verbs with respect to negation and adverbs is target-like. We agree with a reviewer's suggestion that they may have misanalysed the data, for example, assigning the incorrect phrase structure for the quantifier's referent. The higher percentage of quantifier errors in Interview III suggests that their command of tout is variable and does not improve over term.

In summary, all three learners’ nearly perfect accuracy scores for verb placement relative to left-edge markers after the first interview reveals that they have mastered the correct setting for verb raising, while their indeterminacy on GJs indicates a specific problem with the lexical item tout.

The written production tasks involved composing 30 sentences with negation (pas, jamais) and adverbs (quality adverbs bien, mal and frequency adverbs souvent, rarement) from subjects and verbs that were provided. For example, three agents (e.g. Paul, je, vous) with three adverbial phrases involving frequency (e.g. 0, 7, 2/week) were provided for boire du lait; the task involved writing one sentence for each agent. Max and Eleanor (90 tokens each) scored 100% on the three tasks; Chloe, who only completed the task in one of the three interviews, had much lower accuracy (20/30 sentences): two omitted, two ‘corrected,’ six wrong and 20 correct). The low score was due in large part to the fact that she ‘corrected’ (crossed out and rewrote) her original placement of the frequency adverbs souvent and rarement. (15a) to place them all sentence finally as in (15b).Footnote ¹⁰ In contrast, she ordered all the quality adverbs post-verbally (15c).

1. (15) Examples, VR written production, Chloe, III

   1. a. Jeanne boit souvent du jus d'orange. (crossed out)

   2. b. *Jeanne boit du jus d'orange souvent. (inaccurate corrected token)

   3. c. Marc, il lit mal ce poème.

Once again we see a difference related to specific lexical items. In her oral production, Chloe consistently uses correct post-verbal placement of adverbs, a difference between oral and written production that has been noted (Bialystok, Reference Bialystok1997). In this case, the additional reflection devoted to the written task seems to have led Chloe to overcorrect. The final set of results we consider in this section concern longitudinal development.

3.4 Input, short-term and long-term L2 development

Extensive studies (within different research paradigms) growing out of the European Science Foundation studies of naturalistic L2 learning in the 1980s and 1990s (cf. Klein & Perdue, Reference Klein and Perdue1992, Reference Klein and Perdue1997) often considered naturalistic input to be the crucial component of L2A, but research has also highlighted the role of additional factors such as learning strategies, feedback and explicit instruction (Carroll, Reference Carroll2001; Doughty & Long, Reference Doughty and Long2003; Herschensohn, Reference Herschensohn2003; Truscott & Sharwood Smith,Reference Truscott and Sharwood Smith2004).

With respect to French verbal morphosyntax (finiteness, verb raising), our three subjects do not appear to show dramatic progression in temporal morphosyntax over the immersion period, since their accuracy levels are nearly at ceiling for all three interviews. Clearly, the immersion experience has not played a significant role here since the subjects are already stabilised in their ability to use finite forms and correct verbal word order. Likewise, we see no clear development with respect to p/n accuracy, either, as their production is fairly stable in all interviews. Max and Eleanor hover around 99%, while Chloe ranges from 96% to 98% (Table 3). Finally, VR data also show only slight variability, not longitudinal progression (Table 4).

In contrast to the near perfect accuracy of inflection and word order, the GJ data does not show improved interpretation of tout over the period of time investigated here for any of the subjects. They persist in the misanalysed phrase structure, eschewing the interpretation of the quantifier as floated from the subject. Furthermore, their production in other aspects of French (beyond the scope of this study) such as verbal aspect is more flawed. For example, Max, when asked what he habitually did on a daily basis (cued by imperfect with imperfect response expected), responded with multiple sentences in the inappropriate passé composé.

Indeed, the fact that these learners are at ceiling for verbal morphosyntax does not, however, mean that they are native-like in their overall production. Putting aside their non-native phonology and slow rate of speech, their nominal morphosyntax (gender and number concord between nouns, determiners and adjectives), for example, is far less accurate than is their verbal morphosyntax (Herschensohn & Arteaga-Capen, Reference Herschensohn and Arteaga-Capen2007); furthermore, they have a limited range of tenses and moods. As is often the case for second language learners, they avoid what they do not easily produce (cf. Gass & Selinker, Reference Gass and Selinker2001), rather choosing to say what they know how to say, resulting in performance that may be accurate but is not native-like. The important point here is that they clearly have well established the French morphosyntactic values of TP and AgrSP—that is, they have reset the verb raising parameter—despite very obvious weaknesses in other aspects of their grammars and lexicons.

In terms of long-term longitudinal development, if we compare Chloe's performance in the interviews reported here to her performance in earlier studies (Herschensohn, Reference Herschensohn2001, Reference Herschensohn2003) when she was 17 years old, we see little change in accuracy rate from the final interview conducted at age 17 (accuracy rate 98% after 6 months spent in a totally francophone environment in France) to the interviews reported here (likewise 98%).Footnote ¹¹ Once again, the high score obscures differences between age 17 and age 22 with respect to her mastery of French. Although Chloe appears to have stabilised at around 98% verbal accuracy, she has a far greater command of vocabulary and varied tenses in the later interviews than in the earlier ones, so a simple comparison of the percentages is misleading. While p/n agreement does not seem to change in Chloe's long-term development, her command of discourse implementation and a much broader range of tenses does mark a difference in her performance from her earlier interviews. Whereas her final interview at age 17 contained 91 accurate verbal tokens; her final interview at age 23 contained 387 tokens and demonstrated her ability to discuss, narrate and debate extended discourse in the past, present, and future tenses. It appears from the data that verbal morphosyntax can be solidified in L2 French earlier than other grammatical phenomena such as clitic pronouns (cf. Herschensohn, Reference Herschensohn, Prévost and Paradis2004).Footnote ¹² Lexical or semantic details—such as infrequent forms of irregular verbs or native-like interpretation of quantifier float—may continue to be variable even at a more advanced stage.

We do not mean to claim that immersion had no effect on our L2 learners; for example, there is a clear improvement with respect to pragmatic and lexical factors, fluency, colloquial use of language, breadth of vocabulary, and range of verb tenses among other factors. However, given the high level of accuracy at the start of the study, there was little opportunity to show further development on an already acquired feature.

In summary, Max, Eleanor and Chloe generally inflect all verbs for tense and p/n, using all persons, with nearly perfect accuracy. As for verb raising, their accuracy rate is generally 97–100% in oral production. Their written production further indicates that they have a very strong grasp of VR with negation and adverbs, and their GJ results confirm this. Their GJ scores also indicate that they have persistent difficulty with the use of quantifier float with the adverb tout, a characteristic of advanced learners that has been noted in previous studies (Hawkins et al., Reference Hawkins, Towell and Bazergui1993, Herschensohn, Reference Herschensohn2000).