Can we repeat what we do not say in L2?

The paper analyzes the results of a sentence repetition task performed by Lithuanian L2 speakers of different language proficiency levels. This paper focuses on a set of targeted verb forms included in the task because they are less likely to occur in free production. The different forms the speakers produced are analyzed by considering their correspondence (or not) to the targeted form, by comparing the speakers among each other with respect to their learner variety and by comparing the task results to the repertoire of the same speaker in the free production data. In multiple cases of failed re-production, the analysis examines what the targeted forms were substituted with and tries to identify possible reasons for such substitution. This analysis confirms the general inability of the basic variety speakers to distinguish between morphologically different forms and the dominant trend showing that L2 learners are more meaning-focused than form-focused. The analysis also shows growing implicit knowledge, or at least gradual passive acquisition, of the less frequent forms as the speaker advances in the post-basic continuum.


Introduction
Acquiring a second language (L2) is a complex process, which involves a variety of factors influencing it and a diversity of aspects at all language levels: phonological, morphological, syntactic and semantic. A person acquiring a second language is faced with all the complexity at once; however, the human brain is able to learn to "dissect" the overwhelming amount of input, make use of what is comprehensible at a given moment and discard the rest or store it for the future. This capacity, using Klein's (2009) terminology, can be considered as "construction faculty" and the "copying faculty" working to serve the "communication faculty".
When considering communication, speaking is the primary and most important means. Naturally, in case of modern languages, the graphical representation of the target language quickly becomes necessary as a person living in the environment of the language is almost constantly faced with writing in that language in such media as newspapers, TV, internet, and social media. All this constitutes an input and can have a positive effect on the acquisition. Nevertheless, the graphic representation implies substantially more "learned" knowledge, such as alphabetic representation of sounds as well as differences between this representation in the target (L2) and the source (L1) language. Therefore, analyzing oral speech is here considered of primary importance in order to study the underlying language acquisition mechanisms.
This paper analyzes data of elicited imitation (sentence repetition) task, conducted on Lithuanian L2 adults living in the target language environment for a varying amount of time. The focus is drawn to verb forms, and the analysis examines whether they are repeated, omitted or modified. The aims are to discern whether the pattern of acquisition in the previously studied free production data from the same speakers is mirrored in their implicit knowledge; to identify whether there are more problematic forms that require greater attention; to learn about the receptive competence of the speaker based on the form modification and gain other possible insights into the general acquisition process.

Theoretical background
The learner-varieties approach to L2 acquisition sees the adult second language learner as an experienced communicator who, even at the very beginning, uses the general semantic and pragmatic principles to convey their message (Klein & Perdue 1997;Perdue 1993;Klein & Dimroth 2009). The learner is driven by the need to communicate and therefore draws on all possible resources in a constructive way and not merely imitates the native's speech. Therefore, the study of spontaneous speech by, preferably, untutored learners can give some insight into the principles of how human mind copes with linguistic systems.
What seems as a mainly "copying" task, elicited imitation (EI) or sentence repetition is in fact a rather complex action involving sentences that exceed short-term memory and thus do not allow "parroting" them (Slobin & Welsh 1973). Therefore, this task is a widely used procedure for language assessment in second language research. While testing adults for assessment, scholars have found that absence of a morphosyntactic structure of the sentence hinders the recall to a greater extent than the semantic implausibility and that prosody has beneficial effects (Miller & Isard 1963;O'Connell et al. 1968;Polišenská et al. 2015). The EI task usually takes place in controlled settings and targets specific grammatical features, often includes ungrammatical sentences and additional secondary-level tasks (Bley-Vroman & Chaudron 1994;DeKeyser 2003;Ellis 2005;Erlam 2006;Wu & Ortega 2013;Sarandi 2015) and offers insights into other aspects (such as working memory, e.g. Baddeley 2000).
Turning back to the learner-varieties perspective, already Klein (1986: 71-74) noticed the interplay of factors that affect the correct repetition or absence of items and the role of learner varieties. In learner varieties that have not yet incorporated a determinate grammatical feature, that item, especially in other than the initial and the final positions of the utterance, will not be reproduced. Table 1 briefly summarizes the main features of learner varieties. The entire language acquisition process is a continuum; however, the first very significant qualitative step occurs when the learner's interlanguage starts to incorporate verbs. This allows the speaker to construct structured phrases and, therefore, be rather efficient in communication. For this reason, it was called the basic variety. The basic variety still has no morphology, usually the most frequent form in the input is used of a given word and carries only the lexical meaning. Another important qualitative step in the process of acquisition occurs (however, some learners never go beyond the basic variety) when a learner starts to notice the existence of different morphological forms and to associate them with different functions. This is true for L2 production; nonetheless, receptive skills can often exceed the production by far, and this is partly what the specific task analyzed in this paper was aimed at.  Pre-basic basic Post-basic * NP = noun phrase, PP = prepositional phrase, V = verb, Cop = copula, Adj = adjective, Adv = adverb, PART = particle.
The inclusion of the sentence repetition task (as the one analyzed in this paper) into the design of a larger study was inspired by the VILLA project (Varieties of Initial Learners in Language Acquisition; see Rast 2008;Dimroth et al. 2013). In this project learners of various L1s were presented with controlled input of a formerly unknown L2 (Polish). A large part of the input was presented as oral stimuli with the task of repeating it (as well as other tasks). This elicited (and other) production of each learner was later analyzed. The project resulted in many important discoveries, one of which was Dimroth's observation of learners not only acquiring some morphological features of the TL but also of some idi-osyncratic processes, which included individual creative constructions and reinterpretation of morphological variation (Dimroth 2018). Therefore, beginning learners not only copied the input but actually started constructing from it and other available means.
In the context of the same VILLA project, related research on morphosyntactic properties of verbal inflection by the grammaticality judgment test was carried out by Latos (2014). Latos tested for 1 st and 3 rd singular verb forms and concluded that the experiment participants were highly sensitive to morphological forms. The group that focused on the meaning scored 70%, and the group that had a formfocused input scored 78%. Even though the author strives to show the beneficial effects of the focuson-form input, the interesting find is that the difference is relatively small. This seems to support the idea that the "copying faculty" is not the only one involved but is only a part of a much larger system.

Task design and data collection
The data analyzed in this paper includes only a relatively small part of a larger data set collected to study the acquisition of Lithuanian as a second language in a natural setting. The entire study was cross-sectional (data collected from speakers of different proficiency levels at one point in time) and consisted of a semi-structured socio-linguistic interview and some specific tasks (for more detailed description of the method, see Ramonaitė 2015c and for the analysis, see Ramonaitė 2015a;2015b;2017a;2017b).
The entire study for which the data was collected was designed to be as communicatively natural as possible. In order to collect free spontaneous speech, the free production of the participants was not corrected, but help to express single constructions was provided if asked for. Therefore, there was much more emphasis on the meaning than on the form. The other tasks used in the study were meaningfocused (e.g. film retelling and explaining a recipe). For this reason, the task analyzed in this paper, the quasi-experimental sentence repetition task, was the only one explicitly focusing on the form. It was conducted at the very end of the second meeting with each participant as not to have an effect on the rest of the free production during the conversation.
The EI task consisted of repetition of seventeen sentences. The sentences were previously recorded by the same researcher and played one by one, pausing the registration after each sentence and having the participant repeat it. Before the start of the task, the participant was told that they would hear recorded Lithuanian sentences, some of which might be short and others long, and they would be asked to repeat what they heard. However, they were only asked to be attentive but were not explicitly encouraged to concentrate on the form. The primary goal of the task was to learn about the ability of identifying and repeating the verb forms that were less likely to occur in free speech, such as feminine, plural, the frequentative past tense, different participles and compound tenses. The lexemes were high frequency words, presumably known to the participants, thus allowing the analysis of morphology. The collected data will be available for study of acquisition of other categories of Lithuanian L2 for future research and might present some interesting insights. The sentences used for this task with target form glosses and translation can be found in the Appendix.
Given the natural background of the entire study, the sentences proposed were pronounced naturally, at medium-speed, in clear pronunciation, tended towards the standard, were grammatically correct and had a complete meaning. The mean-length of sentences was 13 syllables, but the set also included one very long sentence (28 syllables) and three rather short ones (8 and 9 syllables). The short sentences were placed between the longer ones and were intended as a motivation for the weaker participants.
The sentences attained to pro-drop where possible; also, frequent proper names or other references (e.g. su broliu 'with brother' in sentence 13 in the Appendix) were used to indicate the subject of the sentence.

Participants
The participants of the study were foreigners who at the time of the study were living in Lithuania and had plans to live there at least in the nearest future (i.e. not in exchange programs). Table 2 presents the information about the sixteen research participants. The participants are listed according to their learner variety and sequence in the post-basic continuum, as determined in previous works (Ramonaitė 2015a). The names are pseudonyms, in the transcriptions the speaker is marked with the first three letters of the name, e.g.: Agata = AGA. ** Hours of formal instruction of Lithuanian L2, calculated approximately according to the information provided by the participants. *** Employee refers to employees of international companies that have their branches or large offices in Lithuania.
Half of the participants live in Vilnius city/region, and the other half in Klaipėda city/region in Lithuania. As indicated in Table 2, most of the participants (except for Šarūnas) have had at least some hours of formal instruction of Lithuanian as a foreign language. The students of one English-medium higher-education institution have an obligatory course of one semester and can freely choose whether to continue the classes or not after that. The adults who came to live in Lithuania for work or marriage have often started attending a Lithuanian L2 course but quit for various reasons or just had a short/ introductory course. However, some participants -Juan, Eunok, Nicolas and Elena -have attended an intensive course at a university (they attended the same course but in different periods; however, the input is definitely comparable). Nevertheless, considering the length of stay in the country and the amount of formal teaching (if any), all of the participants have had a lot more exposure to the language from the environment than from L2 classes.

Form analysis
The elicited imitation task involved meaningful sentences that incorporated different targeted verb forms that were less likely to appear in free speech (see the Appendix). There were some forms of the simple tenses (present, past, frequentative past and future), imperative and conditional mood, the three different participles and some forms of compound tenses. In order to have natural sounding sentences, some other, previously non-targeted, forms were included as well, such as present tense, past tense or infinitive forms. It should also be noted that this task was designed before collecting and analyzing Lithuanian L2 data; therefore, at that point there was no way to know which forms or specific lexemes should be avoided as early route-learned or for other reasons. Figure 1 presents the general view of the task data: exact or very close repetition of the targeted items was considered "positive", whereas the replacements with other forms or invented forms were considered "negative". The forms participants failed to repeat were marked "0"; therefore, the size of the column also shows the ability to repeat. In the following sections focusing on different grammatical categories, more detailed information will be provided.

Present simple tense
Only three (of total six) present tense forms were intentionally targeted in the task. It was the 2SG, 3ref (containing a reflexive particle) and 2PL. The 2SG form in the task was turi 'you have (to)'. This form (infinitive turėti 'to have (to)') was used in the task because it differs from the 3 rd person form of the same lexeme only in the syllable stress: turì 2SG and tùri 3SG. The results in Table 3 show that two of the basic variety participants (JUA and ANG) reproduced a non-target form or did not reproduce it at all. One participant (VIC) modified the form into a conditional of the 3 rd person. Although most participants reproduced the form without much difficulty, four of them, three being the most advanced participants, modified the form into the 1SG, i.e. turiù. This particular form, although rather frequent in the input, is overextended by tùri for very long in the post-basic continuum (Ramonaitė forthcoming). It is therefore plausible to attribute this modification to the increased awareness of the form with the stressed final syllable. The participant who is at the very beginning of the post-basic continuum (SUS) did not use the turì 2SG form in her free production, whereas the three most advanced speakers did use it but significantly less frequently than the 1SG form. The fact that none of the participants modified this form into the expected 3 rd person form is of interest. On the surface it would seem to show that learners of the post-basic continuum have already acquired the two differently stressed forms; however, the overextension of the 3 rd person form in free production and the 2 Here and in the following sections: + form reproduced correctly; ~ form reproduced with minor (phonetic) changes; ? form resembling but not corresponding to any target language forms; form not reproduced. In cases of form modification, the actually produced form is indicated. For affixed forms (prefixes, reflexive particles, negative particles), if the affix was omitted but the verb form remained unchanged, a "-" is indicated in superscript. The "l" in superscript indicates change in the lexeme but not the form.
modification into PRES1SG in the repetition task by the advanced speakers indicate the situation might not be as simple and would require further investigation.
The 2PL form galite 'you (plural) can' was reproduced correctly by all the participants, except for one speaker of the basic variety ANG and beginning post-basic SUS who modified it into their respective base forms (3 rd person form and participle galima). The 3 rd person form with the reflexive particle (the last of the three 3 rd person forms in Table 3) was prausiasi 'one washes oneself'. The lexeme is a typical beginner reflexive word and was placed at the very end of the sentence to facilitate focus. It was reproduced correctly when produced; however, six participants did not produce it at all, one reproduced it mechanically, and later stated they did not understand the meaning. The form was correctly reproduced by the advanced speakers, starting from GAN, and, surprisingly, by the pre-basic variety speaker DIE.

Past simple tense
Of the six past simple tense forms that occurred in the EI task, two were targeted for the morphological person form (girdėjau PAST1SG of girdėti 'to hear' and buvome PAST1PL of būti 'to be'), two were 3 rd person forms with the difference that one is a suffixed stem (važiavo from važiuoti 'to go/drive'), and the other one is without a suffix (grįžo from grįžti 'to return'), one was a prefixed form with the reflexive particle (susitiko from susitikti 'to meet'), and the last one was a form of the verb būti 'to be', used as a copula, which in Lithuanian is optional in the present tense but obligatory in the past.   Table 4 shows a clear difference between the pre-basic and basic variety speakers and the rest. These beginning speakers, however, are also not uniform. The only pre-basic variety participant DIE, whose positioning in the pre-basic variety was previously determined from the free production, tends to behave similarly to the post-basic competence on this passive knowledge level of Lithuanian. This is most likely due to his 10 year-long residence in Lithuania. The two basic variety speakers JUA and ANG, as well as SAR, who is "on-the-verge" towards the post-basic variety, show a rather indistinct variation between the different forms. They used present, past, future, infinitive and conditional forms or omitted the form in the repetition altogether. This is typical behavior in the basic variety that does not have inflectional morphology (Klein & Perdue 1997) and when morphologically different forms are used in free variation, i.e. without the specific morphological function that the form has in the target language. In fact, based on the free production of these speakers (Ramonaitė 2015b), they do not distinguish between the different morphological forms but use mostly the base form to express a certain lexeme 3 . The base form is most often the PRES3 or INF. It is worth noticing that in this repetition task in some cases, these speakers produced the PAST3, although overextended (i.e. used for other persons as well). This could prompt to the PAST3 form being more salient and thus used as a kind of 'baseform' of the tense (as will be argued further on).
A different behavior is exhibited by the SUS who has just passed over to the post-basic variety and is very conscious of different forms having different functions; she has just not yet acquired all of them. This seems to be reflected in her non-reproduction of the other (1SG and 1PL) forms but correct reproduction in most PAST3 forms.
Concerning the repetition of past forms by the post-basic variety speakers, it is very important to notice that there are quite a few overextensions, but all of them are overextensions of the PAST3 form to other forms. In fact, the use of PAST3 instead of PAST1SG is registered most frequently even in the group of advanced speakers. This is in line with the findings related to the sequence of acquisition of Lithuanian L2 person markers (Ramonaitė forthcoming): the marker -o is acquired rather late due to the overlap in the present and past tense paradigms and is thus long overextended. That is to say, forms like girdėjo (3 rd person) are used instead of girdėjau (1 st person singular) even by very advanced speakers.
Unfortunately, the results of the repetition task do not shed more light onto the hypothesis expressed so far. When analyzing acquisition of person markers, it was hypothesized that the natives' realization of the final vowel/diphthong lacks saliency and thus obstructs acquisition in L2. However, the diphthong in the ending -au in the repetition task was pronounced clearly but was still not reproduced by many participants. The morphological verb form was the only indication of the subject in this sentence; therefore, further testing is necessary to study this problematic form in more detail.
The overextension of PAST3 onto 1PL is less frequent and occurs only in the pre-basic, basic and speakers in the beginning of the post-basic continuum. The indication of plural subjects in the sentence was evidenced by su broliu 'with brother'. This could be another indicator that the PAST3 form is acquired first in the past tense and is used as a certain base form of the tense.
The suffixed and non-suffixed stems do not seem to be treated differently; both are successfully reproduced. The same can be said about the reflexive particle, which was nearly never omitted. Interestingly, three post-basic variety speakers omitted PAST3 used as a copula, which is obligatory in the past in Lithuanian.

3
For instance, forms einu, eini, eina in the target language have different morphological meanings to indicate tense (present) and person (1SG, 2SG and 3 rd respectively), however, in the basic variety, this morphological function of the form is ignored and these forms can be used interchangeably only for their lexical meaning 'to go'. At this stage these additional functions are carried out by other means, such as using an overt pronoun to express person or an adverb to express tense, e.g. an utterance tu vakar eina [you yesterday go:PRES3] would be used to say 'yesterday you went'.

Future tense
The two future tense forms included in the task show a clear correspondence to the sequence of acquisition (Ramonaitė 2015a), i.e. the future tense forms are naturally acquired later, only after the present and the past forms. The forms were not reproduced by any of the beginning speakers. Only SAR, who incorporates the form bus 'will be' in his variety and makes abundant use of it in his free production, manages to reproduce it during the task as well. Further into the post-basic continuum, these forms do not seem to be difficult to reproduce; see Table 5.

Frequentative past tense
The frequentative past tense is a rather simple and transparent tense that is, however, rather infrequent and therefore acquired late. This is reflected in the repetition task as six beginning speakers up to GED do not even attempt to reproduce it. The more advanced speakers who did reproduce the form show a clear preference towards FPAST3 overextending 1SG. Given the morphological analogy, the reasons for this are likely to be the same as those for overextending PAST3. Speakers GED, ALE, AGA, and EUN do not use the FPAST1SG form in their free production. Speakers GAN, MAD, DIN, NIC, ELE, and ANN do use it although the data is insufficient to confirm acquisition.
Considering the failure to reproduce the elicited sentence, further investigation into the distinction between the forms in -au and in -o would be necessary.

Imperative
The results of the only imperative form pasakyk 'say/tell' included in the task present an interesting insight. All but the basic variety speakers reproduce it correctly, while all the basic variety speakers substitute it with the infinitive form (pasakyti 'to say'); see Table 7.
Although the acquisition of the infinitive itself is not so straightforward (Ramonaitė 2017b), the results of this task show that the form of the imperative permits the recognition of the infinitive stem and thus does not obstruct comprehension. The correct reproduction by DIE is again attributable to his long permanence in the country; in addition, his two Lithuanian native children probably pro-   vide input in family communication where the imperative form is particularly frequent. The imperative forms, even if they do occur sporadically in the basic variety speakers' free production, are not used with the imperative function. Curiously, SAR does use imperative more and apparently purposefully but fails to reproduce the same form in the repetition task.

Conditional
Of the two conditional forms targeted in the task, one cannot be used for the purposes of this analysis due to the choice of the lexeme. It was not known in advance, but the specific form norėčiau 'I would like' is route-learned early as a polite request and is present in all speakers' data. The other form atsineštų 'he/ she/they would bring (along)', however, confirms that the conditional in the learner variety appears a lot later; see Table 8. None of the speakers of basic or the beginning of the post-basic continuum variety reproduce it. Interestingly, SAR and VIC do use this form in their free production but only with one other lexeme: SAR būtų from būti 'to be' and VIC galėtų from galėti 'to be able to'. The incorrect reproduction of DIN could be attributed to mispronunciation, as his produced form is very similar (atsinešu 'I bring (along)'); however, it is likely that he misinterpreted the meaning and actually intended to utter the PRES1SG form.
The missing reflexive marker in case of three post-basic variety speakers (EUN, NIC and ANN who produced atneštų 'he/she/ they would bring') indicates that the reflexive particle is an ulterior complication, and since this verb form is already complex, the particle is not dealt with and is left aside. The prefixation (prefix at-), however, does not seem to cause difficulty.

Infinitive
Due to the focus on flectional morphology the infinitive form itself was not a target; however, three forms of different complexity occurred in the natural meaningful sentences. One was prefixed nueiti 'to go (away or to a certain place)' (perfective meaning), one was prefixed and contained a reflexive marker pasikviesti 'to invite someone to one's place' and a third one without any additional markers žiūrėti 'to watch'.
The beginner speakers and VIC failed to reproduce the prefixed form. Moreover, one basic variety speaker (ANG) evidently noticed the prefix but replaced it with the negative particle (ne). This shows the speaker recognizes the lexeme but does not have (yet) the notion of prefixes, thus the only acceptable situation in the system of his variety where a marker is in front of the verb is a negation particle. The elicited form nueiti and  The form pasikviesti with both a prefix (pa-) and the reflexive particle (-si-) was reproduced also approximately in form by the basic variety speakers, although with an invented and meaningless form (JUA: *pasikiki, ANG: *pasikieti). The plain infinitive form did not seem to cause problems for anyone except SUS, who substituted it with PAST2SG and made other changes in the sentence completely losing the meaning.

Participles
Three different participles were included in the task: the genitive form of the masculine past participle (the nominative occurs in compound tenses) nukritusių 'fallen', feminine of the half-participle eidama '~while going', and the gerund lyjant '~raining' (given the rare occurrence of this form in natives' speech, the most typical construction lietui lyjant was chosen). Interestingly, the repetition task results show a very different situation of each of the forms. It should be said in advance that half-participle forms appear in free production of only three most advanced speakers (NIC, ELE and ANN) and EUN, while very few instances of gerund forms appear in DIN, ELE and ANN. Participles, both present and past, occur in many of the participants free production; however, they are only used as adjectives (no compound tenses are formed) up to advanced stages of the post-basic continuum.
At it can be seen in Table 10, in the repetition task the participle nukritusių is not reproduced by any of the basic variety and the beginning post-basic variety speakers. In addition, two of the post-basic variety speakers (EUN and MAD) reproduce it without the final ending, thus turning the form into the nominative case of feminine participle (nukritusi) which in its turn does not match the following noun. Therefore, the repetition at this stage seems to be more of form than of form and meaning.
The half-participle apparently is a more transparent form to the beginning speakers as more attempts to repeat it are registered. The stem is clearly recognized (except by DIE who tries to repeat it producing a non-existing and incomprehensible form) and in the effort to reproduce it the speakers attempt a form of that lexeme that is otherwise unknown to them. JUA, who has not acquired the future tense and never uses it in his free production, attempts with FUT2SG (although it differs significantly in form), ANG and GED attempt with different forms of FPAST, which are not part of their active forms in the free production, but the half-participle suffix -dam-bears resemblance to the frequentative past suffix -dav-and it is this resemblance they most likely count on. SUS also attempts with a form she is starting to insert in her repertoire, PRES1PL, also likely to the vague resemblance (following her own modification she changes the second verb in the sentence to PASTPL-like form as well). The gerund form, when not known but attempted to reproduce, is unanimously substituted with PRES3 by the beginning speakers DIE and JUA and GED as well. The correct repetition of this form by SAR, SUS, ALE and EUN seems to be more mechanical.

Compound tenses
The sentence repetition task included three most common compound tense forms: the compound present perfect tense yra matęs 'has seen', the compound past perfect tense with the feminine active participle buvo nuėjusi 'had gone', and the compound past tense with the passive participle buvo iškeptas 'was baked'.
The common feature of the task results, as shown in Table 11, is that the compound forms fail to be repeated by all the beginning speakers: pre-basic DIE, basic JUA and ANG, transitional SAR, and even SUS and VIC, who have just entered the post-basic continuum (SAR and VIC reproduce one full compound form). However, the speakers try to solve this issue in different ways or according to their capabilities.
In some cases, the speakers of the very beginning varieties manage to reproduce the auxiliary. When repeating the compound present perfect tense, JUA and ANG try to repeat the second part of the compound as well, although the result is incomprehensible. In the case of the compound past perfect tense, DIE and SAR reproduce only the auxiliary in the past. This could be partly due to the meaning, as stating 'had gone' or 'was' ultimately has the same result (see also Section 5.2).
There were three instances in the beginning varieties (DIE, JUA and SUS) as well as one in the postbasic variety (MAD), where the only part of the compound reproduced was the lexical verb. However, the beginning speakers managed to repeat it only with the dropped prefix, while MAD repeated the full form.
A more common way of solving the issue, also occasionally used by some more advanced speakers, was substituting the entire compound with a simple form of the same lexical verb. The two basic variety speakers JUA and ANG substitute it with their typical base form PRES3. However, PAST3 forms are preferred in the output of SAR and DIE (again, surprisingly, as PAST forms are not part of DIE's free production repertoire) and in nearly all the cases in the post-basic continuum. In such cases the form is abandoned in favor of transferring meaning. The use of FPAST3 by GED just confirms the previous hypothesis that the speaker is still trying to figure out the function of the past frequentative form and therefore tries to fit it in where he can. The use of FUT3 by EUN is hard to explain as not only the verb form elicited but also another time reference (vakar 'yesterday') occurred in the sentence.
A unique case in the task was the ignored prefix on the participle nuėjusi '~had gone'. This lexeme is the same as the one that occurred in the infinitive form discussed earlier. However, in that case only the beginning speakers omitted the prefix, while in this compound (buvo nuėjusi), both the beginners and advanced speakers did this. Besides, not only most of those who substituted the compound with the simple form (JUA, ANG, GED, EUN, and DIN), but also those who maintained the compound (NIC and ANN) omitted the prefix nu-. In this last case, it particularly differs from the natives' use, as the prefix adds to the perfective meaning. It may seem that prefixation, even though very frequent in native's input, is somewhat a challenge even in advanced L2 varieties when used in more complicated forms.

Transferring meaning
The EI task analyzed in this paper is a task focused on form, and therefore the elicitation should evoke the copying faculty. As was already seen in the form analysis, not only the copying faculty but also the construction, or rather re-construction, faculty is at use. This happens when a speaker cannot reproduce the form that is unknown or too difficult but re-constructs another form to convey the message. However, there were cases where nothing resembling the form elicited was reproduced. In some of these cases, the speaker just gave up due to difficulty exceeding his/her abilities, but in others, even though the forms were not reproduced, the speaker put some effort into conveying the meaning using other available means. This section will discuss these instances.
The most difficult sentence to repeat was as expected to be the one that was both long and complex. It included a past perfect compound and a FUT1PL form (as well as other parts of speech that are not discussed here but would merit separate attention). There were some speakers that did not manage to repeat it exactly, but because they understood it, they chose to recreate it in their own way. The following three examples come from ANN, who is the best speaker in the sample, JUA, who is a typical basic variety speaker, and DIE, who is pre-basic in free-production but, as already mentioned, has more advanced passive knowledge of Lithuanian. The sentence provided in the task was as follows: Lina buvo nuėjusi į kirpyklą todėl šiandien pamatysime jos naują šukuoseną As can be seen in examples 1-3, all the speakers understood the sentence and made an effort to express the 'final result'. ANN remains quite faithful to the elicited sentence. She forms the compound tense that has the past perfect meaning; however, she ignores the change of the subject and the tense in the second clause. JUA embeds the entire utterance in the present and ignores the change of the subject in the second clause as well. DIE manages to express the past (the speaker does not do that in his free production), but no other verb appears in the utterance (a typical feature of the pre-basic variety), and he finishes the utterance by recurring to English.
There were some examples of speakers making an effort to express a specific grammatical feature they did understand but were not able to reproduce. One example of this type is an utterance produced by SAR with a following explicit explanation (see example 4). The speaker only manages to repeat the first part of the elicited sentence (man gražu 'it is beautiful/nice for me'), maintains the infinitive form although changing the lexical verb into a synonym (žiūrėti 'watch' into matyti 'see'), adds some more words on his own, modifies the generally neutral word katė 'cat' into a more marked katinas 'male cat' (however, the speaker lives in Samogitia, where the neutral word in the dialect is the masculine form) and then goes on to explain that it "does something on its own", thus expressing the reflexive meaning without actually repeating the elicited form. 4) *SAR: man gražu matyti atrodė kažkaip nu kai kai... katinas... padaro kažkas pats nežinau.
Since quite a few speakers ignored the prefixed form of the passive participle in the compound buvo iškeptas 'was baked', it is interesting to see DIE's variant (see example 5). The speaker is successful at repeating the first clause kai jis grįžo namo 'when he came home', and even though he does not manage to reproduce the rest, he considers it important enough (as it is the main clause) to produce the participle and does not stop there. Interestingly, he adds geras 'good', which in this situation, given this speaker's limited production abilities, probably stands for the "result state", i.e. the cake is "good" when it is "done/baked". This result meaning is important to the meaning of the entire sentence, and this was the only way the speaker was able to express it.

5) *DIE: kai jis ga [/] grįžo namo... aš tortas keptas ah geras.
The last observation concerns again the data of DIE, a unique pre-basic variety speaker whose passive knowledge of Lithuanian is actually much more advanced. In the sentence that aimed to elicit a gerund and the past frequentative tense, the speaker fails to reproduce any of the targeted forms but tries to express as much of the meaning of the sentence as is possible to him. The sentence for elicitation was as follows: lietui lyjant, jis nevesdavo šuns į lauką '~when raining, he would not take the dog outside'. DIE's repetition is provided in example 6. 6) *DIE: čiandien [: šiandien] lyja... nieko su su šuo. {today rains…nothing with dog}

Comparison with free production
In the current analysis, comparison of the repetition test results with free production is made in some more interesting cases. Table 12 summarizes the outcomes of the sentence repetition task by indicating the form being present (+), the form being present on certain lexemes (±), cases of sporadic presence with uncertainty of function (?), the form found in formulas and alike (-) or being absent (empty cell). The marking of the table cells appears in yellow when the task performance corresponds to free production, in red when the speaker underperforms and in green in cases of overproduction when compared to free production.
It can be seen that, besides a couple of individual variations, the past frequentative tense, imperative mood and the rarer participles are in fact quite transparent to L2 learners, who do not encounter many difficulties when repeating these forms even if they do not produce them themselves. The findings also suggest that advanced speakers who did not use some of the targeted present tense forms in their free production, are able to do so when elicited. On the other hand, the overlay table also draws our attention to some forms that cause more difficulty. PRES2SG seems to be uncertain for advanced speakers even if the less advanced did not seem to find it troublesome. The PAST1PL form apparently causes some difficulty for speakers who have passed into the post-basic continuum but are still not very advanced. However, the greatest attention is drawn to the PAST1SG and FPAST1SG, which can be considered together given the same form and function of the morphological marker -au, which the speakers fail to repeat in the task. This is true not only for the beginning speakers, who do not yet use the past tense forms in their free production, but also to those who do. More specific research should be conducted in order to understand this peculiar difficulty.
DIE's data, where he repeats the targeted present tense and imperative forms even though he never uses them himself, shows that these forms are rather transparent and make part of the receptive or passive knowledge of Lithuanian L2.

Conclusion
To sum up, the sentence repetition task results show that in general L2 speakers do not re-produce verb forms they normally do not use in their free speech. This is especially true for basic variety speakers. Post-basic variety speakers are usually able to repeat transparent, even though rare verb forms. Passive knowledge of the most frequent forms is acquired over longer time spent in the country, although this does not seem to influence the more complex forms. The sequence of acquisition of tenses and moods, discovered in previous research, is generally confirmed.
If a form is modified, it is usually substituted with the 3 rd person form of the same tense. This finding did not emerge in other language L2 acquisition research that looked more generally into the sequence 4 Free production data: + form present; ± form present with certain lexemes only; ? form sporadically present but function is uncertain; -form found but possibly unintentionally (unfinished words, formulas and alike). An empty cell means the form is not found in the speaker's free production data. The task outcome is overlaid as follows: yellow indicates correspondence; red, underperformance; and green, overperformance.
of tense acquisition. The 3 rd person form in Lithuanian is the same for singular and plural, which makes it the most frequent form. This could be one of the reasons why not only PRES3 but also PAST3 and FPAST3 forms seem to be used as a type of "starting" forms in their own tense. In turn this could suggest that each new tense/mood that is being acquired has a certain "base" form, and this form for Lithuanian L2 is the most frequent 3 rd person form.
Less advanced speakers may substitute the unknown form with another form that they have started to notice in the input but have not yet acquired the function of, such as FPAST used instead of the halfparticiple. Reflexive participles and prefixes (which often have a perfective meaning in Lithuanian) also seem to be acquired according to the sequence of acquisition, i.e. firstly on simple forms and with more difficulty in more complex forms, such as the conditional mood and compound tenses. Naturally, further large-scale or cross-linguistic research would be very expected to confirm or to add to these findings.
Can we repeat what we do not say in L2?
S u m m a r y The paper analyzes the results of a sentence repetition task conducted with foreigners living in Lithuania. Natural-sounding and meaningful Lithuanian sentences were formed including some forms that are less likely to occur in non-natives' speech. The paper analyzes different forms of the verb: second person singular and plural of the present tense in the indicative mood as well as the reflexive third person form, first person singular and plural and the reflexive third person form of the simple past tense, third person and first person plural of the future tense, first person singular and third person forms of the past frequentative tense, second person singular of the imperative mood, third person of the conditional mood, the prefixed infinitive form, participle, half-participle and gerund as well as three instances of compound tenses. The paper not only looks into whether a certain form was repeated or not, but also analyzes what form the speaker produced as a substitute for the one he/she was not able to re-produce. Speech data of the same speakers had been gathered for a more comprehensive study of Lithuanian as a second language (L2); therefore, the paper also compares the results of the repetition task with spontaneous speech data.
The analysis confirms the sequence of verb form acquisition in Lithuanian determined in previous studies and reveals that in general basic variety speakers do not distinguish different verb forms according to their function.
The new findings in this study include the trend that when L2 speakers start noticing from the input a form that is new to them but are not yet certain about its function, they use it where there is need for the form that goes beyond their competence. Moreover, the third person form, which coincides in both singular and plural in Lithuanian and therefore is the most frequent form, tends to be acquired first not only in the present tense but also in other tenses of the indicative mood. The present analysis has demonstrated that even in such focused-on-form task speakers strive to convey the meaning of the sentence they understood, using all the resources available to them, even if they stray away from the exact form.
The comparison with spontaneous speech data shows that L2 speakers are able to repeat the majority of forms that they have already acquired and use in their spontaneous production, but they are not able to re-produce those forms that are not part of their repertoire and therefore substitute them with less complex known forms. Additionally, it has been observed that some forms, even though frequent, present difficulties for the L2 speakers (e.g. first person singular of the past tenses); however, some, namely the half-participle and gerund, although not used spontaneously, are rather transparent for post-basic variety speakers and are successfully re-produced.