1. Introduction and theoretical background

2The present study simulated a situation in which a learner takes an intensive language course in a new L2 and then does not use it for several months. This scenario aligns with the experiences of many learners who undergo intensive language exposure while studying abroad, but then receive significantly poorer input throughout the following school year due to reduced hours of instruction compared to an immersion setting. In an alternative, specular scenario, attriters experience significantly less exposure during the summer holidays (in the absence of language use, e.g. study stays abroad) than throughout the school year, when they receive regular language instruction.

3The paper considers three main aspects of this phenomenon: a) the attrition of vocabulary, b) the attrition of morphosyntax, with a focus on case marking, and c) the potential effect of task type. It is hoped that thanks to the innovative methodology of the experiment, the present data will contribute to the scientific debate on L2 attrition: based on their systematic review of L2 attrition studies, indeed, Bardovi-Harling and Stringer (2010: 22) note that “only 27 of the 49 L2 studies analysed language samples, and only a small portion of those conducted interlanguage analyses comparable to those found in SLA studies”.

4Several points should be stressed to adequately contextualise this study. The first concerns the specifics of the attrition process of initial interlanguages. Rather intuitively, language loss has been typically reported to be greater if the learners’ initial proficiency was low to begin with (De Bot and Clyne, 1989; Nagasawa, 1999; Reetz-Kurashige, 1999), as summarised by the aphorism “the more you know, the less you lose” (Hansen, 1999: 151).

5Another point concerns the differences in attrition trajectories between lexicon and grammar. The former has generally been found to be more susceptible to attrition than the latter (Kuhberg, 1992; Moorcroft and Gardner, 1987), although a study by Yoshitomi (1992) on initial L2 learners suggests otherwise. Furthermore, inflected elements comprised in formulae and fixed expressions may be retained as unanalysed blocks with greater ease than in more productive language use (Berman and Olshtain, 1983). In any case, it is worth stressing that in attrition research, the lexicon has been much more widely investigated than any other language layer has.

6Finally, in relatively recent years, research on the role of input in L2 processing and acquisition has gained momentum thanks to studies conducted within the “usage-based” framework (Wulff and Ellis 2018), following a period during which some interpretations of the “poverty of the stimulus” argument had somewhat limited scholars’ attention to this critical set of variables (Schwartz and Sprouse, 2012; Rankin and Unsworth, 2016; Crain et al., 2021). The “usage-based” approach claims that the statistical distribution of target language input properties strongly influences language acquisition. Input is nowadays considered a fundamental component of any SLA phenomenon, “the raw material of second language acquisition” (Carroll, 2001), to the extent that even hotly debated topics such as the existence of a “critical period” have been at least partly explained and interpreted in terms of by-products of this crucial variable (Flege, 2008).

7Among the various factors investigated in relation to the role of input, two are particularly relevant to this study. First, frequency has been shown to exert an undeniable influence on the acquisition process (Goldschneider and DeKeyser, 2002; Ellis, 2002). Frequency can be considered from two perspectives (Bybee and Hopper (eds.) 2001), namely type frequency—how many different lexical items (types) occur with a given morphological construction, and token frequency—the frequency of the item in question regardless of the word forms in which it appears. The present study relied on measures of token frequency.

8Second, lexical transparency has been argued to play an important role in comprehension and lexical learning. More specifically, transparency facilitates the recognition of the lexical meaning of a target language item, often along with the part of speech to which it belongs; in turn, this facilitates the early recognition of word order patterns and morphological forms (Singleton, 1999; Rast, 2008). Finally, transparency has been shown to be partly responsible for the overall processing ease of an unknown foreign language in the context of such areas as L2 acquisition (e.g. Watorek et al., 2020) and intercomprehension (Castagne, 2011; Heeringa et al., 2023).

9Due to the evident difficulty of controlling L2 input in a naturalistic setting, research on this topic has often favoured first-exposure studies (Han and Rast (eds.), 2014). However, these are also characterised by their own methodological challenges, in addition to those imposed by input-related research in general. Consequently, experiments investigating the initial stages of acquisition have often relied on artificial languages, or have been limited to the analysis of participants’ performance after only a few minutes of exposure to the input (e.g. Ellis and Sagarra, 2010; Gullberg et al., 2010; Carroll and Widjaja, 2013). Rast’s (2008) study of the first eight hours of exposure to L2 Polish made a significant contribution to the development of a novel methodology for studying the role of input upon first contact with a new language and in the minutes and hours that follow. The VILLA project (Dimroth et al., 2013), in which the present paper is framed, emerged directly from that experience (see also Rast (2017) for a follow-up to the 2008 study).

2. Research questions and hypotheses

10This study aimed to characterise the extent and speed of language attrition following very limited exposure to a new L2. This goal was achieved by comparing the results of a delayed post-test with those obtained at the end of the acquisition process, which was conducted in a highly controlled experimental setting. The data were obtained through a structured task (Q&A) and a Route Direction task (RD).

11Against the background presented in the previous section, the general research question of this study may be articulated into several more specific points.

RQ1: Which language component is affected the most: the lexicon or inflectional morphology?

HY1: Since inflectional morphology still exhibited significant variability at the end of the acquisition process (Hinz et al., 2013; Saturno, 2015), it is expected that attrition patterns will be more evident in the domain of the lexicon.

RQ2: What alternative strategies do learners adopt to compensate for their reduced linguistic ability?

HY2: In line with research on the early stages of naturalistic L2 acquisition (Klein and Perdue, 1992; Perdue (ed.), 1993; Giacalone Ramat, 2003), it is expected that learners will rely more heavily on extralinguistic resources, such as context, while simplifying grammatical meaning to focus on the lexical layer.

RQ3: How do the transparency and frequency of an item affect the production of (nominal) inflectional morphology following L2 attrition?

HY3: In the delayed data collection session, transparent lexical items will be produced with greater accuracy than opaque items. This question mainly concerns the structured Q&A task: concerning the Route Direction task, indeed, Watorek et al. (2016) have shown that learners choose lexical items according to the communicative goal rather than according to their characteristics (transparency/frequency). Moreover, extralinguistic material such as the city map employed in the task may suggest the target language word corresponding to useful referents, thus blurring the effects of input frequency and transparency.

3. Data collection

3.1 The VILLA project

12This study is a spin-off of the VILLA project (Varieties of Initial Learners in Language Acquisition), devoted to the study of the earliest stages of adult second language acquisition in a highly controlled experimental setting (Dimroth et al., 2013). Since the initiative aimed to describe interlanguage development over an extended period of exposure while maintaining full control over the acquisition experience, the input was provided in the form of a 14-hour Polish course.

13The course included no grammatical explanations, so learners had to autonomously establish form-function associations. There was a conspicuous interaction with both the teacher and the other learners. To ensure that all participants were exposed to the same amount and quality of broadly defined input (including practice outside the classroom), no homework was given. Similarly, participants were not allowed to take notes.

14The VILLA project was replicated with identical methodology in five different countries (Germany, the Netherlands, France, the United Kingdom, and Italy) and in two editions (meaning-based and form-based, depending on whether the input included focus-on-form interventions). Only the two Italian editions are considered here.

3.1.1. Frequency and transparency

15One of the main methodological novelties of the VILLA project is the thorough control of the input that the learners were exposed to. All classes were audio- and video-recorded; the resulting tracks were transcribed and annotated and can now be searched for relevant trends. Altogether, this methodology makes it possible to thoroughly correlate learner output with the input they received, thus answering some of the research questions of this study.

16Frequency was coded as a binary, rather than numerical variable. Based on Rast (2008), the nouns occurring in the language tasks were classified as “high-frequency” if they reached a token frequency of 20, “low-frequency” otherwise.

17Furthermore, lexical items were classified into two categories with respect to transparency. The dichotomy “transparent” vs. “opaque” is based on whether a speaker of one of the L1s of the project could be thought to be able to intuitively identify the lexical meaning of a given lexical item. This was assessed through a transparency test carried out prior to the beginning of the Polish course with a separate but comparable group of participants (Valentini and Grassi, 2016). Speakers of French, Italian, English, German, and Dutch with no previous knowledge of the target language translated a set of Polish words into their respective languages. Based on a 50% threshold of accuracy, this procedure resulted in a group of Polish nouns that speakers of all five languages could be thought to be able to easily recognize without any prior knowledge of the target language.

18Based on their frequency and transparency, the lexical items employed in the language tasks can be classified into four sets:

- high frequency (HF) and transparent (T).

- high frequency (HF) and opaque (O).

- Low-frequency (LF) and transparent (T).

- Low frequency (LF) opaque (O).

3.1.2. The languages in contact: Polish and Italian morphosyntax

19Polish was chosen as the target language of the VILLA project for various reasons. First, it is not commonly taught as a foreign language in the countries where the initiative was run, which facilitated the recruitment of participants with no prior exposure to the language. Second, it differs from the participants’ native languages in several respects. Of particular interest is the rich and complex nominal morphology, which contrasts two numbers, three genders in the singular and two in the plural, and—crucially—as many as seven cases (nominative, genitive, dative, accusative, instrumental, locative, vocative). Case is particularly relevant from a contrastive point of view because Italian only inflects full nouns with respect to number, with traces of case opposition emerging only in the pronominal paradigm.

20Table 1 presents a summary of the inflectional categories which are morphologically encoded on various parts of speech in Polish and Italian.

Table 1: morphosyntactic features of the source (Italian) and target (Polish) language

	nouns	pronouns	adjective	determiner
Polish	number, case	gender, number, case
Italian	number	gender, number, case	gender, number

21Polish nouns are characterised by a complex system of inflectional morphology. In this domain, word forms are determined by the interaction of number (singular, plural), gender (masculine animate, masculine inanimate, feminine, neuter), and case (nominative, genitive, dative, accusative, instrumental, locative, vocative). Table 2 presents the most common paradigms of nouns occurring in the VILLA input (mostly limited to the singular). The shaded cells contain word forms that were not present in the VILLA input.

Table 2 : Polish nominal paradigms comprised in the VILLA input, singular (adapted from Saturno & Watorek, 2020)

22Case marking plays several important functions in Polish. First, it distinguishes syntactic functions independently of the position of an item in the utterance: the SVO sentence in 1a and its OVS counterpart in 1b express the same referential meaning but differ in terms of information structure. Note that information structure and pragmatics were not discussed in the VILLA course, so that for the VILLA learners, the two utterances in 1 are identical in all respects (except frequency).

(1) a. [pjes-∅ ˈgɔni ˈkɔt-a]

dog-nom.sg chases cat-acc.sg

‘The dog chases the cat’

b. [ˈkɔt-a ˈgɔni pjes-0]

cat-acc.sg chases dog-nom.sg

‘It is the dog that chases the cat’

23Case may also be directly associated with specific semantic roles. In (2a), the instrumental case expresses the semantic role “instrument”. In (2b), the same case is used to mark the focus component of a copular structure.

(2) a. [ˈpiʃe ˈpjur-em]

write.1sg pen-ins

‘I write with a pen’

b. [ja ˈjestem stuˈdent-em]

I(nom) am student-ins.sg

‘I am a student’

24Finally, nominal elements within prepositional phrases (henceforth PP) always occur in a case other than the nominative. Different prepositions may require different cases (3).

(3) a. [ˈɔbok ˈsklɛp-u]

by shop-gen.sg

‘Next to the shop’

b. [za ˈsklɛp-em]

past shop-ins.sg

‘Past the shop’

25In contrast, Italian nouns distinguish only between a singular and a plural form. Italian nouns do not change their word form depending on their syntactic function or semantic role. These values are mainly expressed through the position of the noun relative to the verb (4a vs. 4b), or the preposition it modifies (4c and 4d).

(4) a. il cane insegue il gatto

the dog chases the cat

‘The dog chases the cat’

b. il gatto insegue il cane

the cat chases the dog

‘The cat chases the dog’

c. presso il negozio

by the shop

‘By the shop’

d. oltre il negozio

past the shop

‘Past the shop’

3.1.3. Participants

26The present contribution focusses on the Italian edition of the VILLA project, which saw the participation of 31 learners with no previous experience of Polish or any other Slavic languages. Most were university students in subjects unrelated to linguistics, although other backgrounds were also represented. The mean age was 24 (min = 19, max = 35). In addition to their L1, most VILLA learners knew at least one foreign language at various degrees of proficiency (most commonly English). Knowledge of Latin to a various degree was also widespread among Italian learners. This may be problematic for a first exposure study because Latin remarkably resembles Polish in terms of morphosyntax, most notably in the morphological expression of case. For learners with Latin experience, the category “case” and free word order may not be as new as for learners without this language background.

27The participants’ lack of Polish experience was carefully assessed through language profile questionnaires and a specially designed language sensitivity test (see section 3.1.1.). This was a necessary requirement to ensure that grammatical features and categories were processed based on the experimental input alone, rather than previous linguistic knowledge.

3.2. Language tasks

28During the course, empirical language data were collected using various (psycho)linguistic tasks. The present contribution considers the “Question & Answer” task (henceforth Q&A) and the “route direction” task (RD). Combined, they offer a comprehensive view of morphosyntactic processing in different elicitation settings (structured test vs. communicative task) and grammatical domains (case marking within PPs or expressing core syntactic functions).

3.2.1. The Q&A task

29In the Q&A task (Hinz et al., 2013; Saturno, 2015, 2020), participants were asked to describe a given character using the correct copular structure and the details supplied in the prompts. Initially, learners were presented with an icon indicating the gender of the character to be described (Figure 1, left). They were then presented with the question ‘who is this/he/she?’, which was formulated as either of two types of interrogative structure (see below). Finally, the nationality or profession of the target referent was presented (Figure 1, right). The learner’s task was to answer the question using this information and the syntactic structure required by the stimulus question (e.g. [on jest piˈlɔtem], ‘he is a pilot’). With few exceptions, all referents had been introduced in the input and should have been familiar to the learners. The task comprised 36 target sentences, with target nouns equally distributed across the masculine and feminine genders.

30In the Q&A task, copular structures were instantiated by two different syntactic constructions (Table 3). In NOM-type structures, the noun appears in the nominative case (in bare consonant for masculine nouns, in -a for feminine nouns) and is referred to by the interrogative pronoun [kto] ‘who(nom.sg)’ and the determiner to ‘this(nom.sg)’. In INS-type structures, the noun appears in the instrumental case (in [-em] for masculine nouns, in [-ɔw͂] for feminine nouns) and is referred to by the interrogative pronoun [kim] ‘who(ins.sg)’ and the personal pronouns [ɔn] ‘he(nom.sg)’ or [ˈɔna] ‘she(nom.sg)’.

Table 3 : Polish copular structures

type	question	declarative	structure	gender
NOM	[kto to jest?] ‘Who is this?’	[to jest ˈstraʒak-0] ‘This is a fireman’	to + -0 (nom.sg)	m
		[to jest akˈtɔrk-a] ‘This is an actress’	to + -a (nom.sg)	f
INS	[kim ɔn jest?] ‘Who is he?’	[on jest straˈʒak-jem] ‘He is a fireman’	on + -em (instr.sg)	m
	[kim ˈɔna jest?] ‘Who is she?’	[ˈɔna jest studɛntk-ɔw͂] ‘She is an actress’	ona + -ɔw͂ (instr.sg)	f

31In the data coding phase, responses in which one or more elements (pronoun, copula verb, or noun) were omitted were excluded from the analysis. Additionally, it was noted that Italian speakers tended to insert a centralised epenthetic vowel at the boundary of word-final syllables ending in a consonant, which resulted in the pronouns on ‘he’ and ona ‘she’ often sounding quite alike. Since it is unlikely that the participants may have had difficulties with the expression of a referent’s sex through personal pronouns, and since the selection of either pronoun required that the noun should appear in the instrumental case, for the purposes of the present analysis the two forms were merged together in a single ‘personal pronoun’ category, on(a), which contrasts with the determiner to. With this caveat, pronominal and nominal elements were considered to have been correctly supplied if, taken in isolation, they matched the corresponding expected form. In contrast, complete copular structures were considered correct if both pronominal and nominal forms matched the expected form. As the task did not target vocabulary, the accuracy of nominal inflection was computed solely based on the ending supplied, independently of whether the lexical morpheme had been correctly selected or produced accurately.

3.2.2. The RD task

32In the RD task, learners were individually interviewed by a Polish-speaking interviewer. The task is a role-play in which the interviewer impersonates a tourist wishing to reach a given address from the train station (this background information was provided to the learner in Polish). Showing a city map (Figure 2), the interviewer asked the learner for directions (again, in Polish): “Excuse me, how can I go from the train station to [uˈliʦa ˈDɔbra] number 4?”. Both the departure and destination points were highlighted on the map, the train station by a red dot and the word [ˈstaʦja] (bottom right), and the final address by the home icon (top left). The suggested (though not the only possible) route was highlighted in red. The map also shows several buildings and points of interest, identified by the corresponding Polish word and sometimes by an icon.

33This task, including its background situation, the request for directions, and the necessary vocabulary and grammatical structures, was not new to the learners, as it had been extensively practiced during the preceding 14 hours. The pattern was introduced by the teacher on several occasions during classes. In the excerpt in (5), the teacher demonstrates how to give directions. Note that the target address was indifferently referred to as [dɔm koˈvalskix] (line 1), ‘the Kowalskis’ home’, after the name of the family who lived there in the fictional context of the course, or [uˈliʦa ˈDɔbra ˈnumer ˈʧterɪ], ‘Dobra Street 4’ (line 8). It is worth noting that the points of interest depicted on the map were used as waypoints in the form of prepositional phrases, e.g. [za ʃpiˈtalem] ‘past the hospital’.

(5) 1 [pʃeˈpraʃam, jak iɕʨ do ˈdɔmu koˈvalskʲix?]

‘Excuse me, how can I go to the Kowalskis’?’

2 [ˈjestem na ˈstaʦɪi].

‘I’m at the train station.’

3 [ˈprɔʃe iɕʨ ˈprɔsto uˈlicow͂ ˈnʲiskow͂]

‘Please go straight along Niska Street.’

4 [za ʃpiˈtalem ˈprɔʃe ˈskrenʨiʨ v ˈpravo]

‘Past the hospital, please turn right.’

5 [to jest uˈliʦa juˈljuʃa swovaˈʦkjego]

‘That’s Juliusza Słowackiego Street.’

6 [ˈprɔʃe iɕʨ ˈprɔsto ˈɔbok ˈʃkowɪ ˈɔbok teˈatru aʃ do uˈliʦɪ ˈdɔbrei̯]

‘Please go straight near the school and the theatre, up to Dobra Street.’

7 [za ˈsklɛpem ˈprɔʃe ˈskrenʨiʨ v lewo i iɕʨ ˈprɔsto]

‘Past the shop, please turn left and go straight.’

8 [dom ˈnumer ˈʧterɪ znai̯ˈduje ɕe na ˈlɛvo]

‘The home number 4 is on the left.’

34During classes, learners were also asked to perform the same task aloud, guiding the teacher along an imaginary route highlighted on various maps, including the one used in the test. The task was rehearsed frequently over several days, as shown by the frequency figures in Table 4. These refer to occurrences produced by the teacher (including feedback on learner output). A significant number of additional occurrences were produced by the learners when performing the task.

Table 4: Route Direction task, examples of absolute frequencies

Expression	mean frequency
[iɕʨ] ‘go’	271
[ˈskrenʨiʨ v ˈpravo] ‘turn right’	142
[uˈlicow͂] ‘street:ins.sg’	75
[uˈlicow͂ ˈnʲiskow͂] ‘along Niska Street’	29
[za] ‘beyond’	302
[za ˈsklɛpem] ‘past the shop’	26

3.2.3. Data collection timeline

35During the experimental course, the Q&A task was administered twice, i.e. after 4h30 (T1) and 10h30 (T2) of exposure to the input. The RD task was administered only once at the end of the course, i.e. after 14 hours (T1). Within the broader setting of the VILLA project, the present work represents a follow-up study unique to the Italian edition of the initiative. Approximately 8 months after the end of the Polish course (8m), learners were asked to voluntarily repeat some of the experimental tasks, which were not modified. As the objective of the experiment was to evaluate L2 attrition in the absence of further L2 input, all participants who had been exposed to Polish or other Slavic languages in the meantime were excluded from the delayed data collection process. Of the original 31 participants, only 14 performed the task again at 8m. Table 5summarises the experimental design.

Table 5 : data collection timeline

task	T1	T2	8m
Q&A	4h30	10h30	8 months
RD	14h	-
no. subjects	31	31	14

4. Results

36This study adopted a mixed-method approach. First, a quantitative analysis of the interlanguage produced at different times was performed to identify possible common trends. Following this initial recognition, a few interlanguage samples were analysed in greater detail to explore the strategies adopted by learners to compensate for the attrition of specific language layers and ultimately verify whether the learner was still able to perform the language task successfully.

4.1. The Q&A task

37Figure 3 presents the participants’ mean accuracy rates by test time and target structure. Individual participants are identified by their four-digit VILLA code (e.g. 5101); the score achieved by the same person at the three different test times are connected by green (T1 to T2) and red segments (T2 to 8m). Standard boxplots summarise the distribution of mean scores at every test time; the figure in a larger font indicates the mean score of the group.

38A marked improvement occurred between T1 and T2, followed by a decline after 8 months of lack of input exposure. However, the mean scores did not fall lower than at T1.

39A Bayesian model was fitted using the package brms (Bürkner 2017) for R (R Core team 2021) with accuracy as response and session as independent variable, in addition to by-participant random intercepts and by-session random slopes. The model showed that while the scores at T2 were considerably different from their counterparts at both T1 (mean = -0.95, sd = 0.34, 95% CI = -1.65—-0.29) and 8m (mean = 0.74, sd = 0.37, 95% CI = 0.03—1.48), the latter were more similar to each other (mean = -0.20, sd = 0.24, 95% CI = -0.68—0.28).

40Turning to a qualitative analysis of the structures produced by the participants, Figure 4 describes the output as a combination of pronouns (personal pronoun or invariable to) and inflectional endings. The four geometric shapes represent the target structures. For example, the responses to the target structure “on(a) + -em” (personal pronoun + instrumental -em ending of masculine nouns) mostly consisted of these two elements combined; however, a few respondents associated the personal pronoun with a consonant ending (C), typical of the nominative case, while a few other participants combined the pronoun with the -ą [ow͂] ending of the feminine instrumental. No learner produced the invariable pronoun to in response to this target structure. What matters most for the present argument is that the learners’ preferences do not seem to change dramatically over time, although one may observe slightly greater dispersion across non-target-like combinations.

41At 8m, numerous participants session produced “creative forms” using lexical bases that were not mentioned in the VILLA input, e.g. [to jest ˈʧinema] ‘this is a cinema’ or [on jest ˈʧinemaˈtɔrski] ‘he is *filmic’, cf. the target [on jest akˈtɔrem] ‘he is an actor’. This can be interpreted as evidence of the loss of the appropriate lexical morphemes introduced during the VILLA course.

42Finally, Figure 5 depicts the effects of frequency and transparency over time. Across all groups, mean scores decrease by 8-10% relative to T2, following an initial increase from T1 to T2 (less noticeable in the case of opaque words).

43A Bayesian model was fitted using the package brms (Bürkner 2017) for R (R Core team 2021) with accuracy as response and the interaction between session, frequency and transparency as independent variable, in addition to by-participant random intercepts and random slopes for the interaction just mentioned. The model revealed no significant inter-group differences in the decrease in scores between T2 and 8m.

4.2. The RD task

44The focus is on case marking in prepositional phrases: by morphosyntactic accuracy, it is meant whether the noun appears in a form that is compatible with the preposition that governs it. A slight decrease in scores between the two test times can be observed, even though accuracy was already quite low at T1.

45A Bayesian model was fitted using the package brms (Bürkner 2017) for R (R Core team 2021) with the accuracy of prepositional phrases as response and session as independent variable, in addition to by-participant random intercepts and by-session random slopes. The model showed that the 95% CI of the difference between the scores at 8m and T1 comprises 0 (mean = -1.15, sd = 0.96, 95% CI = -3.40—0.54).

46From a qualitative perspective, it is striking that learner output seems to dramatically decrease in the number of prepositional phrases over time. It is worth noting that at T1, prepositional phrases were commonly used to indicate waypoints (6, cf. 7, elicited at 8m).

(6) 1 [ˈprɔʃe iɕʨ uˈlicow͂ ˈnʲiskow͂ za ʃpiˈtal-u]

Please go street-nom.sg Niska-ins.sg past hospital-loc.sg

2 [ˈprɔsto skʃɪʒoˈvan-je]

straight crossing-acc.sg

3 [ˈprɔʃe ˈprɔsto ˈɔbok teˈatr-u ˈʃkow-u]

please straight near theatre-gen.sg school-?

(7) 1 [ˈskrenʨ v ˈpravo uˈliʦ-a juˈljuʃ-a]

turn in right street-nom.sg Juliusz-gen?

2 [i ˈskrenʨ v ˈlɛvo uˈliʦ-a ˈdɔbr-a]

and turn in left street-nom.sg Dobra-nom.sg

3 [i dom koˈvalski ˈʧterɪ]

and home Kowalski four

47In the delayed task, prepositional phrases became less common or disappeared altogether, depending on the individual participant. In the latter case, route directions consisted only of adverbs and a few street names (8).

(8) T1 [ˈnumer ˈʧterɪ znai̯ˈduje ɕe na ˈlɛvo]

number four is located on left

8m [dɔm ˈnumer ˈʧterɪ na ˈlɛvo]

home number four on left

48Similarly, finite verbs tend to disappear in the delayed task. Table 6 lists a set of words that were produced at T1 but omitted (by the same learner) in the delayed task. These include prepositions, verbs, and several nouns. As can be seen from the last column, some of these words were common in the input.

Table 6 : RD task, lexical items omitted at 8m

49Conversely, Table 7 lists the words that were most commonly retained in the delayed task (again, relative to the output of the same learner at T1). Many of these are space adverbs and motion verbs. Together, these lexical items make up the core vocabulary needed to carry out the route direction task.

Table 7 : lexical items retained at 8m

word	retentions (out of 14)	input freq. (14h)
[ˈlɛvo] ‘left’	14	170
[ˈpravo] ‘right’	13	215
[ˈskrenʨiʨ] ‘turn’	13	189
[iɕʨ] ‘go’	11	308
[na] ‘on’	11	942
[ˈprɔsto] ‘straight’	11	174
[v] ‘in’	9	983
[i] ‘and’	8	1619
[jest] ‘is’	6	5236

50However, even the lexical items listed above are far from indispensable, since given an appropriate context and extra-linguistic tools—the map, in the present case—one could in principle carry out the task with hardly any linguistic items, as in the minimalistic performance illustrated in (9).

(9) 1 [uˈliʦa uˈliʦa ˈnʲiska]

street:nom.sg street:nom.sg Niska:nom.sg

2 [uˈliʦa juˈljuʃa in swovinˈskjego]

street:nom.sg Juliusza in(?) Słowackiego

3 [end uˈliʦa dɔbra]

and(EN?) street:nom.sg Dobra:nom.sg

4.3. Task comparison

51A Bayesian model was fitted using the package brms (Bürkner, 2017) for R (R Core team 2021) with accuracy as response and the interaction between session and task as independent variable, in addition to by-participant random intercepts and random slopes for the interaction just mentioned. The 95% CI of the interaction coefficient amply comprises 0 (mean = 0.69, sd = 1.28, 95% CI = -1.48—3.66), possibly due to the relative paucity of the data.

5. Discussion

52The results of the delayed data collection are summarised as follows. The analysis of the RD task has shown that the most characteristic feature of attrition in this task is the notable simplification of texts, mainly due to the elimination of all waypoints that were not strictly necessary. A significant omission rate was also observed for numerous lexical items, possibly because of the difficulty in managing noun morphology within prepositional phrases. However, elements expressing basic motion concepts are typically—though not always—retained. In spite of their relying on contextual and extra-linguistic tools, learners seem to still be able to cope with a complex communicative task in a foreign language after several months of attrition: the task is still carried out efficiently with the aid of a minimal repertoire of lexical items, such as ‘straight’, ‘left’, ‘right’, ‘street’, and the street names depicted on the maps. The map is an essential component of this strategy.

53The omission of waypoints probably stems from the learners’ wish to simplify the structure of their output, rather than from the unavailability of lexical items, most of which were transparently indicated in writing on the map. Specifically, known difficulties with nominal morphology in prepositional phrases, perhaps along with limited lexical access to the prepositions, may have prompted learners to avoid these structures altogether.

54The fact that the learners did not produce any of the place complements which were the target of this task does not immediately warrant the conclusion that they are no longer accessible—regrettably, the VILLA project comprises no source of additional data in this respect (e.g. a picture matching task). The test-like context, in contrast, together with the habits developed by classroom practice, implicitly prompts learners to do their best not only in terms of accomplishing the task, i.e., guiding the tourist to the required destination, but also to show as much proficiency in the target language as possible. Indeed, most learners tried to include the expressions they had practiced during the course, not because they were vital—it has just been argued that they were not—but because the learners probably felt obliged to.

55Given the repetitiveness of the task and its close resemblance to classroom practice activities, caution should be exercised not to confuse creative oral production with the use of memorised language chunks. Unfortunately, this question seems difficult to address experimentally in the context of the VILLA project. The possibility of producing texts other than the rehearsed model is strongly limited by the reduced range of the available vocabulary. Learners could remember the Polish words for ‘turn right’, ‘street’, or ‘shop’, but would have to improvise had they needed other words, such as ‘traffic light’ or ‘square’, which were not introduced during the course.

56It should also be acknowledged that when making reference to space in a realistic communicative situation, learners would probably rely heavily on gestures (Gullberg, 2024) or, possibly, other foreign languages, the use of which was explicitly prohibited during the VILLA course. Since the VILLA project does not include video recordings of the tasks discussed here, it is unfortunately impossible to investigate this phenomenon with the available data. However, it is easy to imagine that in a real communicative context, gestures would be widely used to compensate for linguistic deficits.

57Regarding the Q&A task, a moderate decrease in the mean scores was observed between the second and the delayed data collection times. Similarly, responses were slightly more dispersed across non-target-like combinations of pronouns and inflectional endings. Compared to T1 and T2, creative forms based on the participants’ language repertoire (as opposed to existing Polish words introduced during the course) became more common at 8m. This probably indicates the loss of some lexemes. Altogether, one may conclude that following attrition, learners adopt strategies that resemble the “communication at all costs” approach that one typically observes in communicative tasks: when cognitive (Skehan, 2009) and expressive resources are limited, priority is given to the expression of lexical meaning to the detriment of conformity to the target language model. If individual differences play a prominent role in language acquisition, they can be hypothesised to do so to an even greater extent in language attrition, due to the greater number of variables involved.

58In summary, the research question may be answered as follows:

RQ1: Which language component is affected the most: the lexicon or inflectional morphology?

In the RD task, morphosyntax almost entirely disappears, along with a conspicuous proportion of the lexicon. In the Q&A task, however, morphosyntax was still partially present in learner output. Thus, despite its greater vulnerability compared to the lexicon, the retention of morphosyntax is sensitive to task type and possibly—in turn—to the retention of specific lexical items, which may have been memorised in an inflected form due to their distribution in the input (Janda, 2019).

RQ2: What alternative strategies do learners adopt to compensate for their reduced linguistic skills?

In the RD task, learner output at 8m comprised almost exclusively nouns, to the detriment of prepositional phrases and verbs. Extra-linguistic elements—the map—played an even more substantial role than at the end of the VILLA course. No alternative strategy seems possible in the Q&A task.

RQ3: How does the transparency and frequency of an item affect the production of (nominal) inflectional morphology following L2 attrition?

59No statistically significant tendencies were observed with respect to the role of transparency and frequency.

6. Conclusion

60Overall, it can be concluded that in a highly structured task such as the Q&A task, substantial lexical attrition can be observed, along with a moderate decrease in morphological accuracy. In contrast, a complex communicative task like the RD task produces a sharp impoverishment of morphosyntax, as evidenced by the absence of finite verbs and prepositional phrases. This decrease is probably related to significant lexical attrition.

61Inflectional morphology is a sensitive point in both L2 acquisition and attrition. Therefore, it is not unexpected that in a communicative task such as the RD task, morphosyntax suffered from the participants’ lack of L2 practice. It is more surprising that morphosyntax was partially maintained in the highly structured Q&A task. However, other studies have reported early sensitivity to inflectional morphology in the VILLA data, consistent with the findings of other first-exposure studies, such as Gullberg et al. (2012). The format of the VILLA experiment may have made inflectional morphology particularly evident, encouraging participants to notice it (Izumi, 2002) and experiment with it. The VILLA course was no doubt quite intensive, since it took place every day for 1,5 hours a day for two weeks. It also comprised very frequent repetitions and focused on a limited number of lexical items and grammatical structures. Nevertheless, the unusual degree of control it allows makes it an excellent source of data for investigating issues of L2 acquisition and—as this paper demonstrates—L2 attrition following minimal exposure to the input.

Authorship attribution

62This study is the result of a joint effort by the two authors. Nevertheless, Jacopo Saturno was more directly involved in sections 1, 2, and 4, Marzena Watorek in the remaining sections of the paper.