Formulaic language in L1 and advanced L2 English speech: multiword structures in the speech of two Swedish groups compared to a group of L1 English speakers

: The aim of the study was to compare the use of formulaic language, here called multiword structures, in advanced L2 English speech of two Swedish groups with a group of ﬁ rst language (L1) speakers of English. One Swedish group was resident in London and one in Stockholm, thus implying different degrees of exposure to English. Three categories, notably multiword phrases, multiword utterances and metalinguistic multiword structures, were investigated by comparing quantity and distribution. The groups completed two oral tasks, a role play and a retelling task. The results of the Swedish group resident in London were similar to the L1 English group on all measurements of multiword structures in the role play. In the retelling task multiword phrases were signi ﬁ cantly fewer in both second language (L2) groups compared to the L1 English group, although there were L2 individuals within the L1 speaker range. Another question concerned the extent to which the three groups used multiword structures which were equivalents , i.e. transferable between English and Swedish. In the retelling task both Swedish groups produced signi ﬁ cantly more equivalents than the L1 speakers. The implications of these ﬁ ndings are discussed.


Introduction
Formulaic language cuts across most languages (Buerki 2020).According to Ellis (2006) formulaic language plays a vital role for acquisition, proficiency and idiomaticity in both L1 and L2.In the present study instantiations of formulaic language are called multiword structures (henceforth MWSs). 1 English MWSs were studied in the speech of two L2 English groups with Swedish as their L1 and one L1 English group.MWSs are topical issues in studies focusing on Second Language Acquisition (SLA) research when L2 Swedish speakers at very advanced (near-native) levels are compared with L1 Swedish speakers (Abrahamsson and Hyltenstam 2009;Ekberg 2003; for L2 English, French and Spanish Erman et al. 2015; for L2 English Erman et al. 2016).They all show that MWSs distinguish L2 from L1 production.Indeed, results from studies of formulaic language have shown that it is linked to high-level proficiency in both speech and writing (Boers et al. 2006;Forsberg 2008;Forsberg Lundell and Erman 2012;Hargreaves 2001;Lewis 2009;Stengers et al. 2011).
MWSs are referred to by different names in the literature, such as formulaic sequences (Schmitt 2004;Wray 2002), multiword units (Granger and Paquot 2008), and more transparent terms such as conventionalized word combinations (Foster et al. 2014).In the present study the term multiword structure is used to allow for the fact that, although some multiword structures form units as in (1), many MWSs are discontinuous as a result of grammatical adaption as in (2) or expansions, here through an embedded MWS (a bit of) as in 3). 1) no problem 2) the nuts were tightened 3) get into (a bit of) an argument As the examples show, MWSs include a variety of word combinations.Our definition of an MWS is: A conventionalized combination of at least two graphic words, which is accepted and expected in a speech community, e.g.fall behind, swat a fly, light a cigarette, have a bad day, in the afternoon, conveyor belt, how are you doing, no problem, here we go.
Conventionalisation is a common term in SLA research for the kind of linguistic phenomenon that is the focus of this paper (cf.Yorio 1980;Nunberg et al. 1994;Stubbs 2002;Forsberg 2008;Foster et al. 2014).It refers to a standardized combination of words (see Section 3.3.1).The examples above and throughout the paper are from our dataset.
This paper, as part of a large-scale project addressing near-native performance, involves the analysis of MWSs in entire texts (e.g., Erman et al. 2016;Erman et al. 2015Erman et al. , 2018)). 2 Apart from measuring the quantity of MWSs on group level, the analysis included measurements of individual variation within and comparisons between the three groups.The study also investigated the proportion of MWSs that English and Swedish have in common, and thus MWSs in both languages, here called equivalents (EQs; Section 3.4.4).
The aim of the study was to compare the use of English MWSs in the spoken production of two Swedish groups with different experiences of L2 English with that of L1 English speakers across two tasks (see Section 3.2).This study thus focuses on idiomaticity, which in our research corresponds to the use of MWSs appropriate for the context (cf.Erman and Warren 2000;Warren 2005).
The current paper addresses two research questions: 1.What are the similarities and differences in the frequency and distribution of MWS categories on group and individual level across the two tasks?2. What is the proportion of EQs in the two tasks across the three groups?
The first question concerns the frequency and distribution of the three different MWS categories: MW phrases, MW utterances and metalinguistic MWSs on both group and individual levels in a role play and a retelling task.Our expectation was that the London Swedes would match the L1 production in the role play, given their daily input and use (Kecskés 2000).Our second question addresses the proportion of MWSs that are similar in Swedish and English.Since English and Swedish are related languages it follows that they have a number of MWSs in common, here EQs.EQs are more easily acquired and accessed for the L2 user (cf.Jarvis 2000).Therefore, the two L2 groups were expected to have more EQs than the L1 group in both tasks (e.g., Granger 1998;Wolter and Gyllstad 2011).
The structure of the paper is as follows: we first provide the literature review, including theoretical background and relevant earlier research of MWSs in advanced L2 production.Subsequently our data, methodology and analytical procedure are presented.This is followed by data analysis of group and individual tasks.The paper ends with a discussion of the central findings and the conclusion.

Literature review
This section is divided in two parts, the theoretical background forming the basis for this research and earlier studies pertaining to our research questions, i.e. how L1 and L2 English speakers use multiword structures.

Theoretical background
The early 1990s was the starting point for using corpora in applied linguistics.New ideas and insights pointed in the same direction.Data emerging from the study of corpora led Sinclair (1991) to formulate his idiom principle, i.e., the proposal that words are co-selected rather than selected one at a time.This principle, inspired by corpus research, is comparable to Ellis's (1996) psycholinguistic account of associative learning and chunking, explaining how chunks of words are developed in long-term memory.His account suggests that chunks are acquired in interaction with input from the environment in both L1 and L2, constituting an important factor in the development of fluency in language production.Stubbs (2002: 57) even claims that "… all the most frequent words in the language occur in lexicosemantic units, with often surprisingly high strengths of attraction between collocates", implying that these words belong together forming a standard conventionalized expression.
Although frequency is an important variable in language acquisition, there are also other factors to consider.For instance, perceptual salience and usefulness underlie noticing sequences of language (Lenko-Szymanska 2014).Context is another important aspect of associative learning (Ellis 2006).The importance of context and the needs of the individual for acquiring formulaic sequences in a specific situation are pointed out by Wray (2002) and Fitzpatrick and Wray (2006).
Context is an important variable also in Hoey's (2005) account of lexical priming.Lexical priming involves the interaction between the repeated experience of language in context and the mental lexicon.Meaning can be expressed in a number of ways, but the use of appropriate word combinations for the context makes language performance sound natural, with each word inherently linked to another one (Hoey 2005: 6).Unusual word combinations may express the same meaning (cf.Hoey 2005;Pawley and Syder 1983), but with unexpected vocabulary for the context, it will sound unnatural.According to Hoey (2005: 184) all speakers, i.e., both L1 and L2 speakers, are "in a permanent state of learning", implying that the more we learn about different areas and topics the more associative links are formed.

Earlier research
In what follows we focus on work where differences between L1 and L2 speakers have been found, notably, (1) formulaic language in advanced L2 production including processing advantages of using it and (2) L1 influence on L2 production of formulaic language.Earlier studies within the project involving the same material as in the present study include the use of voiced pauses (Erman and Lewis 2013) and advanced vocabulary (Erman and Lewis 2015a, 2015b, 2019).

Studies of MWSs in advanced L2 production
The acquisition of MWSs is favored by time spent in the target language country.Kecskés (2000) maintains that it takes at least two years' stay in the target language community for advanced learners to use utterances appropriate for the situation in ways that approach L1 speaker usage (see also Forsberg 2008;Forsberg and Fant 2010;Erman and Lewis 2015a;Erman et al. 2015).Groom (2009) investigated phrases with frequent prepositions for two groups of Swedish writers of L2 English.One group had spent less than one month and the other more than 12 months in the UK.The results of his analysis showed that time spent in an L2 environment had a positive effect on learning outcomes.
Among studies of spoken production linking formulaic language to proficiency the following are noteworthy: Boers et al. (2006), Forsberg (2008), andStengers et al. (2011).Forsberg (2008) who investigated spoken L2 French found that the quantity of MWSs distinguished different levels of proficiency.Only the most advanced L2 speakers resident in France were nativelike.Erman et al. (2015) found that the resident L2 speakers of English, French and Spanish, having lived at least five years in the respective target countries, reached nativelike levels of MW utterances, corroborating Kecskés' (2000) finding.In a study of verb-noun collocations, however, Forsberg Lundell and Lindqvist (2014), investigating very advanced L2 French (average length of residency was 14.5 years, with a minimum of five years), found significant differences between the L1 speakers and the very advanced L2 speakers (for an overview of difficulties the L2 adult learner faces in acquiring formulaic language, see Arnon and Christiansen 2017: 621-623).There was, however, no correlation between length of residency and the test results.This was suggested to confirm Cummins's (1981) claim that length of residency effects tend to diminish after five years in the target language community.Another study (Forsberg Lundell and Sandgren 2013) showed that individual traits such as aptitude and personality were more important factors for the acquisition of collocations than length of residency.In L2 compositions and gap tests the quantity of appropriate MWSs were found to correlate with proficiency levels in that the total number of MWSs increased with higher levels of proficiency and so did the number of correct answers in the MWS gap tests (Lewis 2009).
Furthermore, studies indicate that there are processing advantages of using MWSs in both L1 and L2 (cf.Wray 2002;Wiktorsson 2003;Siyanova and Schmitt 2008;Wolter and Gyllstad 2011;Carrol and Conklin 2020).By using eye-tracking Wiktorsson (2003) and Carrol and Conklin (2020) found that MWSs were read faster than non-MWSs.Conklin and Schmitt (2008) also found significantly shorter reading times for both literal and metaphorical formulaic sequences compared to nonformulaic control phrases for L1 and L2 speakers alike, although the times were generally longer for the L2 speakers.
L1 and advanced L2 English speech

L1 influence on L2 production
According to Goulart (2019: 13) "L1 plays a role in the correct use of collocations by learners".Words and phrases that are similar between languages are commonly referred to as cognates when sharing the same historical background (e.g.Kroll et al. 2002;Lotto and de Groot 1998), or translation equivalents (e.g.Wolter and Gyllstad 2011) when translatable between languages verbatim.Peters (2015) found an L1 influence in particular on the ease of acquiring adjective-noun L2 collocations.Similar results were shown in Granger (1998) for adverb-adjective collocations.
The role of similarity between L1 and L2, which is important in L2 acquisition, has been discussed by several researchers (cf.Lotto and de Groot 1998;Jarvis 2000).Ringbom (2007) has pointed out the advantage of cross-linguistic similarity in learning a foreign language through his emphasis on the role of positive transfer.This is shown in Nesselhauf (2003: 237) who studied verb-noun combinations (cf.Howarth 1998) in L2 English writing and found an L1 influence not only on mistakes but also on the wording of correct collocations, transferable between L1 and L2.Since Swedish and English are related it is therefore to be expected that some MWSs are transferable between the two languages.
L1 also plays a role in psycholinguistic research measuring reaction times.Wolter and Gyllstad (2011) compared reaction times between English L1 speakers and Swedish speakers of L2 English for (i) collocations with translation equivalents, (ii) non-translation equivalents and (iii) unrelated items.Both the L1 and L2 speakers displayed significantly shorter reaction times for the collocations than for the unrelated combinations.However, the L2 speakers had significantly shorter reaction times for translation equivalents than for non-translation equivalents, a difference not surprisingly absent in the L1 data.This result implies an ease for the L2 speakers of processing translation equivalents that map onto corresponding expressions in L1, a finding which seems particularly relevant for two related languages such as English and Swedish (cf.Jarvis 2000 regarding L2 English by L1 Finnish and L1 Swedish speakers).

Data and methodology
This section starts with a presentation of the corpus, followed by a description of the two tasks: the identification of MWSs and the categorization of MWSs used in this paper.

The corpus
The material involved three groups of participants, 10 Swedes who had lived and worked in London for about 7 years, 10 Swedish university students of L2 English for English teachers, and 10 L1 speakers of English for control.The Swedes were thus advanced speakers of L2 English.The majority of the London Swedes (henceforth LS) and the L1 English speakers had an academic degree, whereas the Swedish University students (henceforth SU) were halfway through their academic English Teachers' program at the time of the recordings.The groups were matched for age, although the Swedish university students were somewhat younger (see Table 1).Both L2 groups had had nine years of L2 English at school and had been exposed to English from early childhood through input from computers, music, English films and TV programs.Nevertheless, English has the status of a foreign language in Sweden.Both the Swedish and English speakers had Swedish and English as their respective mother tongues. 3 The material is controlled in that all three groups were matched for educational background performing the same two tasks.Furthermore, they all belong to the same generation with an average age difference of six years between the Swedish groups see Table 1.
The LS and SU groups met the following criteria: aged between 20 and 38, with L1 Swedish, and L2 English primarily studied in the Swedish school system.The LS group was required to have at least five years' residency in the L2 country (the UK), while the SU group was allowed a maximum stay of three months in an Englishspeaking country, and no English-speaking parent or partner.The last two criteria  3 This is evident from questionnaires completed by the participants (see Appendix).
L1 and advanced L2 English speech would prevent L2 acquisition through exposure in the target language country for the SU group.

Two tasks
The participants performed two tasks: a role play between a legal expert at a company and her/his boss, and a retelling of a film clip of a silent movie seen on a computer screen.Both activities were recorded and transcribed.The role play involved two speakers, an employer (an external L1 speaker) and an employee (only the production of the employee was analyzed).The participants were given a written scenario, in which the employee was requested to call her/his employer and ask for two days off, since a close relative was getting married.However, the wedding day would clash with an important business meeting.The employee was given 5 min to read and contemplate the scenario and then make the call.The mean duration of the task was 5 min.The biggest challenge for the L2 user was to solve the problem over the phone.However, this task involved feedback from an L1 speaker, facilitating the participants' production (see Du Bois and Giora 2014 and Giora and Du Bois 2014 regarding dialogic resonance).
In the retelling task, the participants were asked to simultaneously retell a film clip of the first 14½ min from the silent film Modern Times with Charlie Chaplin shown on a computer.In this task the participants were not allowed any planning time and were only told to imagine that they were describing what they saw on the screen to someone who could not see it.This task thus involved a cognitively demanding situation and, unlike the role play, the participants only had visual support.Table 2 shows the number of words across the tasks and the groups (totaling 61,851 words).
The retelling task was expected to be demanding for all three groups through its monologic character, in part specialized vocabulary, and time constraint.This was assumed to result in fewer MWSs in the two L2 groups than in the L1 group (cf.earlier research e.g.Boers et al. 2006;Forsberg 2008;Forsberg and   3.3 Analytical procedure
The first analytical step involved scanning the transcriptions for possible MWS candidates.These were subsequently checked in Longman's CD dictionary (2005) based on 300 million words and the British National Corpus (henceforth BNC) based on 100 million words.As part of the identification we included MWSs inspired by Mel'čuk's (1998: 38) framework.An example from the material is the noun lever, collocating with pull or push.In other words, pull/push the lever would be the expected way to use 'lever' in the context of the film Modern Times (cf.Erman and Lewis 2015b).In Longman's CD dictionary (2005) both verbs cited above are listed as collocates of the noun 'lever', and in the BNC there are 20 examples.
Identifying MWSs in authentic texts, unlike any predetermined selection, involves evaluating MWSs intersubjectively combined with corpus data.Using this method large proportions of MWSs have been attested in spoken and written L1 and L2 English (Erman and Warren 2000;Jaglinska 2006;Lewis 2005Lewis , 2009;;Wiktorsson 2003).The researchers eventually reached an agreement.
It is essential to acknowledge the gradual and continuous change of language and language use, both in speech communities and among individuals (cf.Wray 2019: 267).Nevertheless, a core set of MWSs are shared by most people in a language community.Thus, communication is enabled among speakers in different English-speaking countries at a given point in time (see Erman and Warren 2000: 33;Erman et al. 2013: 80-83, for a discussion of the identification of MWSs).Some MWSs may go unnoticed because they are semantically transparent, and do not cause any problems of interpretation.However, they are difficult to acquire productively as they are non-salient (Lenko-Szymanska 2014: 229).

Categorization of MWSs
As mentioned, in this study we make a distinction between three categories of MWSs: MW phrases, MW utterances and metalinguistic MWSs (Erman and Warren 2000;Forsberg 2008;Jaglinska 2006;Lewis 2009;Wiktorsson 2003).Non-target combinations (e.g., 'lay a puzzle' for do a puzzle) have been excluded.The reason for their exclusion is that this study focuses on nativelike production and idiomaticity.An MWS embedded in a superordinate one, such as a bit of in the MWS I've got (a bit of) a problem was singled out and counted separately.
L1 and advanced L2 English speech

MW phrases
An MW phrase is a structure that has at least one element with a referential or semantic meaning.MW phrases occur in all kinds of texts and are made up of all phrase types, thus coinciding with grammatical phrase categories.Examples in our material include the following: These grammatical phrase categories have not been quantified.

MW utterances
Like MW phrases, MW utterances have elements with referential or semantic meanings.From a formal point of view MW utterances are grammatical clauses or reduced clauses.They have roots in the spoken language and are used in everyday situations and thus framed in discourse.Examples of MW utterances from our material include: I have no idea, that makes sense, that's a point.Examples of reduced grammatical clauses include: no problem, too bad.Some MW utterances are a starting point when learning an L2 as apparent in textbooks.

Metalinguistic MWSs
Metalinguistic MWSs include structures that function as operators on or qualifiers of other structures.Unlike MW phrases and MW utterances, metalinguistic MWSs are not a semantic but a functional category.They have different functions in discourse, e.g., as text-organizers, marking the speaker's attitude, keeping the floor, clarifying intentions.This category thus includes many different kinds of linguistic expressions.Examples of metalinguistic MWSs from our material include: here we go, 'cause obviously, I was meaning to ask, you see, and so on, here we are.Metalinguistic MWSs may also function as downtoners, e.g., you know, sort

Noun phrases:
a close-up shot, corn on the cob, next week, metal nut Verb phrases: keep up, file one's nails, wipe one's mouth, hang around Preposition phrases: at the last minute, down the street, in the morning, Adjective and adverb phrases: black and white, worse for wear, once again, all of a sudden of, I suppose, I guess, a bit of.It is the context that ultimately determines their function.

Equivalents (EQs)
Cutting across the three categories above are EQs.The main criterion for an MWS to be categorized as an EQ is that there is a semantic, structural and functional match between the English and Swedish structure.For example, the English MWS for all involved has a word-for-word structural, semantic and functional match in the Swedish MWS 'för alla inblandade' and would be used in similar contexts, hence an EQ.EQs are thus perfectly acceptable English and Swedish MWSs.
An example of a different kind is an MWS which changes meaning depending on context such as the English MWS run out, which in a concrete sense is equivalent to the Swedish phrase 'springa ut'.However, in run out of petrol there is no equivalence to the Swedish corresponding phrase 'bensinen tar slut' (English 'the petrol takes end').In other words, run out in run out of in the example above is non-transferable between English and Swedish.Other typical examples of nontransferable MWSs between Swedish and English are common MW utterances, for example, how are you?, Swedish 'hur mår du' (English 'how feel you') and I'm fine thanks, Swedish 'tack jag mår bra' (English 'I feel fine').

Data analysis
Since this study examines only one variable (exposure) the chi-square test has been used. 4The threshold for significance is set at p < 0.05.
The importance of the MWS category usually tallied with its size, so that a quantitatively large category is more relevant than a small one.MW utterances constitute a large category in the role play, securing that the dialogue runs smoothly.Metalinguistic MWSs constitute another important category in this task, adding to the rapport between the interlocutors.MW phrases are a large category in the retelling task pertaining to lexical precision in the narrative and moving the story forward.The results are presented in the order of the research questions starting with the quantity and distribution of the MWS categories, followed by the results for EQs.L2 results that are non-significant in relation to the L1 group are considered nativelike, which is the term used in the following sections.

RQ1 quantity and distribution of MWS categories
The quantity of MWSs was measured in numbers of types (N in the tables below) and MWS types per 100 words.The distribution is displayed over the three main categories of the study, MW phrases, MW utterances and metalinguistic MWSs in the two tasks (Tables 3 and 4).

The role play
Both L2 groups were nativelike in their use of MW phrases in the role play.In fact, the LS group was nativelike on all three categories, confirming our expectation.
The SU group used more phrases than utterances and significantly fewer utterances than the other two groups, thus displaying a non-nativelike distribution of categories in this task (see Table 3).Table 3 also shows that the SU group used significantly fewer metalinguistic MWSs than the other two groups.Because of its elusiveness this category is presumably harder to acquire than the other two.Both MW utterances and metalinguistic MWSs are characterized by social involvement, which is adequate for the fulfillment of this interactive task.
The fact that the LS group was nativelike on MW utterances in the role play corroborates Kecskés's (2002) claim that residency in the target language country has a positive effect on the acquisition of utterances appropriate for the situation.The results for the SU group showed a lack of MW utterances and metalinguistic MWSs, which indicates a lack of practice even at advanced levels.In fact, MW utterances and metalinguistic MWSs may be rare at advanced levels in an educational setting.
The result for the LS group points in the same direction as two previous studies involving the same material as in the present study, viz.voiced pausing (Erman and Lewis 2013) and advanced vocabulary knowledge (Erman and Lewis 2019).In both studies the LS group performed like the L1 group in the role play task.

The retelling task
The largest category in the retelling task was MW phrases.In accordance with our expectations the two Swedish groups used significantly fewer types of MW phrases than the L1 group (see Table 4).
MW phrases are tied to topics and domains (cf.Hoey 2005;McCarthy and O'Dell 2005) and learnt throughout life.As has been shown MW phrases in spoken production are a hurdle even for very advanced L2 speakers (cf.Cummins 1981;Boers et al. 2014;Forsberg Lundell and Lindqvist 2014).Regarding MW utterances and metalinguistic MWSs, which are minor MW categories in this task, there were no significant differences between any of the groups.

Individual variation
The importance of studying individual variation has been emphasized in several earlier studies in SLA (Durrant 2008;Forsberg Lundell and Lindqvist 2014;Forsberg Lundell and Sandgren 2013;Lindqvist et al. 2011;Nesselhauf 2003;Shaw and McMillion 2011).
In Tables 3 and 4 the results for the two tasks have been shown for the groups, which overshadows individual performance.In Table 5 individual results are shown in normalized numbers per 100 words for the largest categories of MWSs in each task, i.e., MW utterances in the role play task and MW phrases in the retelling task.The results for individual variation largely reflect the group results.Table 5, however, gives a more fine-grained picture showing that there are individual nativelike results in both L2 groups.
In the role play task the LSs and L1 speakers had overlapping results, the majority of the individuals being in the spans above 4.0 MW utterances per 100 words.Most of the SU individuals are in the 2.1-4.0 span, where we also find one LS and two L1 speakers.There are two SU individuals in the same spans of normalized numbers as most of the L1 speakers and LSs.
In the retelling task all the L1 speakers produced 5.5 MW phrases per 100 words or more, while the majority of the L2 speakers produced less than 5.5 MW phrases per 100 words.However, the individual results for MW phrases show that three LSs and one speaker from the SU group performed within the L1 range.The individuals in the LS group, in contrast to the individuals in the SU group, were more dispersed across the spans in this task, which may be explained by their different experiences of English in the UK.

RQ2 equivalents
English and Swedish, both being Germanic languages and thus related, share, not only many single words, but also many MWSs.In the role play task there were no significant differences in the proportion of EQs between any of the groups (see Figure 1).This result contrasts with the results for the retelling task, in which both L2 groups used significantly larger proportions of EQs than the L1 group ( p < 0.000), which was in accordance with our expectations.There was no significant difference between the two L2 groups in this task.
In accordance with our expectation the L2 groups used significantly more EQs in the retelling task.The fact that the retelling task was a challenge for the L2 speakers was thus shown in the proportion of EQs.The largest difference in proportions of EQs between the two tasks was found in the SU group; in fact, the difference was close to the threshold for significance (<0.06).The L1 group used almost identical proportions of EQs in the two tasks.In other words, the type of task did not affect their EQ results.
These results are in line with earlier findings, notably that expressions and words that are shared between two languages are more easily acquired and accessed (Granger 1998;Kroll et al. 2002;Lotto and de Groot 1998;Wolter and Gyllstad 2011).In the role play task, however, both L2 groups used proportions of EQs that were similar to those of the L1 group, which can partly be explained by the character of this task, in which the two categories MW utterances and metalinguistic MWSs are not as readily transferable between the two languages as MW phrases.

Discussion and conclusion
Here we recapitulate the most central findings with regard to the use of MWSs in the two tasks, one interactive, the role play task and one narrative, the retelling task.The material consisted of the spoken production of two L2 groups, one group L1 and advanced L2 English speech living in London, the LS group, and the other studying English at a Swedish university, the SU group.An L1 English group was used as benchmark.The analysis of MWSs included the quantity of MWSs on group and individual levels, which was our first research question.Our second research question addressed the role of the L2 participants' L1, Swedish, measured in EQs (equivalent MWSs in English and Swedish).The type of task turned out to have a strong impact on all the results.Likewise, the MWS categories (MW phrases, MW utterances and metalinguistic MWSs) proved to distinguish the very advanced L2 speakers from the L1 speakers, in particular in the use of MW phrases in the retelling task.
As expected the LSs behaved like the L1 group on all categories in the role play, while the SU group used more MW phrases than MW utterances and significantly fewer MW utterances and metalinguistic MWSs than the L1 and LS groups.The latter two MWS categories fulfill important functions in interaction.In the retelling task both L2 groups used significantly fewer MW phrases than the L1 group, which was expected.However, they were nativelike on the use of MW utterances and metalinguistic MWSs, two minor categories in this task.Another aspect of the first research question concerned individual variation.The results showed that some L2 individuals performed within the L1 speaker range on both tasks, in particular from the LS group.
The fact that the retelling task was a challenge for the L2 speakers was also shown in the proportion of EQs.In this task both L2 groups used significantly higher proportions of EQs than the English L1 speakers.The results are in line with earlier findings, notably that expressions and words that are shared between two languages are more easily acquired and accessed (Granger 1998;Wolter and Gyllstad 2011).In the role play task, however, both L2 groups used proportions of EQs that were similar to those of the L1 group, which can partly be explained by the character of this task, in which the two categories MW utterances and metalinguistic MWSs are not as readily transferable between the two languages.
The results for the role play task point to the importance of exposure to the target language (cf.Ellis 2006).Living and working in the target language country benefitted the LS group with regard to all three categories in this task.This contrasts with the results for the retelling task involving many MW phrases, which distinguished both L2 groups from the L1 group.Exposure thus does not seem to improve the quantity of MW phrases (cf.Forsberg Lundell and Lindqvist 2014).MW phrases in particular develop in tandem with general, encyclopedic knowledge, which L1 speakers have acquired from early childhood.The implication is that the more we learn about different areas and topics in L1 or L2 the more associative links are created (Ellis 1996;Hoey 2005).As apparent from the results of the LS group in the role play task, exposure and practice promote the use of MW utterances and metalinguistic MWSs appropriate for the situation (Kecskés 2002).Furthermore, the needs of the speaker have to be considered for their acquisition in a second language (Wray 2002).For example, a recent immigrant, a resident and a student have different needs and as a consequence different repertoires of MWSs.In line with results from previous research our results have shown that an L2 MWS with a matching MWS in L1 facilitates and supports language production.
Unlike several studies on formulaic language, the present study had control of topic and situation, both of which ultimately affected the use of MWSs and enabled comparisons between groups.The main contribution of the study is that it highlights MWSs used in spoken production in controlled situations, in contrast to the multitude of studies focusing on receptive knowledge of MWSs (Wolter 2020: 502).Transcribing spoken data is time-consuming, which partly explains why the analysis of running speech is scarce.

Margareta Lewis
Department of English, Stockholm University, Stockholm, Sweden margaretalewis8@gmail.comMargareta Lewis (PhD in English linguistics in 2008) is affiliated with the English Department at Stockholm University.In 2015 she and her colleagues at an adult college received 'the Stockholm's award' for innovative evidence-based reading project for L2 learners of English.Her thesis and later research include formulaic language, syntactic complexity and vocabulary in spoken and written L2 production, all within L2 acquisition.Her work has appeared in various journals and books.
Nine years at school and an average of  years' residency in London  (-) Ten L English Swedes resident in Stockholm Nine years at school and one year of full-time studies at an English department in Sweden  (-) Fant 2010 for spoken L2; Groom 2009; Lewis 2009 for written L2).

Figure 1 :
Figure 1: Percentages of equivalents of all MW categories in the role play task and the retelling task across the L1, LS and SU groups.

Table  :
Number of words across tasks and groups.

Table  :
Role play; group comparison of the quantity of types of MW phrases, utterances and metalinguistic MWSs; ns = non-significant.

Table  :
Retelling task; group comparison of the quantity of types of MW phrases, utterances and metalinguistic MWSs; ns = non-significant.

Table  :
Distribution of individual results for types of MW utterances in the role play task and MW phrases in the retelling task per  words.