A Corpus-based Study on Question Tags in ELF. An Attempt to Determine an Emergent Construction

Recent research on English as a Lingua Franca (henceforth ELF) has indicated that ELF can be analyzed in accordance with the main principles of the Construction Grammar approach (Pirc 2013). Admittedly, while the majority of studies have been devoted to its creative and unpredictable character, little attention has been paid to how ELF can be conceptualized if its emergent elements are treated as constructions. Thus, the present paper conducts a corpusbased study on the Question Tag Construction (henceforth the QTxC) in ELF with a view to indicating in what way the QTxC can be determined. With the support of the VOICE and BNC corpora, the performed analysis has shown that the analyzed realizations of the QTxC in ELF, while having formal properties that deviate from the norms of English, can also be characterized as exhibiting various sub-types of the prototypical function of the QTxC.


Introduction
In the era of globalization, in which speakers of diff erent mother tongues travel between various communities, English has become a new lingua franca (Seidlhofer 2007) and is the only language that has spread around the world to such an extent (Björkman 2013;Melitz 2018). Since it is seen as a shared language for people who do not have common linguistic backgrounds, ELF is often conceptualized as a variety in its own right (Fiedler 2011) with the norms negotiated by its users (Seidlhofer 2001;Meierkord and Knapp 2002). Thus, due to its dominant role, a great number of studies have been devoted to emphasizing its creative character as well as indicating its preliminary lexicogrammatical features, among which the category of an invariant question tag is included (Seidlhofer 2004, 220). It has been suggested that ELF speakers exhibit a tendency to use an invariant question tag (isn't it) since the complexity of question tags in English seems to be "disproportionate to their relatively simple communicative function" (Seidlhofer 2011, 161). Moreover, along with the growth of research on ELF, it has been claimed that ELF can be analyzed in accordance with the general principles of the Construction Grammar approach (Pirc 2013). However, although Pirc (2013) has analyzed a few examples of the preliminary lexicogrammatical features of ELF as constructions, she has not examined the category of a question tag with reference to the main tenets of the Construction Grammar theory. Hence, the present paper aims to conduct a corpus-based study with a view to indicating in what way the QTxC in ELF can be determined.

Theoretical background of ELF research
Initial attempts to provide the characteristics of ELF have shown that it is a challenging task to defi ne the unstable "beast" that ELF appears to be (Pitzl 2018, 10). Hence, because of its unpredictable nature, "no coherent and comprehensive lingua franca model has been proposed" (Seidlhofer 2001, 140) and various defi nitions of ELF have been suggested (see Firth 1996;Seidlhofer 2001;. However, despite its instability, the postulate of its outstanding creativity resulting from changing communicative needs among speakers is widely accepted. Therefore, ELF is mostly seen as developing in a bottom-up manner in which tentative regularities can be derived from the actual utterances of ELF speakers. Consequently, several corpora have been created, e.g. Vienna-Oxford International Corpus of English or the Corpus of English as a Lingua Franca in Academic Settings, and function as the empirical base on which ELF research should be grounded. However, since a detailed account of studies devoted to ELF is beyond the scope of the present work, only a brief overview is presented. The list of preliminary lexicogrammatical features of ELF is provided by Seidlhofer (2004, 220), who claims that the elements, even though seen as deviant from the norms of English, appear to be comprehensible in the effi cient-oriented communication of ELF speakers. The creation of such forms is said to be motivated by a number of dominant processes, e.g. approximation, simplifi cation, or exploiting redundancy (see Cogo and Dewey 2012;Mauranen 2018), which result in such changes as, for example, the omission of both -s in the third person singular (he look very sad) and articles (our countries have signed agreement), the use of relative pronouns interchangeably, irrespective of the type of a noun, and the creation of an invariant question tag (isn't it) (Nagy 2016). Moreover, ELF speakers tend to use the non-standard forms of verbs (teached) as well as idiomatic expressions (Pitzl 2018) and rely on lexical rather than grammatical elements during interactions (Alptekin 2013). Overall, the negotiated norms appearing in ELF are often conceptualized as hybrid variants that are created to construct shared meanings (Seidlhofer 2009;Mauranen 2012) underlying the emergence of a variety in its own right. Consequently, the creativity of ELF speakers results from the need to modify the utterances so as to facilitate mutual comprehension in the context of ELF interactions. However, since the main focus of the present paper is the analysis of question tags in ELF, it is worth presenting how the category of a question tag is understood in relation to the norms of English in order to illustrate how the negotiated regularities of question tags emerge in ELF.

The category of a question tag
According to the norms of English, a question tag, commonly used to elicit confi rmation, is seen as a short clause, embedded to the main one, that transforms a statement into a question. Its form consists of a pronoun and an auxiliary verb which agrees with the auxiliary verb of the main clause and is "mirror-like to it in that the verb changes from positive to negative or vice versa" (Vettorel 2014, 144, after Crystal 2003. Unsurprisingly, in ELF, the complexity of question tags seems to be "disproportionate to their relatively simple communicative function" (Seidlhofer 2011, 161). Thus, the process of simplifi cation underlies the application of a form that is considered more economical (Hülmbauer 2009) Moreover, it has been noticed that this form seems to be easier for people who speak languages in which question tags have forms diff erent from the English ones, e.g. German (nicht?) or Italian and Spanish (no?) (Vettorel 2014, 145). As stated by Seidlhofer (2001, 138), it is "important to realize that native-speaker language use is just one kind of reality, and one of very doubtful relevance for lingua franca contexts". In this view, instead of comparing ELF to the norms of English, it has been claimed that research on ELF should be conducted with the attention paid to fi ndings concerning World Englishes, pidgins, and creoles (Björkman 2013). In the aforementioned studies, apart from the analysis of an invariant question tag (isn't it), the most frequently described question tags are yes or not ?, yes?, right?, no?, and okay? (see Takahashi 2014;Täuschel 2014;Criado-Peña 2016;Percillier 2016), which stresses the tendency to apply the other forms of question tags that still maintain their prototypical function. Therefore, it seems that the adoption of the traditional defi nition of question tags in English may lead to the omission of diff erent elements that function as question tags in ELF. In other words, since research on World Englishes, pidgins, and creoles emphasizes the tendency to use the forms inconsistent with the norms of English, it seems that the analysis of question tags in ELF with the focus on the important role of their function may facilitate the establishment of certain regularities concerning the aforementioned category. In fact, it has recently been proposed that the analysis of ELF may be conducted with the adoption of the Construction Grammar approach (Pirc 2013). In accordance with this theory, constructions are defi ned as form-meaning/function pairings (Hoff mann and Trousdale 2013;Fried 2015), which allows for taking into account the crucial role of function as an integral part of a question tag. Thus, since the present paper relies on the assumptions of the Construction Grammar theory, the next section provides both the description of the main tenets of the Construction Grammar approach and the indication in what way the Construction Grammar approach complies with the studies on ELF.

Construction Grammar and ELF
Due to the fact that there seems to be no uniform theory concerning the Construction Grammar approach, I limit my discussion to the usage-based models that are seen as corresponding to ELF studies (see Pirc 2013). First of all, the proposals may be characterized by the main shared tenets that language is not an innate cognitive system and all constructions, understood as the basic units of linguistic analysis, are emergent by means of general cognitive processes. Secondly, along with the perspective that Construction Grammar is a usage-based theory (see Croft 2001;Goldberg 2006), the constant emergence of new constructions is seen as a common phenomenon in which "the speech events with their context and discourse environment infl uence the form of the constructions" (Pirc 2013, 59). Moreover, instead of off ering a clear-cut separation of lexicon and syntax, Construction Grammar assumes that all constructions, from basic units (such as morphemes) up to complex schematic patterns (Croft 2001, 17), are the part of the lexicon-syntax continuum (Hoff mann and Trousdale 2013). Consequently, they are organized in taxonomic networks "in which more abstract constructions are stored as superordinate nodes to more specifi c instances" (Hoff mann 2018, 268). Thus, the Construction Grammar approach provides a description of constructions with their "morphological, syntactic, lexical, semantic, pragmatic, or discoursefunctional properties" (Croft 2001, 25) and aims to capture "cross-linguistic generalizations in terms of form-meaning/function constellations" (Fried and Östman 2004, 24). Therefore, while Goldberg (2013) states that cross-linguistic regularities may be explained by both domain-general cognitive processes and the function of a construction, Croft (2001;2003) stresses that constructions are language-specifi c since functionally similar or equivalent constructions usually diff er in their syntactic properties. However, the majority of Construction Grammar theories focus on monolingual descriptions that mainly concern native languages. The attempt to go beyond monolingual models is proposed by Höder (2012;, who focuses on the type of constructions resulting from the blend of at least two languages, i.e. diaconstructions. In his view, the emergence of diaconstructions is said to be motivated by the process of interlingual identifi cation of similar elements in languages a multilingual person speaks. Nevertheless, it seems that the Construction Grammar analysis which would focus on the superordinate processes and distinctive features of ELF has not been proposed so far. As Pirc (2013) notes, the Construction Grammar approach seems to comply with research on ELF for several reasons. Since ELF is commonly conceptualized as developing in a bottom-up manner, in which various non-standard constructions arise due to the communicative needs, it corresponds to the Construction Grammar theory stressing the usage-based model of language in which the forms of constructions are adapted to a given context. Furthermore, due to the main tenet that all constructions, from prototypical to more peripheral, are considered the basic units of linguistic analysis showing the characteristics of a given language, the adoption of the Construction Grammar approach to ELF allows for analyzing all emergent elements in such a way that could indicate tentative features by means of which ELF may be defi ned. In other words, since ELF is defi ned as having its own principles and, at the same time, the forms of constructions are seen as shaped by the context in which they are used, research on constructions emerging in ELF could indicate certain regularities of ELF constructions. However, as Pirc (2013) notices, ELF is still an emerging phenomenon; thus, it seems that the characteristics of constructions in ELF can be determined for each interaction individually. Nevertheless, since ELF speakers are said to exhibit the general ability to adapt the forms to the needs of ELF contexts (Cogo and Dewey 2012) and scholars aim for the establishment of general tendencies of ELF at various levels of linguistic analysis, it seems that certain regularities concerning constructions in ELF can also be delineated. Moreover, since it is claimed that the forms appearing in ELF should not be analyzed in separation from their function (see Cogo 2008;Björkman 2013), the need for the Construction Grammar approach to ELF seems to be emphasized. Therefore, heading towards the Construction Grammar approach to ELF, the establishment of tentative regularities regarding the QTxC, along with the study conducted by Pirc (2013), may be seen as its starting point.

The description of the study
As shown above, the complexity of question tags in English leads to the application of the negotiated form in ELF (isn't it). However, the question arises whether the adoption of a view in which a question tag is defi ned as a form-meaning/ function pairing may lead to the extension of the question tag category by other forms which, in addition to the above-mentioned invariant question tag (isn't it), are used by ELF speakers to elicit confi rmation. Consequently, since it is believed that ELF research should not be conducted in separation from the studies on World Englishes, pidgins, and creoles, the present paper focuses primarily on the analysis of the most frequently described question tags in the aforementioned studies, i.e. okay?, right?, yes?, no?. Secondly, the paper aims to establish whether there are other forms in ELF that can be regarded as the realizations of the QTxC. Moreover, since it is assumed that ELF speakers tend to apply lexical rather than grammatical elements, the next question arises whether the forms such as okay? right? yes?, and no? are more frequently applied than an invariant question tag (isn't it), consisting of a pronoun and an auxiliary verb. At the same time, if ELF is said to be a variety in its own right, it is worth establishing whether the analyzed forms of question tags are more common in ELF than in the English language. Thus, because methods which are only quantitative are said to inhibit a detailed description of the distinctive features of ELF (Björkman 2013) and "qualitatively oriented studies are now becoming ELF mainstream" (Kalocsai 2013), the present study combines two methods due to its underlying aim to indicate both certain regularities of the QTxC and the diff erence in the application of the QTxC between ELF and English.
Since the use of corpus data is "in no way unique to work in Construction Grammar" (Hilpert 2013, 458) and considered helpful as the empirical base for ELF research, the present study relies on the data collected from the VOICE corpus, which is the fi rst computer-readable inventory of ELF interactions. It includes the transcript of naturally occurring speech of approximately 1,250 speakers of 50 diff erent mother tongues (in total 1,023,127 words) and comprises of recorded interactions in various contexts, among which the most dominant are social sciences (29%) and technology (19%) (Björkman 2013). The data collected from the VOICE corpus is compared to the data available in the British National Corpus (BNC) that comprises of 96,263,399 words of spoken (10%) and written (90%) British English in various domains. Nevertheless, the corpus search in BNC is narrowed to the spoken data (in total 9,963,663 words) which complies with the spoken data available in the VOICE corpus. Moreover, due to a great diff erence in the total number of words between the two corpora, the normalized frequency (per 1,000,000 words) is calculated for each form in both corpora. However, it should be noted that the preliminary search in the VOICE corpus has shown that question tags are not always conclusively tagged by means of the question mark following a given form of a question tag. Thus, if a form without a question mark exhibiting the function of eliciting confi rmation is spotted, it is added to the total frequency of a given form. 1 Next, in order to measure the statistical signifi cance of the diff erences between the raw frequencies of the forms in two corpora, the log-likelihood test is used (available at: http://ucrel.lancs.ac.uk/llwizard.html).
Overall, with the adoption of the Construction Grammar approach, the present paper conducts a corpus-based study on the QTxC that entails the following steps: 1) The retrieval of the elements eliciting confi rmation, i.e. right, okay, yes, no, from the VOICE corpus, 2) The retrieval of other possible realizations of the QTxC from the VOICE corpus, 3) The analysis of the realizations of the QTxC from steps 1 and 2 with a view to indicating the regularities of both form and function, 4) The determination of the raw frequency of an invariant question tag (isn't it) in the VOICE corpus, 5) The determination of the raw frequencies of the forms analyzed in the steps above in the British National Corpus, 6) The calculation of the normalized frequencies of analyzed forms in both corpora, 7) The comparison of the diff erences between the raw frequencies of forms in the VOICE and BNC corpora by means of the log-likelihood test.
The fi rst element that appears to function as the QTxC is mentioned in examples (1a) and (1b), i.e. the lexical element right?. As shown in line 760 and 16, the form of the QTxC right appears to be the ellipted form of a question (Am I right?/Is it right?). In example (1a), S3 applies the QTxC to elicit confi rmation of whether S3 is right (or not) about the aspects concerning people from Malta. Likewise, in example (1b), S1 wants to make sure whether S1 is right about their plans. Thus, since the aforementioned realization of the QTxC exhibits the same formal and functional regularities in diff erent conversations, it can be initially assumed that the QTxC right is used to elicit the sub-type of refl exive confi rmation.
(2) copy it on the: er word fi le. (13)  Examples (4a) and (4b) show a diff erent lexical element that is used as the QTxC. In the fragments above, S4 and S1 apply the lexical element no to elicit confi rmation. In example (4a), S4 aims to make sure whether S4 is right that the timetable belongs to S4. In example (4b), S1 tries to elicit confi rmation about S1's assumption about the present situation. Similarly to right and yes, the lexical element no appears to have the function of eliciting refl exive confi rmation and, unlike right, does not seem to be the example of any reduced form. However, in example (4b), the lexical element remember, which appears later in the same conversation, is also marked in bold. In the present fragment, it is used by S1 to make sure whether the person who is talking with S1 can follow what S1 is referring to. Therefore, it seems that this element can also function as the QTxC that is applied to elicit confi rmation concerning the knowledge of interlocutors. The present fragment shows similar properties of the QTxC as in example (4b) in which, along with the QTxC no, the lexical element remember is presented. While talking about the linguistic aspects concerning English and Maltese, S2 uses the lexical element remember to ask for confi rmation. Likewise, this confi rmation seems to be listener-directed (S3), since S2 wants to fi nd out whether S3 knows what S2 is talking about. With reference to the form of the QTxC in both examples (4b) and (5), it appears that remember is the ellipted form of a question (Do you remember?).
(.) EDint331:114 S2: so you have managed to: to translate all <3> these technical terms </3> EDint331:115 S1: <3> no no the-these were </3> NOT <4> translated </4> EDint331:116 S2: <4> ah okay yes </4> yes yes @ EDint331:117 S1: they just (1) WROTE them (.) understand?{S1 gets up and takes books from a shelf} i <5> can show </5> you some of them <6> over here </6> EDint331:118 S2: <5> okay hh </5> In the conversation EDint331, on which the description of the QTxC yes is based, it is possible to notice another realization of the QTxC that shows similar properties as the QTxC remember. In lines 113 and 117, S1 applies the lexical element understand to fi nd out whether S2 can follow what S1 is talking about; thus, it has the function of listener-directed confi rmation. Moreover, similarly to the QTxC remember and right, the QTxC understand seems to be the ellipted form of a question (Do you understand?).

Discussion
The performed analysis allows for establishing certain regularities concerning both qualitative and quantitative aspects of the QTxC. Due to the adoption of an approach that goes beyond the traditional understanding of a question tag, it is possible to notice other forms which, in addition to an invariant question tag (isn't it), are used to elicit confi rmation in ELF. In accordance with the fi ndings concerning World Englishes, pidgin, and creoles, there is a noticeable preference for applying other forms that are said to deviate from the English standards (the quantitative aspect of the analysis is presented in Table 1 and Figure 1).  Table 1 shows the raw frequencies of forms together with their normalized frequencies (per 1,000,000 words) in both corpora. Moreover, it includes the value of the LL ratio that indicates whether the diff erences between the raw frequencies in the VOICE and BNC corpora are statistically signifi cant. Having calculated the normalized frequencies of elements in both corpora, the immediate observation is that ELF speakers tend to rely on the forms such as right ?, okay?, yes?, no? to elicit confi rmation. In all cases, the normalized frequencies of nonstandard forms in VOICE outnumber their normalized frequencies in BNC. The comparison of the normalized frequency of an invariant question tag (isn't it) shows that isn't it in VOICE is the least frequently applied form whereas in BNC it is the most frequently used tag among the ones analyzed in the paper. In both corpora, the forms remember and understand are not frequently applied; however, their normalized frequencies are greater in ELF. Moreover, it is possible to notice the diff erence between the normalized frequencies of the forms no? and yes? in VOICE and BNC. In VOICE, the former has a higher value of normalized frequency than the latter whereas the latter appears in BNC more often than the former (the values of normalized frequencies are summarized in Figure 1).  Moreover, the values of the LL ratio indicate that in almost all cases the diff erence between the VOICE and BNC corpora is statistically signifi cant, i.e. the LL ratio is above 3.84 at the p <0.05 level (Rayson and Garside 2000). Only the diff erence concerning one form, i.e. the QTxC remember, is not statistically signifi cant (0.17 < 3.84).
Thus, it demonstrates that ELF is a distinct phenomenon from the English language and develops the negotiated norms characteristic of a variety in its own right. In fact, it complies with the assumption that the norms of native languages are said to be questionable in relation to lingua franca contexts (see Seidlhofer 2001). Thus, if qualitative methods of analysis may have an impact on the development of ELF studies and, at the same time, the need for a Construction Grammar approach is emphasized, it is worth presenting a number of regularities concerning both the form and function of the analyzed realizations of the QTxC in ELF (the results are summarized in Table 2). As indicated in Table 1, the analyzed realizations of the QTxC in ELF concern the application of one lexical item that is used to elicit confi rmation. Thus, it can be assumed that, apart from an invariant form (isn't it), the dominant realization of the QTxC in ELF has its form -the [lexical element] QTxC with the most schematic function of eliciting confi rmation. In fact, it corresponds to the theory proposed by Alptekin (2013), who stresses the superiority of lexical elements in the context of ELF. However, while the analyzed examples exhibit formal realizations deviating from the norms of English, they also appear to exibit various sub-types of question tag function, i.e. listener-directed, refl exive, and reciprocal confi rmation. Furthermore, in relation to the dominant processes underlying ELF, it is possible to notice that the emergence of certain question tags, i.e. right? understand? remember?, is motivated by the mechanism of simplifi cation (it underlies the application of a reduced form of a question that is substituted by one lexical element). Undoubtedly, with reference to the previous fi ndings, the results of the analysis exemplify the attempts of ELF speakers to apply as simple forms as possible to facilitate mutual understanding among them.
On the other hand, the adoption of the Construction Grammar approach may lead to the new direction of studies on the dominant processes in ELF. As noticed by Alptekin (2013), there is a need for cognitively-oriented research on the relationship between the context of ELF and mental processes and representations. Thus, if constructions are said to be acquired by means of general cognitive processes, it appears that research on constructions appearing in ELF may also point to certain cognitive regularities regarding the variety in its own right.

Conclusion
The paper focuses on the analysis of question tags in ELF. While it has been noted that ELF speakers exhibit a tendency to apply an invariant question tag (isn't it), the author aims to go beyond the present theories and tries to answer the question whether the category of a question may be extended by other forms which, in addition to the above-mentioned invariant question tag (isn't it), are used by ELF speakers to elicit confi rmation. Moreover, the author aims to establish which forms of question tags are the most frequently applied in ELF and whether there is a discrepancy between the application of the analyzed forms between ELF and the English language. With the adoption of the Construction Grammar approach, which allows taking into consideration the crucial role of function as an integral part of a question tag, and with the reliance on the fi ndings concerning World Englishes, pidgins, and creoles, which are considered helpful in ELF research, the performed analysis has revealed certain regularities concerning both qualitative and quantitative aspects of the Question Tag Construction.
It has been shown that ELF speakers use forms such as right? (nf. 233.6), okay? (nf. 255.8), yes? (nf. 107.5), no? (nf. 167.1) to elicit confi rmation more often than an invariant question tag (isn't it) (nf. 37.1) At the same time, it has been presented that an invariant question tag (isn't it) (nf. 323.2) is conversely the most frequently applied form in BNC from the ones analyzed in the paper (the quantitative aspect of the analysis is presented in Table 1 and Figure 1). Moreover, the raw frequencies of the analyzed forms have been compared by means of the log-likelihood test to establish whether the diff erences between corpora are statistically signifi cant. From 7 forms that are analyzed in the paper, the diff erence between 6 forms is signifi cantly diff erent (the LL ratio > 3.84). Thus, it has been shown that the norms of the English language seem to be questionable in the context of ELF and that ELF speakers negotiate their own norms concerning the use of forms applied to elicit confi rmation. Moreover, the analysis has shown that the analyzed realizations of the QTxC in ELF, while having formal properties that deviate from the norms of English, can also be characterized as exhibiting various sub-types of the prototypical function of the QTxC, i.e. listener-directed confi rmation, refl exive confi rmation, and reciprocal confi rmation; however, the forms used to elicit listener-directed confi rmation are not frequently applied (the results are summarized in Table 1).
Furthermore, the need for ELF studies conducted in accordance with the Construction Grammar approach is emphasized. While most of the Construction Grammar theories focus on the monolingual descriptions of native languages, research on constructions in ELF may indicate certain regularities of lingua franca constructions. However, it should be noted that the analysis of the QTxC, along with the study conducted by Pirc (2013), appears to be the starting point of research on constructions in ELF. Thus, the claims above are only tentative suggestions for the Construction Grammar approach to ELF.

Notes
1 However, as a consequence of the limitation resulting due to the lack of uniform tagging, some forms may be omitted in the total frequency of a given realization of the QTxC. 2 Due to space limitations, the analysis presents only two examples of each realization.