Innovation at the Grammar-lexis Intersection in World Englishes 1

It has often been claimed that the English complementation system lies at the intersection of grammar and lexis, an area where, in the process of nativization in new varieties of English, innovation and change commonly occur (Olavarría de Ersson and Shaw 2003; Schneider 2007). The current study aims to contribute to this field by looking at the complementation profile of the verb REGRET as used on the internet ( Corpus of Global Web-Based English (GloWbE), Davies 2013). The manual analysis of more than 18,000 examples unveils new distributions, uses and patterns of complementation with this verb in 16 different L1 and L2 varieties of English worldwide, for which tentative cognitive explanations are offered.


Introduction
The diachronic evolution of the English complementation system has been widely studied (see Warner 1982;Fischer 1988Fischer , 1989Fanego 1990Fanego , 1992Fanego , 1996aFanego , 1996bFanego , 1998Fanego , 2004aFanego , 2004bFanego , 2007Fanego , 2010Fanego , 2016Rohdenburg 1995Rohdenburg , 2006Rohdenburg , 2014Rudanko 1998Rudanko , 2000Rudanko , 2011Miller 2002;Los 2005;Vosberg 2006;De Smet, 2008, 2010, 2013. However, complementation has not received much attention in the area of World Englishes, despite it being considered an area prone not only to variation and change but also to innovation (Mukherjee and Hoffmann 2006: 148;Schneider 2007: 86). Notable exceptions are studies that look at frequency and distributional variation in the various canonical patterns of the complementation system, mainly the ditransitive pattern and the frequency of to-infinitives and gerunds following a large number of verbs (cf. Olavarria de Ersson and Shaw 1 I am very grateful to Elena Seoane for her help with interrater reliability tests and her comments on an earlier draft of this paper, and to the anonymous reviewers for their valuable comments. For funding, my gratitude goes to the Spanish Ministry of Economy and Competitiveness (grant FFI2017-82162-P) and the University of Vigo. institutionalized L2 varieties of English but also native and learner varieties as well as pidgins and creoles. The term "World Englishes" is also considered the most neutral term and is the one that gives the name to the research subdiscipline concerned with the study of these varieties which emerged in the 1980s.
The new varieties of English differ from the standard one, as codified in grammars of English, at all levels of analysis: phonological, lexical, grammatical, pragmatic, and so on. According to Schneider (2007: 86), many innovative patterns tend to occur "at the intersection between grammar and lexis" and "a classic example is the complementation patterns which verbs and adjectives typically enter." On similar lines, Olavarría de Ersson and Shaw (2003: 138) describe complementation as "an all-pervading structural feature of language and thus likely to be more significant in giving a variety its character than, for example, lexis." Hence, verb complementation is an area of innovation and change which merits attention, as shown in previous studies dealing with ditransitive verbs (e.g. provide, supply, give, send…) and the variation between to-infinitive and -ing/gerund-complement clauses (cf. Olavarria de Ersson and Shaw 2003; Mukherjee and Hoffmann 2006;Mukherjee and Schilk 2008;Mukherjee and Gries 2009;Schilk, Bernaisch and Mukherjee 2012;Schilk, Mukherjee, Nam and Mukherjee 2013;Bernaisch 2013;Nam, Mukherjee, Schilk and Mukherjee 2013;Deshors 2015;Deshors and Gries 2016;Gries and Bernaisch 2016). This article seeks to contribute to the understanding of complementation in World Englishes by looking at the complementation profile of one specific verb, REGRET, across different L2 varieties of English.
REGRET is a retrospective verb, that is, it can be followed by a gerund complement (with or without an expressed subject) which has a retrospective or 'anterior' meaning-it encodes the actual performance of the action expressed in the complement clause-as in example (1), and also by an infinitive complement, which has a prospective or 'posterior' meaning-the meaning is thus also more hypothetical or potential-as in example (2) (Quirk et al. 1985(Quirk et al. : 1193. (1) I regret telling you that John stole it. ["I regret that I told you that John stole it" or "… that I am now telling you…"] (Quirk et al. 1985(Quirk et al. : 1193 (2) I regret to tell you that John stole it.
["I regret that I am about to tell you that John stole it"] (Quirk et al. 1985(Quirk et al. : 1193 In addition, it can also be followed by a finite that/zero declarative complement clause, which can have 'anterior', 'posterior' and 'simultaneous' meanings (Noonan 2007: 110-114), as in examples (3)-(5) respectively. According to Huddleston and Pullum (2002: 160), 'simultaneous' meaning can also be expressed by the gerund, as illustrated in (6).
(3) 'anterior': I regret that I didn't do this earlier.
(GloWbE-United States) (4) 'posterior': We regret that due to a high volume of applications, only those applicants selected for further discussions will be contacted.
(GloWbE-United States) (5) 'simultaneous': We are regretting that we now know what we know.
(GloWbE-United States) (6) I regret being alive still. (GloWbE-United States) Therefore, the scope of variation in the complementation of REGRET for the expression of 'anterior' and 'simultaneous' meanings includes the variants gerund and that/zero complement clause; for the expression of 'posterior' meaning the variants are to-infinitive and that/zero complement clauses.
Other complementation patterns are possible with REGRET, namely a noun phrase (henceforth NP), as in (7), an interrogative complement clause (8), and the it that construction (9); (cf. Huddleston and Pullum 2002: 937-939, 963, and FrameNet 3  As noted above, this study seeks to identify innovative patterns or uses that may take place within the complementation profile of the verb REGRET. Innovation is here understood in Bamgbose's (1998: 2) terms: for him, an innovation is "an acceptable variant" and should be distinguished from an error, the latter being simply "a mistake or uneducated usage." He argues that "the main question that arises with innovations is the need to decide when an observed feature of language use is indeed an innovation and when it is simply an error", and in order to make this decision, defining both innovations and errors is crucial. To make such a distinction, Bamgbose (1998: 3-5) proposes five internal factors that lead to a linguistic feature qualifying as an innovation: a. Demographic: the number of users of a pattern in the acrolectal variety. When a form is not used in the acrolectal variety, it is more "likely to be stigmatized as non-standard" (Bamgbose 1998: 3). b. Geographic: whether the feature is spread across a country and is not limited to a single geographical region or language background (Hindi English, Marathi English). c. Codification: whether the feature is found in grammars, dictionaries, course books, and reference manuals. d. Authoritative: whether the feature is used by professional writers, teachers, media practitioners and publishing houses. e. Acceptability: whether the feature is accepted by the linguistic community. 4 Section 3 below describes the methodology used to identify innovative patterns in the clausal complementation of the verb REGRET. 4 Other categorizations include linguistic innovation vs linguistic conventionalization (Croft 2000) and error vs conventionalized innovation (van Rooy 2011). In both cases, linguistic innovation and conventionalized innovation require the new patterns/forms to be spread, accepted, and selected by the speech community. That is, they need to fulfill at least some of Bamgbose's criteria.

Methodology
As mentioned above, the data for the study was retrieved from GloWbE. 5 This corpus contains data from the years 2012-2013 and is divided into two main text-types: Blogs, which accounts for about 60% of the corpus, and General (40%), which contains other web-based materials, such as newspapers, magazines and company websites. One interesting feature of this corpus is that the language it represents sometimes resembles that of conversations and reflects the turn-taking system, especially in internet blogs, as in examples (10)  The characteristics of the corpus do not allow for an analysis of all 5 factors cited by Bamgbose (1998, cf. Section 2). The first of these, demographic, considers the number of speakers in the acrolectal variety that make use of a specific structure. The GloWbE corpus does not contain metalinguistic data on speakers, and thus it is not possible to know if they use a basilectal, mesolectal or acrolectal variety. However, since writers are using English on the internet, presumably for communicative purposes as an interethnic language, we might assume that they will try to use the most proficient variety of English that they have in their repertoire. In fact, if we have a look at GloWbE, we see that the varieties writers use are not of a mixture of English with their substrate languages but are clear and intelligible for any reader with a 5 It becomes necessary to acknowledge some limitations of the GloWbE corpus. Firstly, the lack of accuracy in determining the geographical background of the writers, since there are no metadata available. Secondly, the comments section of newspapers and blogs are included in the corpus, so that it becomes imposible to ascertain the country of origin of the writers, and thirdly, the difficulty to know the particular variety used by the writer, that is, acrolectal, mesolectal, or basilectal (cf. Davies and Fuchs 2015: 26;Mukherjee 2015: 35;Hoffmann 2018: 179).
command of English. Therefore, the first factor, demographic, will be analyzed in terms of the number of tokens retrieved in each variety of English under study. The second factor, geographic, considers whether the feature is spread across a country. The GloWbE corpus, again, does not contain metadata about the origin of the writers and therefore it is not possible to know whether they come from the same or from different parts of the country. Thus, we will deal with this factor by looking at the spread of a specific feature across the different varieties in the data; if a feature is spread across different countries we must assume that it is spread across the English-speaking population in general. Therefore, I will consider that a feature is geographically spread when it is present in all the four subgroups considered. The other three factors proposed by Bamgbose (1998), authoritative, codification, and acceptance, cannot be directly derived from GloWbE and hence will not be considered here. Future studies will be required to see whether the features found in the 14 L2 varieties in this study increase over time, as well as whether they are codified in grammars and are accepted by the linguistic community. As noted in Section 1, the study examines 16 varieties of English (see also  (12) and (13) respectively), which were excluded from the analysis. Other examples that had to be discarded were instances in which the original source was not the internet, as in example (14), which cites John Milton's "Second Defence of the People of England" (Milton 1653); further examples taken from songs and poems were found and discarded.
(12) I would say this record displays a wide range of themes-family, love, regret, fear, youth, aging, desire, etc. (GloWbE-Great Britain) (13) …whose style of preaching you find painful below that of his regretted predecessor? (GloWbE-United States) (14) a beginning) was glorious; but, with deep emotions of concern will they regret, that those were wanting who might have completed the structure. They will lament (GloWbE-United States) The presence of such examples also led me to manually assess precision and recall, defined as a "measure of the ability of the system to retrieve relevant documents while at the same time holding back nonrelevant ones" (Van Rijsbergen 1979: 11). Precision and recall were analyzed for each variety under study, following Hundt and Schneider (2009), who analyze the suitability of using a parser "as a tool for descriptive linguistics" (Hundt and Schneider 2009: 1) and who argue that parsing performance may vary in different varieties of English due to unexpected differences in the various L2 patterns with respect to L1 ones. As shown in Table 2 below, both precision and recall rates are above 90% in all varieties with the sole exception of American English, where recall is 89.2%. This shows that the search regret*_v* is above average (see for example Hundt, Schneider and Seoane 2016). The following section presents the results obtained after the analysis of the complementation patterns of the verb REGRET in each variety.

Results
This section reports on the manual analysis of the 12,917 valid attestations of the verb REGRET retrieved from the corresponding components of GloWbE. It also offers an in-depth analysis of variation found in the complementation patterns (that/zero clauses vs. to-infinitive and gerund clauses) with special attention to new uses of the canonical patterns and new emerging patterns. Table 3 below shows the number and percentage of tokens of the verb REGRET classified according to the complementation they take (rows 1 to 8). They are further classified into L1 vs L2 varieties, with the latter being grouped geographically (Asia, Africa, and the Caribbean). Rows 1 to 6 include the patterns that are covered in reference grammars, that is NP, gerund with and without an explicit subject, toinfinitive, that/zero CCs, wh-CCs, and the it that construction (see examples (1) to (9) in Section 2). From this group, it is interesting to note that the it that 6 construction, which is recorded in the grammars, is only registered once, in Philippine English (marked in bold in Table 3; cf. example (15)). According to Huddleston and Pullum (2002: 963), REGRET and RESENT are the main two verbs that display this "behaviour", with both the it and the that-complement clause as direct object of the verb, and with the possibility of omitting the it direct object without any change in meaning. Clearly, this construction is not at all frequent in current internet language, which might imply that it is becoming obsolete or simply that it is more frequent in other oral or written text-types.
(15) I know and I regret it that most website owners love to have kinds of flashy stuff. (GloWbE-Philippines) The next group, rows 7 and 8, covers complementation patterns not previously reported in the literature of the verb REGRET and thus potentially innovative. Row 7 deals with REGRET as an intransitive verb, which appears in the corpus 927 times, as in example (16). This use is not acknowledged in any of the reference grammars checked (Quirk et al. 1985;Biber et al. 1999;Huddleston and Pullum 2002) or in FrameNet, though it is mentioned in one dictionary (Merriam-Webster Dictionary), although without being exemplified. However, there are two main reasons why this pattern might not be considered a new or innovative pattern. Firstly, according to the OED, when REGRET entered the English language from the French verb regretter, it could be used both transitively and intransitively (see examples (17) and (18) respectively). The intransitive use already had the meaning 'to feel or express sorrow or regret'. Therefore, the current intransitive construction is not new to REGRET from a diachronic 6 The exceptional nature of this example led me to explore the possibility that it might be a dummy it (cf. Mondorf 2016), but an examination of the broader linguistic context of this example suggested that it should be interpretated as an instantiation of an it that construction.
perspective. The second reason is that even though the examples are syntactically intransitive, there is usually a retrievable direct object in the context. This can be seen in example (19), where the previous clause We had hurt someone is understood as the direct object of the verb REGRET (We regretted at that very moment that we had hurt someone). In the historical examples, (17) and (18) Finally, row 8 in Table 3, with the label "other CC", includes all the new complementation patterns that are not considered in the grammars and are not historically attested. These will be analyzed in more depth in subsection 4.2. Before that, I will provide an overview of the distribution of the available canonical complementation patterns and a brief account of their new uses. The main variability in the complementation of the verb REGRET is that between (i) gerunds and finite complement clauses where the action expressed in the CC is 'anterior' or 'simultaneous' to the time indicated by the main verb REGRET, and between (ii) to-infinitives and finite complement clauses where the action expressed in the CC is 'posterior' to the time indicated by REGRET. As for (ii), the corpus data here confirms that both to-infinitives and finite complement clauses can be used with 'posterior' meanings (see examples (20) and (21) respectively). However, in most cases, alternation between them is not possible: what is expressed with a finite complement clause cannot be expressed with an infinitive one, and vice versa (see example (22)).
(20) I regret to say that accumulated problems in the judiciary do not yield to any immediate and ready-made solution. (GloWbE-Bangladesh) (21) …they regret they will not be able to attend. (GloWbE-Great Britain) (22) If we have not received your completed and signed booking form with the appropriate deposit within 7 days, we regret that your reservation will lapse. (GloWbE-Great Britain) Therefore, for the study of the distribution of the available canonical patterns, I will focus on the alternation between gerunds and finite clauses, in which the meaning expressed by the CC is 'anterior' or 'simultaneous' to that expressed in the main clause. Table 4 sets out the distribution of these two patterns across the different varieties, grouped as L1 and L2 varieties. It shows a preference for the gerund in all the varieties; however, this preference is lower in L2 varieties and there is a stronger tendency for the use of finite patterns. The Fisher exact test shows that the difference between the four groups with regard to the choice between the gerund and finite patterns is significant at p<.05; p-value is 9.215075 x 10 -7 . These results seem to confirm previous research that revealed a stronger tendency towards a more transparent mapping of form and meaning in L2 varieties of English (cf. Nesselhauf 2009;Steger and Schneider 2012;Laporte 2012;Gilquin 2015;Callies 2016; among others). More precisely Romasanta (2017) studies the complementation of the verb REGRET in British English, American English, Hong Kong English, Nigerian English, and Jamaican English, and concludes that the stronger tendency for the use of finite patterns in L2 varieties may be explained by both substrate influence and cognitive processes of transparency and isomorphism. 7 On the other hand, Romasanta (forthcoming) concludes that other syntactic and semantic factors may also play a role in the choice of more transparent patterns, such as animacy of the subject in the main clause, coreferential subjects, voice of the complement clause, length of the complement clause, presence of negative markers in the complement clause, presence of intervening material between the main clause and the complement clause, among others.

New Uses of the Canonical Complementation Patterns
The verb REGRET exhibits new uses of the canonical patterns described above. The innovation has to do with the meaning of the CC, in particular with the temporal relation between the time expressed in the main clause and the time expressed in the complement clause. As seen in section 2, complement clauses can express three different meanings, that is, 'anterior', 'simultaneous' or 'posterior' meanings. That complement clauses can express the three meanings, gerunds can express both 'anterior' and 'simultaneous' meanings, and infinitives have a 'posterior' meaning.
Starting with the 'anterior' meaning, the canonical patterns are that complement clauses, as in example (23), and gerunds, as in example (24). However, as can be seen in Table 5 below, examples with 'anterior' meaning expressed with to-infinitive CCs have also been attested in the corpus. In many examples like these, which express 'anterior' meaning with a to-infinitive clause, the verb of the complement clause is usually a verb of the senses, such as SEE and HEAR, as in example (25); other verbs such as LEARN, MEET, and LET are also recorded (see examples in (26)  The Fisher exact test shows that the difference between the four groups with regard to the expression of an 'anterior' meaning is significant at p<.05; p-value is 0.00005322458. As mentioned in Section 3, for the study of innovation I follow Bamgbose's (1998), in particular his demographic and geographic criteria, since the other three criteria (authoritative, codification, and acceptability) cannot be examined in GloWbE. From a demographic and geographic perspective, Table 5 shows that examples of this new use of the to-infinitive with an 'anterior' meaning are present both in L1 and L2 varieties, with the Caribbean and Asia displaying the highest percentages (3.5% and 3.4% respectively). Therefore, this pattern seems to be both demographically and geographically spread. The percentages, however, are rather low and future studies are necessary to confirm this apparently innovative use of an already existing pattern.
The next meaning, 'simultaneous' meaning, is usually expressed either with that complement clauses, as in example (27), or gerunds, as in example (28). However, as seen in Table 6 Bamgbose's (1998: 2) demographic factor, therefore, this pattern is not demographically spread. As for the geographic factor, it seems that this pattern is tentatively favored in African varieties (Nigerian, Ghanaian, and Kenyan English) where three of the four examples are found. In general, however, this innovative use of an already existing pattern would not qualify as an innovation, in Bamgbose's (1998) terms, in internet language.
Finally, the 'posterior' meaning is usually expressed either with that complement clauses, as in example (30), or to-infinitives, as in example (31). Table 7, however, shows that gerunds also express meaning: two are attested in the L1 varieties of English (British English and American English) and one in Philippine English (see examples in (32)). Example (32c.) may be understood as expressing 'anterior' meaning; however, if we look at the context of the sentence, the writer is talking about an invitation she received for a future event, and therefore she regrets not being there in the future. The Fisher exact test shows that the difference between the four groups with regard to the expression of a 'posterior' meaning is significant at p<.05; p-value is 0.002894456. The number of examples is very low and therefore it is not demographically and geographically spread.
To summarize this section, we can say that the canonical complementation patterns that are being used with new meanings normally display very low numbers and are not spread across the population or across countries (cf . Tables 6 and 7). Such examples could therefore be considered production errors. However, the pattern with higher percentages of occurrence, that is, the use of to-infinitives with an 'anterior' meaning (Table 5), might be worth exploring in further research using new datasets that allow for an analysis of future developments in this complementation pattern.
In the next subsection, I will focus on the label "other CC patterns", used in Table 3: complementation patterns that are not treated in reference grammars.

Emergence of New Complementation Patterns
The final form of innovation in the complementation profile of REGRET can be found in the emergence of new complementation patterns that are not contemplated in grammars or dictionaries. These other forms of complementation are included in Table 8.
3,780 5,669 3,004 462 12,915 The first pattern shown in Table 8 is the use of the verb REGRET followed by a prepositional phase (henceforth PP; see example (33)). Table 9 provides the number of examples retrieved with this pattern and the percentage it represents within the total number of instances analyzed for each regional group. It shows that this pattern is most productive in Asian and African varieties with 3.3% and 2% respectively, whereas they only represent 0.4% and 0.9% in the L1s and the Caribbean.
(33) But you will NEVER regret for visiting there! (GloWbE-Singapore) Looking at Asia and Africa in more detail, Figure 2 shows that within this continent, Malaysian English and Bangladesh English are the varieties that make most frequent use of this pattern (5.4% and 4.7% each), and within Africa, it is Ghanaian English and Kenyan English that resort to this pattern most often (with 3.4% and 2.3% respectively). Turning now to the specific prepositions used, in Asia we find examples with all the prepositions attested in the data (11 different prepositions in total). Some of these, such as at, from, towards, upon, and with, which only occur followed by a NP, are even exclusive to Asia (see examples (34) -(38) respectively). On the other hand, in African varieties the range of prepositions is more restricted. As for the native varieties and Jamaican English, they only make use of 5 and 3 different prepositions respectively.  -4  2  -6  NP  -1  --1  TOTAL  -5  2  -7   from   -ing  -1  --1  NP  -1  --1  TOTAL  -2  --2  towards NP  -2  --2  upon  NP  1  1  --2  with  NP  -2  --2  TOTAL  16  189  60  4  269 As can be seen in Table 10, the preferred preposition in Asia, Africa, and the native varieties is for, and it is followed by both gerunds and NPs, with 76 and 64 examples each, as in examples (39) and (40) respectively. The next most frequent preposition is about, which is again used in all the varieties and is the preferred preposition in the Caribbean variety, but the total number of examples here is just one third of those in the case of for. As to about, the preference is for NP (see example (41) Table 11 shows the use of the verb REGRET followed by a preposition by grouping the examples in terms of the complement of the preposition instead of the preposition itself. The use of a preposition followed either by a gerund or an NP occurs in all the varieties, and both patterns are productive, with 100 and 172 examples in total, respectively. One possible explanation for this use of the verb followed by a preposition is through analogy with a number of patterns. On the one hand, it may be an analogy with prepositional gerunds like she delighted in doing it, since similar examples with the verb REGRET are found in my data, such as but I don't regret the least in paying nearly double the 2009 pricing (GloWbE-Malaysia). However, analogy with prepositional gerunds would only explain the use of REGRET followed by a preposition + gerund, and not the use of preposition + NP. Another possible source of analogy is the noun REGRET, which can be followed by different prepositions + gerund, especially about, at, for, as in It reflects a moral regret for having violated a situation of trust, and with these and other prepositions + NP, namely about, at, for, in, of, over, as in Audrey has no regret about her decision. Most of the prepositions found in my data can be used with the noun REGRET, but there are some new ones: on, from, towards, upon, and with. Finally, other possible sources of analogy are the semantically related construction be sorry for/about something, which has the meaning of 'regret', and other prepositional verbs such as long for, first attested in c1225, cf. OED s.v. long, v.1. These data may point towards the emergence of a new prepositional verb, especially with the preposition for (to regret for something), a phenomenon investigated in previous studies (cf. for example Schneider 2004;Mukherjee and Hoffmann 2006;Mukherjee 2009;Nesselhauf 2009;Zipp 2010;Schneider and Zipp 2013;Tan 2013Tan , 2016Hoffmann 2018).
This PP complementation pattern of REGRET is both demographically and geographically widespread, and therefore it might be considered as an innovation that could be gaining ground in present-day English, both in L2 and (especially) in L1 varieties.
Returning to Table 8, from rows 2 to 8 we find, firstly, S + toinfinitive, which only appears once, in Bangladesh English (see example (42)) and, secondly, bare infinitives, with one example in each native variety of English (see examples in (43)). Note that in example (43a) there is intervening material between the verb REGRET and the verb of the complement clause, which may be the reason for the use of the bare infinitive instead of the expected to-infinitive. Thirdly, there is also one example of the construction it + to-infinitive, illustrated in (44). This example is from Sri Lankan English and resembles the construction it that found in Huddleston and Pullum (2002: 963). There is also one instance with the complementation pattern that + gerund, and another one with that + to-infinitive (see examples (45) and (46)), both occurring in Sri Lankan English. In example (46), that may be understood as a complementizer or as a fronted object of the verb buy. It is also interesting to note that in both examples the verb TO BE appears before REGRET (You are regret, I am regret). This to be regret construction is also found in other examples, such as those in (47). These examples occur in varieties from both Asia and Africa, but not in the variety from the Caribbean (Jamaica) or in the L1 varieties. The next pattern is the use of the past participle, as in examples (48a) to (48d), which occurs mostly (5 of 7 examples) with the verbs SEE and BE (seen and been respectively). Hundt (2016) investigates the use of seen and been for seeing and being in "unusual auxiliary-participle combination" such as *are been seen and *have been seen, and looks at metalinguistic comments, native speaker judgments, corpus data, historical evidence, evidence from dialects, and child language. Her conclusion is that seen and been may be simply a spelling variant of seeing and being, since the two pairs can be homophones in spoken English. Thus, it seems safe to apply this hypothesis of seen and been being spelling variants to the case of verb complementation. However, two of the examples in my data, one with bought and another one with missed, do not seem to support this, since bought and buying and missed and missing are not homophones. Another hypothesis, as a reviewer pointed out, could be that this is a case of syntactic ellipsis, where the perfect auxiliary having is elided (In fact I regret having seen them…). Future research might clarify whether this is an analogical extension of REGRET + seen and REGRET + been to other verbs. Finally, example (49) illustrates the one case of the complementation pattern for NP -ing, which occurs in Sri Lankan English. Given the low frequency of these new complementation patterns, and that they are not spread across different varieties of English (they occur mainly in four Asian and three African varieties of English), these should probably be considered to be performance errors.

Conclusion
This paper has dealt with innovation in the complementation system of REGRET across different L1 and L2 varieties of English by looking at new uses of the canonical complementation patterns and new patterns found with this verb.
To begin with the general data, some L2 varieties of English exhibit a higher proportion of finite patterns than the L1 varieties. Two explanations for this difference that will have to be tested in future research are the potential influence of the substrate languages, and the effect of cognitive processes such as the tendency towards isomorphism and transparency that take place in situations of language contact and second language acquisition.
As for the innovations, two of the criteria proposed by Bamgbose (1998) were tested for all the new patterns and uses found: demographic and geographic distribution. The only pattern that fulfills these two criteria is the use of REGRET followed by prepositions, especially the preposition for as in I regretted for coming here (GloWbE-Jamaica). Different hypotheses were proposed for this new use of REGRET: i) analogy with prepositional gerunds (as in She delighted in doing it), ii) analogy with the noun REGRET, which can be followed by different prepositions, and iii) analogy with be sorry for/about, which is semantically equivalent to REGRET. Other potential innovations, which do not fulfill Bamgbose's factors (1998) but will have to be explored in the future, are the use of toinfinitives with retrospective meanings, as in I am tired and regret to be a Kenyan (GloWbE-Kenya), and gerunds with prospective meanings, as in I regret reminding you (GloWbE-United States). These new uses are not yet spread demographically and geographically, but future studies might detect a gradual increase in use.
In sum, through the analysis of the complementation profile of REGRET, this paper has found further evidence in support of the fact that, as Schneider (2007: 86) and Olavarría de Ersson and Shaw (2003: 138) among others have noted, the intersection between lexis and grammar in general terms, and complementation in particular, is frequently the locus of variation and change, especially during the process of structural nativization in L2 varieties of English. However, more research is still needed. Further corpus data, for example, containing metalinguistic data on speakers would help to identify more clearly the demographic and geographic factors involved in potential innovations. The other three parameters proposed by Bamgbose (1998), authoritative, codification and acceptability, also merit future research.