A Formalization of the Proposed Distribution

The long-standing claim that this study attempts to challenge is that colloquial Indian English syntax does not allow inversion in main clause questions but has obligatory inversion in embedded questions, as illustrated in examples (1) and (2) (Trudgill & Hannah, Reference Trudgill and Hannah2008, p. 137, Examples 9a, 9b):

1. (1) a. What this is made from?

2. b. Who you have come to see?

1. (2) a. I asked him where does he work.

2. b. I wonder where is he.

Such forms have been noted since the early days of scholarship on Indian English (Whitworth, 1907/2007, pp. 148–150) and are regularly supplied in descriptive grammatical overviews of the language. Trudgill and Hannah (Reference Trudgill and Hannah2008) characterize this aspect of syntax as “exactly the opposite of EngEng [U.K. English] usage” (p. 137), and Mesthrie and Bhatt (Reference Mesthrie and Bhatt2008) state that “the generalization of question-formation strategy in colloquial Indian English is the mirror opposite of the generalization of question-formation strategy in Std Ind [Standard Indian] English” (p. 98). That such forms exist is without question. As shall be shown, they exist in several varieties of English. The issue is whether the alternative formulations are ruled out, such that, as Bhatt (2004/2008) states, “the simple empirical generalization that emerges from the data . . . is that in Vernacular IndEng inversion is restricted to embedded questions [emphasis added]; it does not apply in matrix questions” (p. 551).

Bhatt (Reference Bhatt2000) and Mesthrie and Bhatt (2008, pp. 96–106) go beyond anecdotal observation in providing a formal account of this variation, contrasting Standard Indian English with Colloquial Indian English.^{Footnote 5} They elaborate an analysis in terms of optimality theory (OT; Grimshaw, Reference Grimshaw1997; Prince & Smolensky, Reference Prince and Smolensky1993) to distinguish two distinct formal grammars: one with inversion restricted to and obligatory in main clauses and the other with inversion restricted to and obligatory in embedded clauses. As a complete evaluation of this account is beyond the scope of this article, the following description conveys the basics of the analysis.

The OT framework conceives of grammars as hierarchies of constraints on well-formedness. Particular language grammars differ from one another in how they rank the constraints. These constraints are not inviolable; rather, a constraint may be violated in the case that the structure is preferred according to the values of a higher ranking constraint. In any language, if two syntactic variations are considered, evaluation of the constraints violated in terms of their ranking determines which option is considered “optimal.” In this case, the relevant constraints are as follows:

1. 1. OP-SPEC: Operators must be in specifier (Spec) position.

2. 2. STAY: Constituents must remain in position; no movement is allowed.

3. 3. OB-HD: Heads of selected projections must be filled, either overtly or by a trace (t).

In Bhatt’s (Reference Bhatt2000) and Mesthrie and Bhatt’s (Reference Mesthrie and Bhatt2008) analysis of main clauses, the first two constraints are relevant. The ranking for Standard English(es) is OP-SPEC » STAY, so a sentence such as What would you like to eat? is preferred over What you would like to eat? In the sentence What would you like to eat? OP-SPEC is respected, due to wh-movement to the specifier of the complementizer phrase (Spec, CP), even though STAY is violated twice: once by movement from T (tense) to C (complementizer) and again by movement from object position to Spec, CP. However, in What you would like to eat? which is assumed to be generated by adjunction of the wh-phrase to the tense phrase (TP), OP-SPEC is violated, ruling out the candidate structure. In contrast, in Indian English, the proposed ranking is STAY » OP-SPEC, such that the preferred candidate is What you would like to eat? In this case, although both STAY and OP-SPEC are each violated once, the other candidate is worse: two cases of movement (the modal and the wh-phrase) would amount to two violations of the higher ranking constraint.

In the proposed analysis of embedded clauses, only the last two constraints are relevant, as the wh-word is assumed to have moved to Spec, CP in both language varieties. The ranking for Standard English(es) is STAY » OB-HD, so a sentence such as I wonder what he is eating, with only one instance of movement, is preferred over I wonder what is he eating, which involves two violations of STAY. That OB-HD is violated by I wonder what he is eating, as the head of the CP remains unfilled, does not matter, because STAY is the higher ranking constraint. In Indian English, however, the ranking is argued to be OB-HD » STAY, which means that I wonder what is he eating emerges as the stronger candidate. In this case, OB-HD is not violated, as T-to-C movement results in the head of the CP being overtly filled.

For a more complete explanation of this account including OT tableaux, see Mesthrie and Bhatt (2008, pp. 96–106). For current purposes, it is sufficient to note that the logic of argumentation is sound given the assumptions of the analysis, both overt (the OT constraints and rankings and the adjunction analysis of wh-fronting) and covert (that variation is due to two separate grammars and that there is no optionality involved). Moreover, the general spirit of inquiry, to demonstrate that “Indian vernacular English is just as systematic and logical as any other variety of English” (Bhatt, Reference Bhatt2000, p. 69), is one that I endorse unconditionally.

However, two problems arise that suggest that an alternative analysis is needed. First, despite commonly repeated claims in textbooks and anecdotal reports, there is a lack of empirical support for this distribution. Second, a language with wh-inversion entirely restricted to embedded clauses would stand as an anomaly, given previous research on universal patterns of embedding and inversion.

The Status of Existing Evidence

To date, no systematic empirical study in the World Englishes paradigm has shown this distribution to exist, despite various anecdotal claims and informal investigations. Bhatt’s (Reference Bhatt2000) evaluation of inversion in embedded clauses was based on presenting only six tokens of this construction, as part of a series of 24 sentences all exemplifying Indian English variants, to 27 people (including high school English teachers, two doctors, two linguists, and an engineer) and asking them if the sentences were likely to be produced by a speaker of Indian English or British English.^{Footnote 6} The results for inversion in embedded clauses were as follows: 81% were judged to be Indian English, 0% British English, and 19% neither (i.e., participants could not imagine any inversion in embedded clauses). However, interpreting these results is less than straightforward, for a number of reasons: (a) the numbers of participants and stimuli were very small; (b) the test instrument was not balanced; (c) participants gave not grammaticality judgments but rather perceptions of what sounded “more Indian” and “less British;” and (d) the fact that these forms are attested in Indian English in no way implies that noninversion in embedded sentences is disallowed.

The interpretation of obligatory inversion in embedded clauses also flies in the face of recent corpus studies that suggest that noninversion in embedded clauses is the preferred (though not the only) option in Indian English. Although these investigations may be considered problematic in various respects, they do all point to the same conclusion. In a pioneering corpus linguistic investigation of this issue, Hilbert (Reference Hilbert, Siemund and Kitana2008) performed a custom search using the International Corpus of English, henceforth ICE (for discussion of this set of freely available online corpora, see Greenbaum, Reference Greenbaum1996). She searched the spoken files of the Indian subcorpus (ICE-India), which contain 600,000 words, for patterns of inversion in main and embedded clauses. The rates of wh-inversion reported for main clauses were 81% inversion (540 tokens) and 11% noninversion (76 tokens); the proportions for embedded clauses were 18% inversion and 82% noninversion. However, the precise calculation of the figures remains unclear. There were no stated criteria for exclusion, thus no account was given of the apparent exclusions in main clauses (11 + 81 < 100%), and no raw numbers were given for embedded clauses. The importance of this will become apparent in the Results section; in the current investigation, as many as 35% of strings of wh-elements followed by auxiliaries turned out not to be wh-questions. Possible hits that are candidates for exclusion include wh-clefts, relative clauses, and cases that are ambiguous with respect to movement, such as subject wh-questions and copied auxiliaries or modals (e.g., How should we should count these?). Another issue in Hilbert (Reference Hilbert, Siemund and Kitana2008) was that yes/no questions were identified in the untagged corpus by searching for transcribed question marks. However, not all yes/no questions in the corpus are followed by a question mark (many are inside longer stretches of dialogue). The only information concerning the search for wh-questions is that they “were selected independently of this mark-up” (Hilbert, Reference Hilbert, Siemund and Kitana2008, p. 270).

In a subsequent study, drawing on a subcorpus of the spoken files of ICE-India—namely, the “private conversations” (S1A-001 to S1A-100)—Hilbert (2011) focused on how inversion may be conditioned by the verb type: In main clauses, inversion is preferred, but more so with be (about 95%) than with modals or auxiliaries (about 80%) or with lexical verbs involving do-support (about 60%). Neither raw numbers nor percentages were provided; the preceding figures are estimations based on a bar graph (Hilbert, Reference Hilbert, Mukherjee and Hundt2011, p. 130, Figure 2). Hilbert argued that, in embedded wh-questions, inversion is almost entirely restricted to third-person singular is, leading her to propose an analysis in terms of fixed chunks derived from a process of learning by imitation (to be examined in more detail in the Discussion section). However, these findings are difficult to interpret, as there are no raw numbers, no accurate percentages, no statistics, and, again, no information concerning the search criteria.

Balasubramanian (Reference Balasubramanian2009) also found that the proportion of noninversion with wh-phrases in main clauses in the spoken files of ICE-India was very low, at only 8% (72/880), although a more detailed breakdown is lacking, and the analysis did not extend to embedded clauses. In this case, the search followed the tagging of the corpus. However, as in Hilbert (Reference Hilbert, Siemund and Kitana2008), there was no reference to any criteria for filtering.

Sedlatschek (Reference Sedlatschek2009) reported findings from a smaller corpus of 180,000 words culled from newspapers, broadcasts, and student essays. In this case, the author’s methodology was more explicit (p. 292). First, all wh-clauses in the corpus were identified using the WordSmith concordance program (Scott, Reference Scott2004), with nine different wh-elements as keywords. The data were then filtered manually, eliminating relative clauses, subject wh-questions, and instances of usage that conform to Standard English conventions (such as noninversion following how come). The remaining data were categorized into main and embedded clauses. Proportions of embedded wh-inversion in embedded clauses from the three subsections of the corpus were as follows: 10% from the press (2/20), 13.3% from broadcasts (10/75), and 2.5% from essays (1/40) (p. 297). Despite the limitations of scale and a reliance on written rather than oral production, the analysis was careful, the distribution was clear, and the findings were in accordance with the other corpus studies cited.

Investigations such as Balasubramanian (Reference Balasubramanian2009), Hilbert (Reference Hilbert, Siemund and Kitana2008), and Sedlatschek (Reference Sedlatschek2009) stand as important pioneering studies and deal with more issues than those addressed within the narrow focus of the current investigation. They highlight an exciting new avenue in corpus-based research, which promises not only to shed greater light on the nature of the so-called New Englishes but also to facilitate comparison with classroom varieties of English (see Mukherjee & Hundt, Reference Mukherjee and Hundt2011). The preceding review makes clear that, in addition to a general lack of statistical analysis, each of these initial findings requires further investigation due to partial reporting of the results, because the search methodology was unclear, or because there were no clear criteria for inclusion or exclusion. However, they all trend in the same direction. On the issue of wh-inversion in Indian English, the current study hopes to provide more comprehensive empirical evidence and to elucidate the relevance of the results for L2 research on language universals.

Syntactic and Discourse-Pragmatic Constraints

Whether Indian English exhibits wh-inversion in embedded clauses is not only an empirical question about a particular language variety; it also bears on broader issues of syntactic universals and discourse-pragmatic constraints. The hypothesized syntactic distribution is implausible whether viewed from a formalist perspective (given the history of research on main clause phenomena) or a functionalist perspective (considering the backgrounding of information in embedded clauses).

Several decades of formal research on main clause phenomena lead us to expect that no natural language should allow inversion in embedded clauses while ruling it out in main clauses. The first systematic investigation of the syntax of main and embedded clauses was that of Emonds (Reference Emonds1970, Reference Emonds1976). Emonds (Reference Emonds1970) proposed the structure-preserving constraint, an important feature of which was the restriction of certain syntactic phenomena, termed root transformations, to main clauses. On this approach, wh-inversion is a root transformation, and any apparent exceptions must be otherwise explained. Emonds (1976, pp. 23–25) considered apparent counterexamples such as wh-inversion in embedded clauses in African-American English, as shown in (3) (Labov, Cohen, Robbens, & Lewis, Reference Labov, Cohen, Robbens and Lewis1968, p. 299, Examples 415, 419).

1. (3) a. I don’t know how did I do it.

2. b. Where did she get the coat from I don’t know.

He suggested that such forms may be similar in structure to the “mixed indirect discourse” (p. 25) of German, in which the apparent embedded clause exhibits verb-second (V2) syntax, which is ordinarily a main clause phenomenon (in today’s terminology, the finite verb raises to C, and any single argument or adjunct phrase may raise to Spec, CP). On this account, all apparent cases of wh-inversion in embedded clauses are actually derived parenthetically from the fusion of two main clauses, as in the following example (Emonds, Reference Emonds1976, p. 25, Examples 11, 12).

1. (4) a. John wondered (mused), (why) should he be early.

2. b. John wondered thus: Why should he be early.

Subsequent to Emonds’s (Reference Emonds1970) proposal, Ross (Reference Ross, Corum, Smith-Stark and Weiser1973) developed an alternative statement of constraints on embedding. Referring to main clauses as “upstairs” and embedded clauses as “downstairs,” a jargon still familiar in generative linguistics today, he proposed “a principle whose truth is borne out in myriad cases of Real Apartment Life, and which is therefore dubbed: (1) The Penthouse Principle: More goes on upstairs than downstairs” (p. 397). He proceeded to recast it more soberly as follows:

1. (5) No syntactic process can apply only in subordinate clauses.

Ross (Reference Ross, Corum, Smith-Stark and Weiser1973) argued that apparent exceptions such as NP-raising (noun phrase raising) and that-deletion crucially involve activation of another, higher clause, and syntactic processes such as reference questions, fronting rules, embedded force rules, pseudoimperatives, and various idioms either do not apply in embedded clauses or “are more active, have a wider scope of application . . . when they operate upstairs” (p. 401). Thus, although (5) remains valid, the final statement of the penthouse principle is, in implicational terms: “where ≥ means ‘is as syntactically active as, or more syntactically active than,’ the following inequality obtains: Main clauses ≥ subordinate clauses” (p. 411). The penthouse principle makes different predictions from the structure-preserving constraint. Whereas the structure-preserving constraint implies that wh-inversion should never be found in embedded clauses, the penthouse principle allows for this phenomenon on the condition that it applies freely in main clauses and that its use in embedded clauses is relatively restricted.

In an early functionalist response to these syntactic accounts of the asymmetry between main and embedded clauses, Hooper and Thompson (Reference Hooper and Thompson1973) argued that discourse factors were more useful in explaining such differences. Embedded clauses generally contain backgrounded information, which is less likely to undergo the kind of transformations associated with topicalization, contrast, and presentative focus. However, on this account, such transformations remain possible when the embedded information involves the assertion of a proposition, as shown in examples (6)–(8) (Hooper & Thompson, Reference Hooper and Thompson1973, p. 466, Example 5; p. 478, Example 92; p. 480, Example 123).

1. (6) Alice vowed that under no circumstances would she loan me the key.

1. (7) It appears that this book he read thoroughly.

1. (8) The scout discovered that beyond the next hill stood a large fortress.

As for questions, actual requests for information are fronted in discourse and appear in main clauses, but there are circumstances in which, depending on the type of predicate and the discourse context, the embedded interrogative is also treated as a “real question.” G. M. Green (Reference Green1996) argued that pragmatics sheds light on precisely which contexts are felicitous for embedded inversion in colloquial American English. She observed that embedded questions such as (9) and the truncation of embedded questions (“sluicing”) as in (10) both imply that “the individual to whom the answer is implied or assumed to be relevant is in fact ignorant of the answer” (p. 141). Thus the (a) sentences are better than the (b) sentences in (9) and (10) (G. M. Green, Reference Green1996, p. 141, Examples 23a, 23b, 24a, 24b).

1. (9) a. She wants to know who did I appoint.

2. b. ?She already knows who did I appoint.

1. (10) a. John broke something but he won’t say what.

2. b. ?John broke something and he said what.

In a more recent discussion of main and embedded clauses, Bybee (Reference Bybee, Bybee and Noonan2002)^{Footnote 7} explained the thrust of this discourse-pragmatic approach to the difference between clauses types as follows: “Typical subordinate clauses are pragmatically flat just as they tend to be intonationally flat and less susceptible to permutations for pragmatic purposes” (p. 5). Unlike main clauses, which provide the default location for topicalization, contrast, focus, and questions, embedded clauses have functions that are generally “much more modest” (p. 4): modifying or further identifying the head NP in the case of relative clauses or modifying the main event in terms of causes, times, or locations in the case of adverbial clauses.

The implications of this approach for syntactic distribution seem to align with the penthouse principle, which makes predictions in terms of proportional distribution, rather than the categorical restriction implied by the structure-preserving constraint. Both Hooper and Thompson (Reference Hooper and Thompson1973) and Bybee (Reference Bybee, Bybee and Noonan2002) argue that an approach based on discourse pragmatics is more explanatory than essentially descriptive syntactic accounts pairing transformations with grammatical domains, which are too constrained and cannot convincingly account for the counterexamples. In contrast, proponents of a syntactic approach tend to see purely pragmatic accounts that rely on “vague discourse properties” (Emonds, Reference Emonds, Haegeman, Nye and Aelbrecht2012, p. 28), such as assertion and presentational focus, as failing to account for syntactic constraints in embedded clauses.

A possible bridge between formal syntactic and discourse considerations has emerged in work by Emonds (Reference Emonds, Adger, De Cat and Tsoulas2004, Reference Emonds, Haegeman, Nye and Aelbrecht2012), who argues that the previous category of sentential root may be reconstrued as a discourse projection, defined as an unselected finite inflection phrase (IP)—that is, not an argument or adjunct of an underlying lexical item. The term discourse projection captures the fact that “the constituents of a root clause stand in privileged relations with surrounding categories of discourse” (Emonds, Reference Emonds, Adger, De Cat and Tsoulas2004, p. 81) such as focus, topic, and speaker-oriented interjections. Various types of movement to the left edge of the main clause are analyzed in terms of category-less phrases (XPs) characterized as discourse shells, which dominate discourse projections and whose Spec positions function as unrestricted landing sites for root phenomena. The concept of discourse shells is intended to generalize over and replace previous proposals of topic phrases and focus phrases (Rizzi, Reference Rizzi and Haegeman1997, Reference Rizzi, Cheng and Corver2006).^{Footnote 8}

Regarding seemingly exceptional transformations in embedded clauses, Emonds (Reference Emonds, Adger, De Cat and Tsoulas2004, Reference Emonds, Haegeman, Nye and Aelbrecht2012) argues that many are, in fact, more restricted than generally realized. Cases of what he terms RIDE (root-like indirect discourse embedding) are always finite; they substitute for complements, not adjuncts; they are usually introduced by verbal or adjectival heads, not nominal or prepositional heads; and one argument of the introductory predicate must be animate. The distinction between root and nonroot phenomena is maintained but is loosened somewhat, in that individual languages may allow certain types of embedded clauses as discourse projections as long as RIDE conditions are met and with a degree of parametric variation as to the types of clauses involved. Advantages of this approach over a pure discourse-pragmatic account include a formal characterization of constraints on RIDE and a coherent structural analysis of the phenomena.

There is another possible explanation of certain main versus embedded clause discrepancies, which, although very simple, appears to account for a significant amount of the variation found with wh-inversion. Both Emonds (Reference Emonds, Adger, De Cat and Tsoulas2004, Reference Emonds, Haegeman, Nye and Aelbrecht2012) and the cartographic approaches based on Rizzi (Reference Rizzi and Haegeman1997, Reference Rizzi, Cheng and Corver2006) eschew a simple, unified CP analysis of question formation in main and embedded clauses. Yet there appears to be a straightforward link between the existence of overt complementizers and cases of prohibition on movement in embedded clauses—as shall be shown later in this article—which is easily explained if the structure used for complement clauses and that used as a landing site in question formation is one and the same functional projection. As Lightfoot (Reference Lightfoot, Haegeman, Nye and Aelbrecht2012) puts it, “the reason [for restrictions on syntactic operations in embedded clauses] is that the syntactic enabling conditions are not met” (p. 172). The results of the current study support this conclusion.

To summarize: Across a broad range of linguistic frameworks, whether one adopts an account in terms of general syntactic principles or in terms of discourse-pragmatic constraints, there is consensus in the literature as to the well-documented asymmetry between main and embedded clauses, and as such the traditional claim about inversion in Indian English stands out as an anomaly.

Embedded Inversion in L1 Dialects

Evidence from L1 dialects and colloquial forms of standard varieties can also shed light on the nature of embedded inversion. In various L1 dialects, embedded inversion is relatively commonplace, but only when specific syntactic and pragmatic conditions obtain, and always in conjunction with (at least optional) inversion in main clauses. Belfast English has true embedded inversion, as argued by Henry (1995, pp. 105–123). Example (11) shows the sequence of tense and change of pronouns, and (12) shows inversion in interrogative complements of verbs that cannot select direct speech (Henry, Reference Henry1995, pp. 106–107, Examples 11a, 11b, 19, 20).

1. (11) a. They asked me, “Have you read War and Peace?”

2. b. They asked me had I read War and Peace.

1. (12) a. The police found out had the goods been stolen.

2. b. We couldn’t establish did he meet them.

However, inversion is not a feature of all embedded questions. If the clause is introduced by an interrogative complementizer,^{Footnote 9} inversion is impossible, as shown in (13), and although wh-movement may co-occur with the complementizer that in this variety, such combinations rule out inversion, as shown in (14) (adapted from Henry, Reference Henry1995, pp. 107–108, Examples 25a, 25b, 25c, 27, 33).

1. (13) a. They couldn’t work out {whether/if} we had left.

2. b. They couldn’t work out had we left.

3. c. *They couldn’t work out {whether/if} had we left.

1. (14) a. I wonder which dish that they picked.

2. b. I wonder which dish did they pick.

3. c. *I wonder which dish that did they pick.

These examples all point to inversion by means of a CP at the head of an embedded clause, rather than the fusion of two main clauses. A further possibility in Belfast English that cements the embedded clause analysis involves cases with long-distance movement, as shown in (15), which cannot be analyzed as parentheticals (adapted from Henry, Reference Henry1995, pp. 108–109, Examples 35, 46).

1. (15) a. Who _idid John hope would he see t_i?

2. b. Who _idid John say did Mary claim had John feared would Bill attack t_i?

McCloskey (2006, pp. 89–90) gives the following examples from Irish English (Hiberno English), which he notes are not parentheticals. Note the pronominal binding in (16) and the impossibility of declaratives with a rising intonation pattern in this context in (17a), in contrast with their acceptability in true parentheticals, such as in (17b) (McCloskey, Reference McCloskey, Zanuttini, Campos, Herburger and Portner2006, p. 89, Example 8; p. 90, Examples 13–14).^{Footnote 10}

1. (16) Every male physicist wonders will he be awarded a Nobel Prize.

1. (17) a. *I wonder it’s raining?

2. b. It’s raining?, she mused.

McCloskey (Reference McCloskey, Zanuttini, Campos, Herburger and Portner2006) goes on to observe that, in this variety, embedded inversion is usually found with question predicates such as ask / wonder / inquire / want to know / want to see, but it is characteristic of resolutive predicates such as find out / know / remember / discover only when they are interrogative or negative (pp. 112–115). Moreover, McCloskey notes that embedded inversion is found more generally in the presence of “devices which determine nonveridical contexts” (p. 114). The same principle appears to apply in African-American English, as previously shown in the example in (3) for negation and in the examples in (18) with imperatives and desideratives (L. J. Green, Reference Green2002, pp. 87–88, Examples 20f, 20h, 20e, 20k).

1. (18) a. Tell me do it make any sense.

2. b. Go over there and see did they bring my car in.

3. c. I wanted to see was it the one we bought.

4. d. I wanted to know could they do it for me.

That African-American English also allows noninversion in main clauses does not run counter to the penthouse principle, as such inversion is optional. No speaker disallows inversion in main clauses; in fact, matrix inversion is one of three typical patterns (L. J. Green, Reference Green2002, p. 87).

A further observation may be made with examples from Tyneside English (my own childhood dialect), spoken in the northeast of England. First, not only is inversion possible with interrogative complementizer deletion, but it is obligatory, this being the only manifestation of the embedded clause’s interrogative force, as illustrated in (19). For resolutive verbs, which, unlike question verbs, select both that and if/whether, there is, in fact, no ambiguity in embedded clauses: Inversion indicates if-deletion, and noninversion indicates that-deletion, as in (20).

1. (19) a. Terry asked Julie would she like to gan to the beach. ^{Footnote 11}

2. b. *Terry asked Julie she would like to gan to the beach.

1. (20) a. Me mam wants to know was me dad happy yesterday. (if-deletion)

2. b. Me mam wants to know me dad was happy yesterday. (that-deletion)

Such distribution appears to hold in the other dialects considered here (L. J. Green, personal communication, August 6, 2010; A. Henry, personal communication, June 13, 2011; J. McCloskey, personal communication, June 23, 2011). The relevant syntactic condition appears to be as follows:

1. (21) In languages with inversion in matrix clauses, optional deletion of an interrogative complementizer results in obligatory inversion in embedded clauses.

This brief survey of L1 dialects has established that inversion in embedded clauses is a relatively widespread phenomenon that is not particular to Indian English and is everywhere subject to certain syntactic and pragmatic constraints, as would be expected on any formal analysis in which the CP projection is the target of the relevant movement operations. None of the varieties discussed violate the penthouse principle, in that inversion is allowed freely in main clauses, even though it is not always obligatory.

Embedded Inversion in L2 Varieties

Colloquial Indian English is not a L1 dialect but a L2 system acquired after the supposed critical period for syntax (Johnson & Newport, Reference Johnson and Newport1989; Long, Reference Long, Doughty and Long2003; Meisel, Reference Meisel2009; Newport & Supalla, Reference Newport and Supalla1990), as discussed earlier. The assumption of such a critical period has informed various hypotheses over the years that claim either partial or complete impairment of knowledge of UG following this sensitive window (Clahsen & Muysken, Reference Clahsen and Muysken1989; Hawkins & Casillas, Reference Hawkins and Casillas2008; Hawkins & Chan, Reference Hawkins and Chan1997; Kanno, Reference Kanno1996; Meisel, Reference Meisel2009; Schachter, Reference Schachter, Gass and Schachter1989; Tsimpli & Dimitrakopoulou, Reference Tsimpli and Dimitrakopoulou2007). Second language grammars that purportedly violate universal constraints have been called rogue (Thomas, Reference Thomas and Eubank1991), illicit (Hamilton, Reference Hamilton1998), or wild (Goodluck, Reference Goodluck1991). If inversion were truly restricted to embedded clauses in L2 English in India, then it would be a rogue grammar in this sense.

However, as noted previously, I adopt the alternative hypothesis, which is that L2s are natural languages in the same sense as L1s (e.g., Adjemian, Reference Adjemian1976; Eckman, Reference Eckman1981; Gass, Reference Gass, Behney and Plonsky2013; Schwartz & Sprouse, Reference Schwartz and Sprouse1996; Sharwood Smith, 1988; White, Reference White2003). This is perhaps most succinctly expressed by the interlanguage structural conformity hypothesis (Eckman, Moravcsik, & Wirth, Reference Eckman, Moravcsik and Wirth1989, p. 195):

1. (22) All universals that are true for primary languages are also true for interlanguages.

Embedded inversion provides a testing ground for this hypothesis, yet surprisingly there appears to have been little systematic investigation of this topic in SLA, despite numerous expressions of theoretical interest, anecdotal reports, and minor experimental reports (e.g., Bley-Vroman, 1997; Borer, Reference Borer1996; Escutia, 2002; Finegan, Reference Finegan1999; Johnston, Reference Johnston1985). A recent L2 study takes a more systematic look at this. Pozzan and Quirk (Reference Pozzan and Quirk2013) investigated inversion in main and embedded clauses by Chinese and Spanish learners of English. Two elicited-production tasks were administered to 80 participants (32 L1 Chinese, 32 L1 Spanish, and 16 L1 English controls). If L1 transfer were involved, one may expect greater rates of inversion in the case of L1 Spanish (which has T-to-C movement in both main and embedded clauses) than in the case of L1 Chinese (which has no T-to-C movement). The results relevant to the current study are as follows. First, for each L1 group, there was more wh-inversion in matrix clauses than in embedded clauses (L1 Spanish, matrix: 85%; L1 Spanish, embedded: 29%; L1 Chinese, matrix: 93%; L1 Chinese, embedded: 29%). This was also true for inversion in yes/no questions (L1 Spanish, matrix: 97%; L1 Spanish, embedded: 2%; L1 Chinese, matrix: 100%; L1 Chinese, embedded: 2%). Second, the low rates of inversion in embedded yes/no questions may be plausibly linked to the overt complementizer if, which appears to be obligatory for most learners in this study (as in Standard English) and which blocks movement from T to C. Thus, L2 learners appear to respect the same universal principles of phrase structure as L1 dialect speakers. Third, there appears to be no transfer effect (although further studies with lower-level learners are required to confirm this). This study remains in need of replication, but the results are likely to be generalizable and echo the findings from L1 dialect research.

The question of whether wh-inversion is restricted to embedded clauses in Indian English is, of course, an empirical one. Although this distribution may seem unlikely in the context of the preceding review, it remains the standard analysis and is in need of systematic evaluation. The implications for L2 research are fundamental: Either colloquial Indian English is an exceptional language variety whose violation of universal principles challenges the notion that L2s are of the same ilk as L1s or, alternatively, it conforms to crosslinguistic constraints and supports the idea that L1s and L2s are equally natural expressions of the human language faculty.