1. Introduction
This study of adverb code-switching aims to provide new insights into the nature of adverbs as a syntactic category, the processes of language contact, and the theory of grammar in general. Adverbs and adverbial phrases are examined in mixed and unmixed Haitian Creole–English second-generation grammar by means of the naturalistic Miami corpora from the years 2004 (9 participants; 34,835 words) and 2005 (12 participants; 54,330 words). Naturalistic interviews were exclusively collected with second-generation bilinguals in order to focus on this group and to avoid skewing results by blending generations. The quantification and analysis of both mixed and unmixed Haitian Creole and English adverbs in these corpora is important because it helps advance the understanding of this unstudied language pair and code-switching generally by capturing and counting all adverbs uttered. Examples of mixed adverbs from the Miami corpus can be found in (1), where the mixed English adverb ANYWAY is surrounded by Haitian Creole, and in example (2), where the mixed Haitian Creole adverb konnya is surrounded by English:Footnote 1
(1)
(2)
The following two subsections present the sociolinguistic and psycholinguistic situation of Haitian-Americans in Miami. Next, current research on monolingual and bilingual approaches to the syntax of adverbs is examined. Exemplification of mixed English and Haitian Creole adverbs by means of the quantitative methodology is followed by the aggregate quantitative results which compare mixed and unmixed adverbs in Haitian Creole and English. The syntax and discourse positions of adverbs are illustrated and quantified; the left-periphery and right-adjunction are shown to be especially accessible positions. Common mixed English adverbs which are partial homonyms such as just – jis/jous or back – bak are examined separately since they are most likely borrowings. Lastly, verb and adverb syntactic interactions in monolingual Haitian Creole and English are compared with anomolous examples from code-switching.
In the first place, the analysis of adverbs here provides evidence regarding the theoretical issue of bidirectionality in a context generally characterized by asymmetry. In the second place, light is shed upon the split between lexical and functional categories. Two related questions are explored: (i) if adverbs are a category that includes lexical and functional elements, as Ernst (Reference Ernst2002) argues, then, in a Matrix Language Frame approach, the code-switching of English adverbs in Haitian Creole should be restricted to lexical adverbs and blocked among functional ones; and (ii) if adverbs are Theta-role receiving ‘content’ elements, as Eckardt (Reference Eckardt1998) argues, then they should be expected in an Matrix Language Frame approach to code-switching. In the third place, examined is the role that the coding of language-indexation has in explaining the continuum between code-switching and borrowing. Our hypothesis is that some English adverbs gradually become borrowings-in-progress as they increasingly occur in strings surrounded by Haitian Creole material and are decreasingly used as switch-points into a co-indexed English string. Our coding of the surrounding language-indexation provides a means to test this quantitatively. In the fourth place, I argue that sociolinguistic and psycholinguistic pressures impact the form taken by the two languages in code-switching.
Code-switching is bidirectional even if it appears to be asymmetrical (see Hebblethwaite, Reference Hebblethwaite2007, Reference Hebblethwaite2009b). The appearance of asymmetry in quantitative work is the result of sociolinguistic and psycholinguistic pressures which cause speakers and communities to select the ‘low’ language as the primary functional frame for code-switching and not the ‘high’ language. I reject the claim that morphological feature-valuation (MacSwan, Reference MacSwan1999 and subsequent work) can be used for predicting the results of code-switching. Rather, along the lines of Myers-Scotton (Reference Myers-Scotton1993, Reference Myers-Scotton2002) or Chan (Reference Chan2008), I argue that the lexical and functional properties of words and the sociolinguistic and psycholinguistic realities of the community provide a better framework for predicting outcomes in code-switching.Footnote 2
1.1 The sociolinguistic situation of Haitian-Americans in South Florida
The state of Florida, according to the 2000 Census, has more than 182,224 Haitian-born residents, and when one takes into account the second generation, the state is home to no less than 300,000 persons of Haitian decent (Sohmer, Reference Sohmer2005). According to the Modern Language Association, Haitian Creole is spoken by 1.78% of Florida's population, making it the third most widely spoken language in this state after English (74.54%) and Spanish (18.65%) (http://www.mla.org). Miami-Dade County has one of the largest Haitian communities in the U.S. with at least 95,669 documented persons (Sohmer, Reference Sohmer2005, p. 3). Fort Lauderdale in Broward County, which is to the immediate north of Miami-Dade, is home to more than 62,324 Haitian-born residents. The size and density of the Haitian population in south Florida is one of the conditions underlying the robust use of code-switching.
The sociolinguistic situation in Miami weighs significantly on the way Haitian-Americans use Haitian Creole and English. Social and economic data show that Haitian Creole has a ‘low’ status and English a ‘high’ one in Miami. The sociolinguistic situation in Miami's Haitian community is one characterized by challenges and struggles that come from a legacy of prejudice. Until 1965, Jim Crow laws banned black Americans from cinemas and restaurants and sent them to pay taxes at a separate window at Miami-Dade's courthouse, among other indignities (Stepick, Reference Stepick1998, p. 38). With this background, the media of the 1970s and 1980s transmitted many hysterical reports about the arrival of destitute and unskilled Haitian boat people off the coast of Florida. On top of this, in 1982 the Centers for Disease Control issued erroneous warnings that Haitians were at-risk for tuberculosis and AIDS (Stepick, Reference Stepick1998, p. 35). As a result of these claims and perceptions, a siege-mentality emerged and Miami's more comfortably installed communities sometimes saw Haitians as a pariah. Consequently, members of the Haitian community had difficulty finding even minimum wage employment (Stepick, Reference Stepick, Grenier and Stepick1992, p. 56).
The quantitative results of Stepick, Stepick, Eugene, Teed & Labissiere's (Reference Stepick, Stepick, Eugene, Teed, Labissiere, Rumbaut and Portes2001, p. 241) questionnaires suggest that Haitian-Americans encounter more racism in public schools than other groups: Haitians experience it at 70.3%, other Antillean groups at 56.6%, and all other groups at 55.1%. The annual median salary in the Haitian-American community is the lowest of all groups in Miami at $27,287 (non-Hispanic Black follows with $28,617, Cuban with $33,427, Puerto Rican with $34,854, Nicaraguan with $35,059 and non-Hispanic White at the summit with $49,673; Sohmer, Reference Sohmer2005, p. 6). Haitian-Americans also demonstrate more ambivalence toward Haitian Creole than other groups to their respective languages. Portes and Schauffler's (Reference Portes and Schauffler1994, p. 648) study in Miami shows that 27.50% of Cubans and 22.10% of Nicaraguans in public schools consider themselves weak in Spanish while 67.80% of Haitians consider themselves weak in Haitian Creole. At the same time, 32.50% of Cubans and 41.40% of Nicaraguans consider themselves strong in Spanish whereas only 11.80% of Haitians consider themselves strong in Haitian Creole. These numbers suggest that linguistic assimilation occurs at a rapid pace in all second-generation groups. However, while all groups express a preference for English, the Haitian second generation takes less interest in the parental language than any other group. To explain this finding, Stepick et al. (Reference Stepick, Stepick, Eugene, Teed, Labissiere, Rumbaut and Portes2001, p. 239) reveals that first-generation parents cope with a range of problems. Haitian parents lack high-school diplomas more than other ethnic groups (60.2% for men and 67.5% for women whereas the average for other ethnic groups is 36.0% for men and 29.6% for women). Haitian parents are more often unemployed; when employed, it is more often at minimum wage in the service sector and they are less commonly employed as managers, administrators, or clerks. 81.9% of Haitian parents earn less than $25,000 annually and consequently they lack the means to pay for their children's enrollment in university and they are unwilling to go into debt to pay for it. The lack of parental skills, economic immobility, and a negative reception in the U.S. contribute to the language stress in the Haitian-American second generation. These factors, among others, explain the ‘cultural dissonance’ and indifference to parental emblems like Haitian Creole and they also explain the asymmetry involving the preference of Haitian Creole as the frame for code-switching (see Hebblethwaite, Reference Hebblethwaite2006; Stepick et al., Reference Stepick, Stepick, Eugene, Teed, Labissiere, Rumbaut and Portes2001, p. 231).
1.2 Psycholinguistic support for asymmetrically established languages
Psycholinguistic research supports the view that many bilinguals have asymmetrically established languages. Meuter (Reference Meuter2005, pp. 353–355) shows through calculating “switch cost from the L2 to the dominant L1” that non-balanced bilinguals have asymmetrically established languages. Among her participants, switching from the L2 to the stronger L1 required 143 milliseconds while switching from the L1 to the weaker L2 required 85 milliseconds, which she calls “an asymmetrical language switch cost”. This is due to the greater attention demands made by processing the weaker L2. Brain imaging studies suggest that more brain regions are activated for the L2; additionally, lower proficiency bilinguals require the activation of more brain regions in the processing of two languages (see Dehaene et al., Reference Dehaene, Dupoux, Mehler, Cohen, Paulesu, Perani, Moortele, Lehéricy and Le Bihan1997; Perani et al., Reference Perani, Dehaene, Grassi, Cohen, Cappa, Dupoux, Fazio and Mehler1996). Balanced bilinguals were able to decrease the L1/L2 proficiency disparity with practice. Meuter (Reference Meuter2005, p. 365) finds that “when speaking the L2, perhaps more so in an L2 setting, there is a continual need to suppress the dominant language”. Of course, the issue of which language is the L1 and which is the L2 is murky in the second generation due to the range of acquisition and affective scenarios (e.g. Does the child speak Haitian Creole and English at home? Do one or both parents speak one or both languages? Do siblings speak English, Haitian Creole or both? Does the bilingual go to an English or Haitian Creole daycare setting? Do the L1 and L2 shift based on social allegiances and peer and sibling influence? etc.). While there are complex differences in families, it is clear that the parental language and the societal language are not known or treated in the same ways. Most of the participants in my study display non-balanced knowledge of their two languages and they code-switch asymmetrically.
Grosjean (Reference Grosjean, de Groot and Kroll1997) pointed out that there are many circumstances in which the bilingual must speak in only one language because the interlocutor only speaks one language. Practice, upbringing, schools, and the necessities of interacting in the ‘high’ status language dictate the impermeability of English in the U.S. Very differently, second-generation Haitian Creole speakers within the Miami Haitian community can employ either language; if the speaker selects Haitian Creole, it is usually characterized by code-switching. In production terms, when a Haitian Creole frame is selected, the activation and selectional processes access Haitian Creole and English lexical elements. When an English frame is selected, however, only the English lexicon is accessed while the Haitian Creole lexicon is closed off due to sociolinguistic and psycholinguistic pressures.
2. Monolingual approaches to adverbs
Monolingual approaches to adverbs provide hypotheses and insights that can be extended to adverbs in bilingual production. Adverbs are a diverse and complex group of elements which modify several syntactic categories such as adjectives in (3), auxiliary verbs in (4), verbs in (5), adverbs in (6), clauses in (7), and prepositions in (8):
(3) You are extremely kind.
(4) They certainly have arrived.
(5) We run fast.
(6) In fact we run very fast.
(7) Suddenly she found herself in the same situation.
(8) He walks right into the door.
(see Ernst, Reference Ernst2002; Radford, Reference Radford1997)
Derived from adjectives (slow > slowly) or nouns (love > lovely), adverbs may receive a pronounced or a non-pronounced adverbial affix (e.g. broadly vs. too; Manninen, Reference Manninen2000). As a reviewer noted, the suffix -ly can also turn a noun into an adjective, e.g. compare She looked lovely with She is a lovely lady. This fact might provide a clue about the lexical traits of the suffix -ly since adjectives are a lexical category. Adverbials, which are elements that function like adverbs, may also include entire PPs, exemplified in (9) (Ernst, Reference Ernst2002, p. 1).
(9) He typed in a hurry.
Adverbs are adjuncts because they form an expanded category of the same type as the one they merge with and because they modify verbs (Radford, Reference Radford1997, p. 371). Adverbial adjuncts give supplementary information about a predicate whereas nominal arguments give information required by a predicate (Ernst, Reference Ernst2002, p. 7).
Recent generative work has argued that adverbs are “overt manifestations of functional distinctions similar to particles, overt morphology and auxiliaries” (Alexiadou, Reference Alexiadou2004, pp. 677–678). Pollock (Reference Pollock1989) uses French and English verb placement to suggest that adverbs occupy Specifiers in an expanded functional IP; French verbs move leftward past adverb-occupied Specifiers to a higher functional head. Cinque (Reference Cinque2004, pp. 684–685 and his footnote 4) argues that adverbs are a functional category because their emergence in first language acquisition is connected to the functional heads they correspond to. Evidence of their functional status is also found in the Sino-Tibetan languages Boro and Garo, and in the Uto-Aztecan language Chemehuevi, where adverbs are mostly bound morphemes, like other functional elements. Additionally, Cinque (Reference Cinque2004, p. 685) notes that there is a class of adverbs which has morphological expression through a suffix in every language.
Radford (Reference Radford1997, p. 45) takes the semantic properties of adverbs to be an indication that they are lexical. Cinque (Reference Cinque1999, pp. 77–107) also identifies an extensive list of adverb semantic classes (also see Huang, Reference Huang1975, p. 13).Footnote 3 Cinque (Reference Cinque1999, Reference Cinque2004) argues that adverbs denote functional projections but as a whole he treats them as the intersection of functional and lexical characteristics. Ernst (Reference Ernst2002, p. 9) considers adverbs to have lexical (10a–c) and functional (10d) members:
- (10)
a. predicational
speaker-oriented: frankly, maybe, luckily, obviously
subject-oriented: deliberately, stupidly
exocomparative: similarly
event-internal: tightly, partially
b. domain: mathematically, chemically
c. participant: on the wall, with a bowl, for his aunt
d. functional
time-related: now, for a minute, still
quantificational: frequently, again, precisely
focusing: even, just, only
negative: not
clausal relations: purpose, causal, concessive, conditional, etc.
Functional adjuncts under (10d) are a heterogeneous group, set apart from the others by their non-gradability (e.g. she speaks very/too/extremelyquickly versus *she speaks very/too/extremelynow) and their quantificational meaning (she twice knocked on the door, example from Cinque, Reference Cinque1999, p. 26) (Ernst, Reference Ernst2002, p. 9, 120). Given these predictions, in an approach like the Matrix Language Frame, where functional elements are supplied by the Matrix Language and content elements by the Embedded Language, one would expect code-switching to be restricted to the lexical class of adverbs and blocked among the functional class of adverbs, a matter I turn to ahead.
Samuelsdorff (Reference Samuelsdorff, Olbertz, Hengeveld and García1998, pp. 275–277) argues that there are functional adverbs, such as very, and lexical ones, such as cautiously. In some languages the equivalent of very can be expressed by means of a grammatical suffix, such as -issimo in Italian:
(11) benissimo
“very well”(Samuelsdorff, Reference Samuelsdorff, Olbertz, Hengeveld and García1998, p. 275, ex. (14))
Adjective-modifying degree ‘adadverbs’ like very and absolutely have “grading properties” as a class since they express the “extent of a quality or state” (Samuelsdorff, Reference Samuelsdorff, Olbertz, Hengeveld and García1998, p. 275). Samuelsdorff (Reference Samuelsdorff, Olbertz, Hengeveld and García1998, p. 275) argues that it is not necessary for an extra entry to exist in the lexicon for a manner adverb like cautiously if “expression rules” can add -ly to the adjective. Adverbs which are derived from adjectives should not be placed in separate entries in the mental lexicon since they are generated by derivational morphemes. The stem is lexical and the affix is functional, suggesting that adverbs are composite.
Eckardt's (Reference Eckardt1998) work on adverbs has implications for code-switching. Eckardt (Reference Eckardt1998, p. 10) explores the idea that verbs have both nominal and adverbial thematic or “event” arguments. In the case of adverbs, the event argument “stands for the action going on, the thing that happens” (Eckardt, Reference Eckardt1998, p. 10). The sentence Tom kissed Sylvia tenderly could be represented as KISS (TOM, SYLVIA, e) & TENDER (TOM, e), where there is a two-place relation between the individual and the event (Eckardt, Reference Eckardt1998, p. 10). The event parameter creates “an intermediate level of intentionality restricted to the verb, but not affected by nominal arguments” (Eckardt, Reference Eckardt1998, p. 12). Adverbs are “functors that map binary relations between individuals and events” (Eckardt, Reference Eckardt1998, p. 12). Verbs and their adverbial events have a Thematic relation, for example, SPATIAL (inside), TEMPORAL (yesterday), BENEFICIENT (for Susan), INSTRUMENTAL (with a knife), VOLITIONAL (shyly/carefully/reluctantly), and so forth (Eckardt, Reference Eckardt1998, p. 34).Footnote 4
To illustrate how adverbs are thematic event arguments for verbs, Eckardt (Reference Eckardt1998, p. 38) explores semantic differences that arise when adverbs are added to sentences. The event cooking in (13) ceases to compete with the other hypothetical causes faced by (12) in the explanation of the neighbor's calling the police.
(12) Pat was cooking spaghetti, which caused the neighbor to call the police.
(adapted from Eckardt, Reference Eckardt1998, p. 65, ex. (17))
(13) Pat was cooking spaghetti late at night, which caused the neighbor to call the police.
(adapted from Eckardt, Reference Eckardt1998, p. 65, ex. (18))
Eckardt (Reference Eckardt1998, p. 67) shows that the set of relevant alternative worlds is restricted by the presence of the adverb. Adverbs predicate over events and provide a thematic focus (Eckardt, Reference Eckardt1998, p. 43). Eckardt's (Reference Eckardt1998) idea that adverbs are Theta-role receiving event arguments subcategorized in verbs provides Matrix Language Frame advocates with an explanation for the frequency of adverbial code switching and borrowing since the Matrix Language Frame model predicts code-switching among elements which are either Theta Role-Assigning or Theta Role-Receiving (Myers-Scotton, Reference Myers-Scotton1993, pp. 6–7).
Manninen (Reference Manninen2000, p. 153) argues that Finnish adverbs are lexical because they are a subclass of adjectives and nouns, not a separate category. Finnish adverbs are noun or adjective stems with an attached “adverb case” morpheme (e.g. -ly, -ment, etc.). In situations where the stem and affix are indistinguishable, Manninen (Reference Manninen2000, p. 153) posits an abstract morpheme. Adjectives and adverbs share pre-modifiers like quite, almost, surprisingly, physiologically (ibid., p. 136). Along the lines of Larson (Reference Larson1988), nouns or adjectives become adverbs by means of a Case Phrase (KP) or a PP (Manninen, Reference Manninen2000, pp. 146–148). Attracted by features, the adjective or noun raise to Spec KP or Spec PP and Spec–head agree with K°/P°, which is either pronounced or covert.
Following Eckardt's (Reference Eckardt1998) arguments, I expect adverb code-switching in a Matrix Language Frame approach since they are Theta-role receiving ‘content’ elements. A final argument for the lexical status of adverbs comes from the fact that they are not a closed class such as determiners but rather a fairly large and expandable one (e.g. there are 1,525 adverbs counted in the CD-ROM version of Le Petit Robert, p.c. Julie Auger). The next section turns to work that focuses on adverbs in code-switching.
2.1 Adverbs in bilingual grammar
With a few exceptions, very little work has been done on adverbs in bilingual grammar. Treffers-Daller (Reference Treffers-Daller1994, p. 175) finds 388 mixed French adverbs in her Brussels Dutch–French corpus, noting that they are among the most common mixed categories. With 3,988 French single-word switches in Treffers-Daller (Reference Treffers-Daller1994, p. 99), syntactic categories are ranked by frequency: (i) nouns at 58.4% (2,329/3,988); (ii) interjections at 12.4% (496/3,988); (iii) adverbs at 9.7% (388/3,988); (iv) adjectives at 9.0% (362/3,988); (v) verbs at 8.9% (353/3,988); (vi) conjunctions at 0.8% (33/3,988); (vii) prepositions at 0.1% (4/3,988); (viii) pronouns at 0.02% (1/3,988); (ix) and other elements including numerals, wh-words, quantifiers and particles at 0.5% (21/3,988).
Treffers-Daller's (Reference Treffers-Daller1994) work attends to both Dutch–French code-switching and French–Dutch code-switching. As I have done, she identifies all mixed and unmixed structures. She identifies 8,500 unmixed Brussels Dutch adverbs and 0 mixed Dutch adverbs. Oppositely, she identifies 1,788 unmixed French adverbs and 388 mixed French adverbs (Treffers-Daller, Reference Treffers-Daller1994, pp. 103–104). She also finds 45 (full constituent) French PP tokens in Dutch (ibid., p. 208). Treffers-Daller's (Reference Treffers-Daller1994, pp. 99–104, 205) work shows that linguistic influence in Dutch–French is asymmetrical: across all syntactic categories, 3,988 single-word French switches occur in Dutch clauses whereas only 117 Dutch single-word switches occur in French clauses. Social structure is identified as the source of this asymmetrical influence since “French has been the upper language and Brussels Dutch the lower language for generations of Brusselers” (Treffers-Daller, Reference Treffers-Daller1994, p. 106).
Originally, Myers-Scotton (Reference Myers-Scotton1993, p. 129) classified adverbs as “system” or functional morphemes (as opposed to content-morphemes). She argued that adverbs are system morphemes that “pick out individuals across variables” (Myers-Scotton, Reference Myers-Scotton1993, p. 129). No singly occurring Embedded Language (EL) adverbs (i.e., mixed English) are found in her Swahili–English corpus whereas 20 “Embedded Language Islands”, which are strings that include more than one word, occur with adverbs as their leftmost element, for example m-na-jua VERY WELL “you know very well” or ni-ko VERY SURPRISED “I am very surprised” (e.g. the xYadvy structure in my coding) (Myers-Scotton, Reference Myers-Scotton1993, p. 129). However, Myers-Scotton (Reference Myers-Scotton2002, p. 70) changes her position, now arguing that adverbs form a category that includes both the system members (such as degree adverbs very, too) and content members. Yoon (Reference Yoon1992, p. 438) does not discuss the adverbial tokens he collects from his Korean–English corpus, but, he counts 41 mixed English adverbials, which amount to 7% of all his switched elements. De Rooij's (Reference De Rooij1996) treatment of adverbs in Shaba Swahili–French is limited to a single page and three examples of mixed French adverbs.
MacSwan's (Reference MacSwan1999) minimalist-inspired approach appeals to morphological mechanisms, especially [+] and [−] feature valuation, as putatively the most parsimonious method to account for both monolingual and code-switching modes of speech. Locating grammatical requirements in lexical items, he argues that “conflicts in the lexical requirements of words” explain the grammar of the code-switching clause (MacSwan, Reference MacSwan1999, p. 151). Unfortunately, MacSwan (Reference MacSwan1999) offers only one paragraph on adverbs in code-switching. The lacuna is attributable to his emphasis on grammaticality judgments when a large corpus is needed to uncover a full range of mixed structures. In his modest corpus of 20 minutes he does find a handful of Spanish adverbs in Nahuatl: ya “already”, muy “very”, de nuevo “again”, and más o menos “more or less” (MacSwan, Reference MacSwan1999, p. 142). Regarding the scant literature on adverbs in code-switching, MacSwan (Reference MacSwan1999, p. 142) speculates that as adjuncts they “do not interact with other elements in any obvious way”. He does not state anything else; however, one might surmise, given his focus on phi-feature agreement, that he is referring to the absence of morphosyntactic agreement associated with adverbs or that there is no constraint on which language an adverb can occur in.
While MacSwan (Reference MacSwan1999, etc.) does not analyze mixed adverbs, it is important to attempt to reconstruct how a minimalist approach might be applied to Haitian Creole–English adverb code-switching. MacSwan's theory predicts that there are no particular constraints on adverbs because they do not have features to check. Once we are in the domain of an element that has no particular grammatical constraints on it, one would think that MacSwan is free to appeal, as I do, to the asymmetry of the sociolinguistic and psycholinguistic situation as the reason for one language being selected as the functional frame (in both my corpora and his, the ‘low’ language). However, MacSwan (Reference MacSwan1999, p. 39) explicitly says that “there is little hope that a set of explanatory principles can reliably be developed which predict when and how people will say what” and “while social motivations for code-switching can be catalogued and described in interesting ways, theoretical constructs of general or universal use may be quite beyond our reach”. Given his dismissal of sociolinguistic factors, what poses problems for his approach is the actual identification and valuation of adverb features in code-switching. In Haitian Creole and English, for example, adverbs do not contrast morphosyntactically in any obvious way. Evidence of bidirectional switching found in the naturalistic data is problematic for a feature valuation approach which predicts switching in only one direction; conversely, if adverb features are congruent, the question then becomes why the results are overwhelmingly asymmetrical? The next section turns to recent theoretical discussions about code-switching and it explores the implications they have for adverbs in bilingual production.
2.2 The Null Theory (Chan, Reference Chan2008) and the 4-M Model (Myers-Scotton, Reference Myers-Scotton2002)
Chan's (Reference Chan2003, Reference Chan2008) theory on the role of lexical and functional categories and his implementation of the “the Null Theory” have interesting implications for adverbs in bilingual production. While Chan's (Reference Chan2008) work draws from early proponents of the Null Theory (Mahootian, Reference Mahootian1993; MacSwan, Reference MacSwan1999, Reference MacSwan2000), his position also marks a clear break from them. The proponents of the Null Theory have claimed that it is more economical to rely only on those constraints and principles already accepted for monolingual grammar and they have criticized theories which appeal to constraints and principles specific to code-switching (i.e., Myers-Scotton, Reference Myers-Scotton1993, Reference Myers-Scotton2002). Chan (Reference Chan2003, Reference Chan2008, p. 778) argues that the dichotomy between lexical and functional categories provides the best insight into the grammar of code-switching. Functional categories, he argues, inherently determine the position of their complements, whether in monolingual or bilingual clauses.
Chan (Reference Chan2008, p. 779) advances that directionality parameters account for word order in code-switching better than constraints and principles specific to code-switching proposed by proponents of the Matrix Language Frame model such as the Morpheme Order Principle, the System Morpheme Principle, and the 4-M Model, among others (see Myers-Scotton, Reference Myers-Scotton1993, Reference Myers-Scotton2002). The distinction between lexical categories (N and V) and functional categories (D, I and C) provides the simplest basis for accounting for word order in code-switching (Chan, Reference Chan2008). Rejecting the claims of MacSwan (Reference MacSwan1999), Mahootian (Reference Mahootian1993) and Nishimura (Reference Nishimura1997), who argue that the phi-feature conflict between lexical and functional heads determines word order and grammaticality in code-switching, Chan (Reference Chan2008, p. 780) takes the position that it is the functional category, and not the lexical one, which determines word order and grammaticality. The parameter value, for example Verb–Object/Object–Verb or Determiner–Noun/Noun–Determiner, is not linked to specific verbs or nouns and it is hence not a part of their lexical subcategorial properties or lemma (Chan, Reference Chan2008, p. 787). The bilingual who, say, speaks both V–OX and O–VY languages will then set the parameter value prior to lexicalizing the verb from either language X or language Y. This view explains cases where a V from language X with V–O occurs in O–VX word order or a V from language Y with O–V occurs in VY–O word order (see Chan, Reference Chan2008, p. 787).
In addition to his arguments on VO/OV, Chan (Reference Chan2008, p. 792) provides examples which show that functional categories from either language determine word order in code-switching (and thus he, like myself, predicts bidirectionality). Nouns and noun complements (e.g. the genitive) and functional heads and their complements (D–NP, C–IP and P–DP) show that the language of the functional projection (D, C or P) determines the surface position of its lexical complement (NP, IP or DP) regardless of the language of the complement (see Chan, Reference Chan2008, p. 800). Chan (Reference Chan2008, p. 803) argues that since functional categories are closed classes, it is economical to have “the head–complement order specified in the lexical entries of functional categories”. Chan (Reference Chan2008) does not address adverbs; however, as members of a category that is open class and shares a thematic relation with verbs, within Chan's theory adverbs should be expected in code-switching. Chan (Reference Chan2008) predicts bidirectionality but he does not explain why it occurs simultaneously with asymmetry.
Although the proponents of the Null Theory commonly claim that their proposals are more parsimonious than the Matrix Language Frame approach, it is not obvious how Chan (Reference Chan2008) represents such a shift since he also claims, albeit with different terminology, that there is a basic difference between content (lexical) and system (functional) elements. Likewise, Chan (Reference Chan2008) argues that a bilingual clause entails the selection of functional/system morphemes (and their subcategorial properties) from one (“Matrix”) language while the lexical/content morphemes from either (“Matrix” or “Embedded”) language may be accessed. In my view, the terms “functional morpheme/system morpheme/Matrix Language” and “lexical morpheme/content morpheme/Embedded Language” are basically synonymous.
The 4-M Model developped by Myers-Scotton (Reference Myers-Scotton2002) and Jake and Myers-Scotton (Reference Jake, Myers-Scotton, Isurin, Winford and de Bot2009) classifies morphemes into four types. In this model, one morpheme type is a content morpheme and the three others are different types of system morphemes, the “Early system morphemes”, “Bridge late system morphemes”, and “Outsider late system morphemes” (Myers-Scotton Reference Myers-Scotton2002, p. 73). Bridge late system morphemes and Outsider late system morphemes are retrieved at the level of the formulator and generally come from the Matrix Language. The Bridge late and Outsider late system morphemes primarily build grammatical structure and they are not primarily semantically meaningful. Content and Early system morphemes are conceptually-activated and semantically meaningful (2009, p. 11). Content and Early system morphemes typically involve Embedded Language elements and they are the ones most available for code-switching. The morphemes which are conceptually-activated and meaningful are called forth by the intentions of speakers (rather than being “structurally-assigned”). The most important split in the 4-M Model is therefore between conceptually-activated morphemes (content and Early system morphemes) and structurally-assigned morphemes (Bridge and Late system morphemes).
Jake and Myers-Scotton (Reference Jake, Myers-Scotton, Isurin, Winford and de Bot2009) employ the 4-M Model to explain why some syntactic categories, especially prepositions, do not behave uniformly in code-switching. While some EL prepositions are found in code-switching, others are not, and the authors argue that this is because the category of prepositions includes members from the different morpheme types. Noticing the complexity of prepositions, Chan (Reference Chan2008, p. 802) says that “one may conjecture that the traditional distinction between functional and lexical categories is not fine enough . . . there might be two distinctions cross-cutting various categories, namely, closed class/open class words and conceptual/featural (or Theta-assigning/non-Theta assigning) words”. While Chan (Reference Chan2008) rejects the 4-M Model which was introduced to account for fine distinctions, he suggests that the split between functional and lexical categories may not be adequately explanatory. Chan's (Reference Chan2008) generalizations are interesting but they do not fully explain why some Embedded Language prepositions are code-switches and why others are not; this is where the 4-M Model may provide insights. More quantitative work is needed on different types of prepositions or adverbs, for example, to see if they display different quantitative behavior in naturalistic data.
The code-switching in the Miami Haitian Creole–English corpus does not provide evidence that the category of adverbs is composed of various morpheme types. The data in my corpora suggest that adverbs form a lexical category (though this does not mean that the trend holds cross-linguistically). English adverbs identified as functional by Myers-Scotton (Reference Myers-Scotton1993, Reference Myers-Scotton2002) such as the degree adverb very can also be found in a Haitian Creole frame. This does not fully negate her idea that degree adverbs are different than other adverbs; however, it does suggest that whatever the differences are, they did not block their appearance in Miami Haitian Creole–English. The next section examines the quantitative methodology used to quantify the adverbs in my corpus.
3. The quantitative methodology exemplified with adverbs
I coded the transcribed corpora using a system which codes the language-indexation of the target token in addition to the tokens to the immediate left and right. This approach has gradually evolved from the work of Myers-Scotton (Reference Myers-Scotton1993, etc.), Treffers-Daller (Reference Treffers-Daller1994), and especially Klintborg (Reference Klintborg, Ureland and Clarkson1996) and De Rooij (2000). To obtain results that are complete, I coded the entire corpus including both mixed Haitian Creole and mixed English tokens (in both the 2004 and 2005 corpora), and unmixed Haitian Creole and unmixed English tokens (in only the 2004 corpus). (It is essential therefore to separate the results of the totally coded 2004 corpus from the partially coded 2005 corpus.) Schematically, if English (Eng) is represented by the letter Y and Haitian Creole (HC) is represented by the letter X, this bidirectional coding method produces six mixed types and two unmixed types (the lower-case letters x and y indicate the language indexation of the word that precedes or follows the upper-case target token Y or X):
To illustrate this method, examples of all mixed adverb types are provided below.
Mixed English Adverbs Footnote 5
Type (i) [xYx]: MAYBE
(14)
Type (ii) [xYy]: ANYWAY
(15)
Type (iii) [yYx]: MAYBE
(16)
Mixed Haitian Creole Adverbs Footnote 6
Type (i) [yXy]: konnya “now”
(17)
Type (ii) [yXx]: trè “very”
(18)
Type (iii) [xXy]: twò “too”
(19)
A supplementary coding system is used in cases where the surrounding elements are not representative of the language of the clause as a whole. The supplementary coding system takes into account the language of the clause Io in order to determine whether the token is mixed1 (not the same language as I°) or mixed2 (the same language as I°). Following a number of authors, I assume that I° determines the base or Matrix Language of the clause (see Klavans, Reference Klavans, King and Maley1985; Prince & Pintzuk, Reference Prince and Pintzuk2000 [1983]). The mixed English tokens in (14)–(16) and the mixed Haitian Creole tokens in (17)–(19) can be divided into these two types.
(20) Mixed1 is a mixed type that is not in the same language as I°.
(21) Mixed2 is a mixed type that is in the same language as I°.
On the basis of these definitions, mixed English adverbs MAYBE in (14)–(16) are mixed1 tokens since Haitian Creole supplies the Io (ka, pa and te, respectively). The mixed Haitian Creole adverb konnya “now” in (17) is mixed1 because the Io of the clause is the English copula. In (18), however, trè “very”, is mixed2 since Haitian Creole's null copula is the clause's I°. Lastly, twò in (19) is also mixed2 since it shares the same Haitian Creole language-indexation as the Io of the clause, the negation marker pa. Clearly the clause is Haitian Creole and the English adjective loud has been inserted. The issue of differentiating mixed1 versus mixed2 will be an important consideration further ahead. The next section presents the quantitative results and discusses the findings.
3.1 Results and discussion on adverbs
In Tables 1–4, I give the aggregate quantitative results for all mixed and unmixed adverbs from the Miami Haitian Creole–English corpora. Due to the relatively small number of mixed adverbs, I focus on the aggregate results. Given that individual adverbs occur in only a few tokens or at most under 20 tokens, the aggregate analysis is the best way to obtain a meaningful generalization about the surrounding language-indexation. Notice that the number 1,176 represents the total number of mixed and unmixed adverb tokens identified in the 2004 corpus.Footnote 7 Recall that there are three mixed token types and two directions: therefore, where Y = English and X = Haitian Creole, I find, xYadvx, xYadvy, and yYadvx token types alongside yXadvy, yXadvx, and xXadvy.Footnote 8
In the next sections I interpret the above results and examine what they suggest about the grammar of mixed and unmixed adverbs. Comparing all adverb tokens from the 2004 corpus, unmixed Haitian Creole adverbs account for 66.7% (795/1,176) of all tokens and unmixed English adverbs account for 19.8% (234/1,176). Hence, unmixed adverb structures are preferred in Haitian Creole–English for a combined 86.5%. Mixed English adverbs in Haitian Creole account for 12% (142/1,176), a striking percentage compared to mixed Haitian Creole adverbs which are the least numerous at 1.27% (15/1,176). The commonness of mixed Haitian Creole adverbs and the rarity, but not invisibility, of mixed English ones is accounted for with the notions of asymmetry and bidirectionality. Sociolinguistic and psycholinguistic realities ensure asymmetry in Haitian Creole–English. At the same time, traces of code-switching in the opposite direction, that is English–Haitian Creole, suggest that bidirectionality exists theoretically and abstractly but is suppressed in linguistic performance for the reasons given.
In the mixed adverb results, the HC–Engadv–HC token Type (i) predominates in the 2004 corpus at 38% and in the 2005 corpus at 50.8%. The fact that so many English adverbs are completely surrounded by Haitian Creole suggests that their inclusion is unproblematic. To assess the borrowing versus code-switching status of mixed English adverbs in Haitian Creole, occurrences of Haitian Creole near-equivalents have also been counted.
Four groups of mixed English adverbs can be identified. Group 1 English adverbs in Table 5 occur infrequently while the Haitian Creole equivalents are common. Group 1 English adverbs are most likely code-switches. In group 2, the number of English tokens is still small whereas the Haitian Creole equivalent is more limited than the Haitian Creole adverbs in group 1. These mixed English adverbs are possibly borrowings-in-progress; however, the number of overall tokens is too low for any certainty (see Table 6). In group 3, the mixed English adverb is more common than the Haitian Creole equivalent. Group 3 adverbs are most likely borrowings-in-progress or borrowings (see Table 7). Finally, group 4 includes mixed English adverbs that scarcely occur and whose Haitian Creole equivalent is rare or non-existent. Results are given in Table 8.
Given the predictions of the Matrix Language Frame where functional elements are supplied by the Matrix Language and content elements by the Embedded Language, one would expect code-switching to be restricted to the so-called lexical class of adverbs and blocked among the so-called functional class of adverbs. However, this pattern was not found; for example, among time-related adverbs, identified as functional by Ernst (Reference Ernst2002), now was attested in the speech of two speakers in three tokens, for right now (similar to his for a minute) occurred in one token, and still was attested among four speakers and in nine tokens. Among quantificational adverbs, exactly (similar to Ernst's precisely) occurred in one token. Among focusing adverbs, even was found in two speakers and four tokens; just, although excluded due to convergence with the Haitian Creole cognate jis, occurred abundantly; and only occurred in one token. Among negative adverbs, not yet was found in two speakers and two tokens. Various adverbs which are classified as functional by Ernst (Reference Ernst2002) appeared, counter to expectations, in my code-switching corpus. Therefore, based on my corpus, I take the position that adverbs behave like a lexical category and not one that includes lexical and functional members.
3.2 Mixed adverbs in Miami Haitian Creole–English
Table 9 provides a list of all the mixed English adverbs and adverb phrases found in the Miami Haitian Creole corpora. The list indicates the distribution and frequency of each element in order to provide a basis for evaluating the items as borrowings, code-switches or some mixture of both. In addition, the occurrences of the nearest Haitian Creole equivalent are counted as a means of assessing the code-switching versus borrowing status of the mixed English adverb.
This complete listing shows that the majority of mixed adverbs in the corpus occur only once, suggesting that they are most likely code-switches. Adverbs with higher rates of occurrence like sometime(s) and still are possibly borrowings-in-progress, a topic taken up in the next section.
3.3 Frequent mixed English adverbs in Haitian Creole
In Table 10, I provide the most frequent mixed English adverbs. The total number of tokens for any given adverb rarely rises above 10. Adverbs with several tokens show the range of surrounding language-indexation token types.
The surrounding language-indexation results given in Table 10 provide a few insights into mixed adverbs and the continuum that links code-switching, borrowing-in-progress and integrated borrowing (also see Treffers-Daller, Reference Treffers-Daller2005, p. 500). The mixed English adverbs (e)specially, maybe, still and then are perhaps borrowings-in-progress given that these tokens are most often surrounded by Haitian Creole (i.e. the high proportion of the HC–Eng–HC types). I call these items ‘borrowings-in-progress’ rather than ‘borrowings’ because they all include tokens with adjacent English (e.g. HC–Eng–Eng or Eng–Eng–HC). English adverbs like anyway, even, more and probably, which are often adjacent to other English elements, are more weighted toward code-switching (rather than borrowing) given that they share language-indexation on at least one side in most of the token types (e.g. HC–Eng–Eng or Eng–Eng–HC). Thus the English adverbs with adjacent English show that these adverbs are in between fully integrated borrowings and code-switches. The next section examines the role that discourse position has in code-switching.
3.4 English adverb syntax in Haitian Creole–English code-switching and the role of discourse position
Adverbs modify a variety of syntactic categories and occur in different syntactic domains. There are at least four sentential domains in Miami Haitian Creole–English code-switching; (a) and (d) are peripheral and (b) and (c) occur in the midst of argument structure.
(a) Left-periphery-adverb in CP
(b) IP-adverb
(c) VP-adverb
(d) Right-adjoined (e.g. phrase-final) modifying IP, VP, or CP
Example (22) below illustrates type (a) left-periphery-adverbs that occur higher than the subject in Spec IP. The majority of mixed English adverbs occur in this discourse-prominent position – 54% (n 68/124) in the 2004 corpus and 52% (n 41/78) in the 2005 corpus.Footnote 9 As one reviewer suggested, this points to the accessibility of the left-periphery syntactic domain and its discourse-prominence.
(22)
Example (23) shows the type (b) IP-adverb that occurs after the subject pronoun but before the finite verb bezwen “need”. This position is less accessible since mixed English adverbs only target the position at 9.6% (n 12/124) in the 2004 corpus and 8.9% (n 7/78) in the 2005 corpus.
(23)
The example in (24) illustrates the type (c) mixed English VP-adverb in Haitian Creole–English. Once again mixed English adverbs targeting this position are few in number with only 9.6% (n 12/124) in the 2004 corpus and 10.2% (n 8/78) in the 2005 corpus.
(24)
The final example, in (25), illustrates a type (d) right-adjoined, clause-final adverb. The verb mache “to walk”, takes the adjunct PP avèk ayisyen “with Haitians”. In final position, the adverb unfortunately modifies the clause from a hierarchically superior X-bar position (above IP or CP, as the case may be). Mixed English adverbs targeting this position are higher in number with 25% (n 32/124) in the 2004 corpus and 23% (n 18/78) in the 2005 corpus. The higher number of mixed tokens targeting this position suggests that it is also discourse-prominent like type (a).
(25)
Mixed English adverbs are most frequently used in discourse-prominent syntactic positions. Syntactic positions which are peripheral to argument structure (i.e., subjects, objects and verbs) are the most accessible in adverb code-switching. The left-periphery is the most accessible of all, a matter I attend to in the next section.
3.5 Adverbs and left-peripheral elements
As shown above, mixed English adverbs target the left-periphery in the majority of tokens. Indeed, adverbs share affinities with other left-peripheral elements like discourse markers, conjunctions, complementizers, wh-words, and coordinators. All the categories which target extended CP can serve as left-periphery boundaries for code-switching or can be combined into a left-periphery chunk. Torres (Reference Torres2002) and Hebblethwaite (Reference Hebblethwaite2007) show that the left-periphery serves as one of the most active syntactic domains for code-switching. In addition to [CPX [CPY]] and [CPY [CPX]], the structures [CPX [IPY]] and [CPY [IPX]] are very common.
Example (26) shows a long chunk of left-periphery elements, e.g. discourse markers so and well, the conjunction but, and the adverbial then.
(26)
Of course, as illustrated by the English adverb USUALLY in (27), sometimes mixed English adverbs target the left-periphery as lone elements:
(27)
The temporal adverbial wh-word when is equivalent to Haitian Creole's lè and kou. This HC–when–HC sequence is the only mixed token of lone when in the corpora:
(28)
In example (29), when marks an intersentential code-switch into English:
(29)
Coordinators like and and ak are also left-peripheral elements as they occupy extended Coordinator Phrases (along the lines of Rizzi, Reference Rizzi and Haegeman1997). One frequently finds the adverb then occurring in the chunk, “and then”, as illustrated in (30):
(30)
The left-periphery serves as an ideal position for code-switching due to its discourse-prominence. Situated outside of the argument structure, the domain is seamlessly populated by mixed tokens. As a site which transitions to and from argument structures and establishes illocutionary force, the left-periphery proves to be the most accessible platform for code-switching.
3.6 English adverb borrowings or borrowings-in-progress in Miami Haitian Creole
There is a small group of common English adverbs in Miami Haitian Creole that is likely comprised of borrowings or borrowings-in-progress rather than code-switches. The elements discussed in this section (just and bak) are not included in the count of mixed English adverbs because they are possibly borrowings and because they overlap with Haitian Creole partial homonyms.
Just
The adverb just is very frequent in Miami Haitian Creole and other languages in contact with English, such as Ontarian French (Rehner & Mougeon, Reference Rehner and Mougeon1997). It is a cognate of the Haitian Creole adverb jis/jous “just” and French juste. Twenty-six tokens of the HC–just–HC sequence occur in the speech of several participants:
(31)
There are ten tokens of the HC–just–Eng sequence and eight tokens of Eng–just–HC. Each sequence is illustrated in (32) and (33), respectively. English just in (31) above is likely a borrowing-in-progress that is already closely related to jis/jous. At the same time, just is linked to code-switching, as can be seen in (32) and (33).
(32)
(33)
There are several similarities between Ontarian French and Miami Haitian Creole with respect to this adverb. In both contact varieties just frequently undergoes consonant cluster simplification st > s (Rehner & Mougeon, Reference Rehner and Mougeon1997, p. 95). Rehner and Mougeon (Reference Rehner and Mougeon1997, p. 102) speculate that the increase of juste in these French-weak Ontarian communities is “promoted by a process of English language convergence”. Miami Haitian Creole is also subject to such convergence between Haitian Creole's jis/jous/jus and English's just/jus, which is why I keep such items apart from the count of code-switches.
Bak
The English adverb back has long been borrowed into the Haitian Creole spoken in Haiti. It is possible that back entered the language during the US occupation of Haiti (1915–1934) and it is found in all late 20th-century publications. Examples of common expressions include fè bak “go back”; mennen bak “to watch one's step, to watch over, to be careful” in se lidè ki bon pou mennen bak peyi a “this is the leader who is good for watching over the country”; and pa bak “backwards” in mache pa bak “to walk backwards”. In pa bak notice that bak forms the adverb “backwards” with the help of the preposition pa. Bak produces combinations with Haitian Creole verbs that have no English equivalent, e.g. *make/do back. As one reviewer noted, the fact that bak can be used to create structures and phrases that do not exist in English shows that it is not only borrowed in Haitian Creole but has started leading a life of its own.
Back is borrowed in other communities where there is intense contact with English. For example, back is borrowed as baka in Sranan (Blanker & Dubbeldam, Reference Blanker and Dubbeldam2005) and it is also attested in certain Acadian varieties, notably Chiac, a French–English contact language spoken in Moncton, New Brunswick (Perrot, Reference Perrot, Valdman, Auger and Piston-Hatlen2005, p. 314; Tremblay, Reference Tremblay, Brasseur and Falkert2005, p. 263). Bak has even greater extension in Miami Haitian Creole–English due to intense contact with English back. In Miami Haitian Creole back closely resembles equivalent English phrasal verb constructions in common locutions, such as pale back – speak back, in (34):
(34)
Footnote 10 Tokens of bak are counted as unmixed elements because bak is already established in the Haitian Creole spoken in Haiti and thus facilitates an increase in the use of back and complicates identifying back as a borrowing or code-switch.
3.7 Haitian Creole verb/adverb interactions and an anomaly in Haitian Creole–English
As shown in (37) and (38), Haitian Creole clause-internal elements like adverbs and negation cannot intervene between the verb and direct object:
(35)
(36)
(37) *Bouki konnen deja Boukinèt.
(38) *Bouki konnen pa Boukinèt.
(from DeGraff, Reference DeGraff, Cinque and Kayne2005, p. 301, exx. (3)–(4))
Verbs typically remain in-situ in Haitian Creole, which means that they cannot raise past the adverb or negation, as the ungrammatical sentences in (37) and (38) show.
Although participant (001, 2004) is an outlier who is not representative of the sample, examples (39)–(40) show that in her idiolect, the Haitian Creole verb may occur higher than the English adverb. The adverb occurs between the verb and the direct object, in contrast to the canonical position (see DeGraff, Reference DeGraff, Cinque and Kayne2005, Reference DeGraff, Holm and Patrick2007). Examples (39)–(40) show that this position is productive in this speaker's code-switching and they suggest that the participant has activated parametric syntactic values that are not typically associated with the verbs and adverbs involved.
(39)
(40)
Examples (39)–(40) violate Haitian Creole and English word-order rules because neither allows adverbs of this type to intervene between the verb and the direct object; rather both languages require them either before the verb or after the direct object. Note, however, that in sentences like (a), She quickly purchases the television, and (b), She purchases quickly the television, sentence (b) is less degraded than the monolingual English equivalents of (39)–(40). Could participant (001, 2004) be tapping into a marginal syntactic position and extending its availability to other adverbs? Outliers like participant (001, 2004) might provide clues about available parametric settings in Universal Grammar and grammatical change in innovative individuals.
The in-situ analysis of unmixed Haitian Creole verbs does not consistently hold since participant (001, 2004) displays occasional short verb-movement to the head of the light verb past the adverb adjoined to Spec VP. Short verb-movement in Haitian Creole could be the result of English or Haitian Creole influences if single-place predicates such as (a) She ran quickly into the shop, (b) She walked slowly down the street, or (c) Li devlope rapid nan lekòl la “She developed quickly in the school”, etc., serve as syntactic models for new parametric realizations in Haitian Creole–English in spite of restrictions on two-place predicates in the standard languages, e.g. *She walked slowly the dog or *Li devlope rapid yon plan “*She developed rapidly a plan”. These anomalies may suggest that word order is more receptive to change than other aspects of the grammar.
3.8 Either and neither
Polarity adverbs like either and neither provide interesting clues about language contact. Polarity adverbs are bound by a negation operator higher in the IP functional domain. The adverbs either and neither are co-indexed with pa in (42), (44) and (45), and with or in (43). The mixed negation structure can be represented as in (41).
(41)
In (42) the structure pa . . . EITHER forms a bilingual polarity structure. In the EITHER . . . OR polarity structure in (43), observe that the argument structure intervenes.
(42)
(43)
The English negative polarity adverb neither occurs in a double negation pattern in which Haitian Creole supplies the higher I° negation operator and English supplies the lower associate lexical adverb:
(44)
(45)
These data suggest that the morphosyntactic requirements of Haitian Creole negative polarity structures are satisfied regardless of whether the lower associate lexical adverb belongs to the same language or to a different language. In the examples (42), (44) and (45), the Haitian Creole functional negation element, pa, c-commands the lower English lexical adverb, as shown structurally in (41) above, without regard to its language-indexation, suggesting that English adverbs are perfectly felicitous in structures where they directly depend on Haitian Creole functional heads. Example (43) illustrates the English either. . . or polarity structure, which involves no negation and is selected by the Haitian Creole predicates, di “say” and fè “do”.
4. Conclusion on adverbs in Miami Haitian Creole–English code-switching
I have examined current discussions on adverbs and explored the theoretical challenges presented by mixed English and Haitian Creole adverbs using my quantitative corpus methodology. In his mono-linguistic approach to adverbs, Cinque (Reference Cinque2004, p. 685) says that they are “intrinsically functional” and linked to specific functional projections, while he also says (Cinque Reference Cinque1999) that they fit into a range of semantic classes, suggesting lexical properties. Ernst (Reference Ernst2002) argues that adverbs include both lexical and functional types while Eckardt (Reference Eckardt1998) and Manninen (Reference Manninen2000) argue that most adverbs are lexical, and they develop theories that explore the links that adverbs have with predicates, thematic roles, and derivation.
The qualitative and quantitative evidence in Miami Haitian Creole–English code-switching suggests that adverbs behave like members of a lexical category. Indeed, adverbs need to be examined in many more language pairs in order to discover whether they display differences and functional traits in other corpora. If adverbs of different types display different quantitative behavior, a more refined model of morphemes will be needed along the lines of Myers-Scotton (Reference Myers-Scotton2002) or Jake and Myers-Scotton (Reference Jake, Myers-Scotton, Isurin, Winford and de Bot2009).
Ernst's (Reference Ernst2002, p. 9) typology separates lexical adverbs from functional ones. However, as I noted earlier, almost all of the ones identified as functional are used as code-switches in Miami Haitian Creole–English. Myers-Scotton (Reference Myers-Scotton2002, p. 70) has suggested that degree adverbs, very and too, are functional; however, these and others flagged as functional are also attested in my corpus. There are scarcely any adverbs which have been identified as functional that are truly blocked from occurring in Miami Haitian Creole–English code-switching. Some adverbs which have been called functional are even flourishing in my corpus, e.g. still. Of course, one must recall that the overall number of tokens is limited and the singleton tokens for very and too might in fact indicate inaccessibility; however, many of the adverbs identified as lexical also occur at equally low rates.
There are additional reasons to think that adverbs generally act like members of a lexical category. As we have seen, Mixed English adverbs constitute 12% of all adverbs in the Miami corpus. Mixed English verbs, for their part, constitute 4.5% of all verbs. Mixed English bare nominals constitute 6.3% and mixed DPs (i.e., HC D + Eng NP) are 4.8%, forming a total of 11.1% of all DP/NPs (see Hebblethwaite, Reference Hebblethwaite2007). Given that adverbs code-switch at a higher percentage than categories like nouns and verbs, which are widely regarded as lexical, it appears that any functional trait that they may have imposes no restrictions on their availability. The reviewers expressed surprise at this finding since nouns and verbs are typically thought of as the most code-switched categories. Indeed, while French nouns in Brussels Dutch constitute 58.4% (n = 2,329) of all code-switches in Treffers-Daller (Reference Treffers-Daller1994, p. 99), adverbs are nevertheless the third most commonly switched category with 9.7% (n = 388). In Yoon (Reference Yoon1992, p. 438), adverbs amount to 7% of all switched categories. Therefore the findings for adverbs in other corpora are not entirely different from my own.
Beside Klintborg (Reference Klintborg, Ureland and Clarkson1996) and De Rooij (2000), most work on code-switching lacks the systematic and bidirectional analysis of the surrounding language-indexation of mixed tokens. In the 2004 corpus 38% (54/142) and in the 2005 corpus 50.8% (63/124) of mixed English adverbs occurred as lone (xYx) elements, suggesting that English adverbs, as an aggregate class, insert easily into Haitian Creole and that some even become borrowings-in-progress (longitudinal work is needed to measure this). Moreover, English follows (xYy) in 33.8% (48/142) [2004 corpus] and 38.7% (48/124) [2005 corpus] or precedes (yYx) in 28.1% (40/142) [2004] and 10.4% (13/124) [2005] for a combined average of 56% (149/266), which suggests that adverbs are closely linked to language-switching or occur commonly in chunks with other language co-indexed elements, especially discourse markers. In terms of syntactic structure, my examination of adverbs shows that they most commonly target discourse-prominent positions in, first, the left-periphery and, second, a right-adjunction position whereas they are more rare in the IP- or VP-internal positions. This suggests the accessibility of the peripheries outside of argument structure.
Because of the data in the Miami corpora, the data from Brussels in Treffers-Daller (Reference Treffers-Daller1994), and much recent work, we have discarded, as does Myers-Scotton (Reference Myers-Scotton2002), the analysis in which adverbs are strictly system (= functional) morphemes ill-suited to code-switching (Myers-Scotton, Reference Myers-Scotton1993). Eckardt's (Reference Eckardt1998) thematic adverb subcategorial feature in verbs is just one of several lexical explanations that help account for the presence of mixed adverbs in code-switching corpora and make it possible to predict their occurrence in Myers-Scotton's (Reference Myers-Scotton1993) Matrix Language Frame model. Future work should compare adverbs with derivational morphology (e.g. quickly) and those without it (e.g. too, also) across various pairs of languages to discover differences linked to adverbial suffixation.
In the 2004 corpus, only 15 mixed Haitian Creole adverb tokens are found in English sentences, amounting to 1.27% (15/1,176) of all 2004 adverb tokens, which is equivalent to 9.5% (15/157) of all mixed adverb tokens. This is evidence that mixed English adverbs, at 90.5%, are far preferred over mixed Haitian Creole ones. At the same time, the Haitian Creole adverbial constructions that were found in English utterances provide evidence of the potential for code switching in the opposite direction, what I call bidirectionality. The low number of such code-switches shows the limited access of mixed Haitian Creole adverbs in Miami's sociolinguistic and psycholinguistic context, where languages have a ‘high’ or a ‘low’ social status. The asymmetry between the sociolinguistic and the psycholinguistic roles of the languages in Miami accounts for the asymmetry in which it is common to find mixed English adverbs in a Haitian Creole frame but rare to find mixed Haitian Creole ones in an English frame. Reversed sociolinguistic conditions (in another world), would theoretically give Haitian Creole adverbial switching into an English frame the asymmetrical advantage. The result of asymmetry does not discount the potential for bidirectionality, it merely disguises it.
It is not a conflict of grammatical features that causes asymmetry, it is instead the prevailing sociolinguistic and psycholinguistic conditions where English has a ‘high’ status and Haitian Creole has a ‘low’ status. The ‘low’ social status of Miami Haitian Creole gives it a privileged role as the functional Matrix Language whereas the ‘high’ status of English places limits on its availability as a Matrix Language, on the one hand, while making it the ideal reservoir of lexical elements to supply Haitian Creole-framed code-switching, on the other. An approach like MacSwan's (Reference MacSwan1999), where language-indexation is unexamined, sociolinguistic and psycholinguistic conditions are inadmissible, and morphology (the strength of phi-features) is supposed to account for mixed grammar, the issues of bidirectionality, the lop-sidedness of the results (9.5% mixed Haitian Creole adverbs versus 90.5% mixed English adverbs), and the absence of a plain morphological basis for adverb behavior present serious problems.
The sociolinguistic and psycholinguistic situation is reflected in the quantitative findings which show a preference for code-switching with an Haitian Creole functional frame. The few cases of English–Haitian Creole code-switches provide evidence that code-switching is not absolutely banned in the other direction; it is only disfavored in that direction. MacSwan's (Reference MacSwan1999, etc.) minimalist approach is missing an explanation for bidirectional data which occur in the context of asymmetry. Table 11 shows that the ‘high’ or ‘low’ social status of the two languages predicts which one will asymmetrically serve as the Matrix Language.
Treffers-Daller (Reference Treffers-Daller1994) and Hebblethwaite (Reference Hebblethwaite2007, Reference Hebblethwaite2009b) argue that the asymmetry between Matrix Language and Embedded Language morphemes mirrors sociolinguistic and psycholinguistic pressures. English is not banned as a frame since traces of bidirectionality can be detected. These traces of bidirectionality support the idea that results with the opposite asymmetry are theoretically possible under opposite sociolinguistic and psycholinguistic conditions.
Mixed English adverbs in Haitian Creole show a range of borrowings (just, back), borrowings-in-progress (especially, still, maybe, sometimes, etc.), and code-switches (unfortunately). De Rooij (2000), Klintborg (Reference Klintborg, Ureland and Clarkson1996) and Torres (Reference Torres2002) argue that borrowing and switching are not mutually exclusive processes. There is a continuum in which items enter a language as code-switches and may become borrowings-in-progress over time. Some English adverbs gradually become borrowings-in-progress as they increasingly occur in strings surrounded by Haitian Creole material and are decreasingly used as switch-points into a co-indexed English string. In some cases they may even replace Haitian Creole equivalents and become fully-fledged borrowings.
Although Cinque (Reference Cinque1999, Reference Cinque2004), Ernst (Reference Ernst2002), Myers-Scotton (Reference Myers-Scotton2002), Treffers-Daller (Reference Treffers-Daller1994) and Yoon (Reference Yoon1992) suggest that adverbs include lexical and functional properties, the results here only provide evidence of their lexical status in Miami Haitian Creole–English. Supporting arguments for such a view were given from Eckardt (Reference Eckardt1998) and Manninen (Reference Manninen2000). This paper is one of the first to focus exclusively on adverbs in code-switching by means of a quantitative methodology which codes both mixed and unmixed tokens bidirectionally and for their surrounding language-indexation. This method serves as an additional tool for assessing theoretical claims about the nature of the syntactic category of adverbs. Future work might employ cross-sectional data to analyze individual differences in adverb code-switching. Collecting naturalistic and experimental data from a larger sample of participants may also shed light on some of the issues raised here. Additionally, expanding recruitment to include the first and third generations, in addition to the second generation, will help us understand how different groups employ adverbs in code-switching (see Torres, Reference Torres2002). Future work might also employ a variationist approach like the one used by Mougeon and Beniak (Reference Mougeon and Beniak1991) in order to compare differences in Haitian Creole–English code-switching in Miami, Fort Lauderdale, New York or Boston, for example. While the Miami Haitian Creole–English corpora provide little evidence for the applicability of the 4-M Model to adverbs, an expansion of bilingual data might discover evidence that can be better accounted for by dividing adverbs into different morpheme types. Here I attempt to forge into new territory with a comprehensive quantitative methodology but it is clear that further work in other language pairs and by means of other methodologies is needed in order to gain a better cross-linguistic understanding of adverbs.
