Hierarchicalized matrices: codeswitching among urban Nigerian Arabs.
One of the most developed models for explaining bilingual constituents is the matrix language frame model (Myers-Scotton 1997). Its basic insight is that a bilingual constituent will be determined by the abstract structure underlying only one of the languages of the constituent, the matrix language. While the model has been tested against many language situations, a situation typical for minority linguistic groups in Maiduguri in NE Nigeria has not been studied in detail. The present study centers on Nigerian Arabs. Urban Nigerian Arabs are typically bilingual in Nigerian Arabic (NA) and Hausa, and additionally they use English and Standard Arabic, particularly for lexical concepts. Bilingual speech (codeswitching) is supported by both NA and Hausa matrices. In a number of respects, however, both lexical and constituent insertions into these two languages are different. This article documents these differences on the basis of an extensive spoken corpus, and examines their implication for the notion of matrix language. It is argued that the structural notion of matrix language needs to be supplemented with a perspective allowing matrix languages to be conceived of in hierarchical terms.
1. Matrix (1)
One of the important concepts in codeswitching research has been the postulation of a matrix language as a frame for organizing multilingual speech. Its most explicit formulation was developed by Myers-Scotton in her matrix language frame model (henceforth MLF) (Myers-Scotton 1997; Myers-Scotton and Jake 1995, 2000, 2001). One strength and continuing appeal of the model lies in the fact that it adapted monolingual psycholinguistic theories of sentence production (especially Levelt 1989) to bilingual language production so that observed regularities of outcomes in codeswitching can be related to the way bilinguals process their languages in production and comprehension.
Briefly, in this model (Levelt 1989: 9; Myers-Scotton 2002: 24 (2)), a conceptualizer sends a message to a formulator whose function it is to convert a message to a linguistic structure. The formulator calls up lemmas. These are abstract, language-specific entries specifying semantic information and syntactic information (e.g. grammatical class like N, V, predicate-argument structure of a V). In addition, lemmas may have grammatical information such as number added by way of diacritics (Roehlfs 2000: 74). Once activated, a lemma recovers the relevant morphological properties to encode the appropriate word form.
The syntactic organization underlying Levelt's model is one based on the sentence, and Levelt outlines an algorithm by which a sentence is generated (Levelt 1989: 236-246). While detailed formulations of language production may differ, there appears to be broad agreement within psycholinguistics that a unit corresponding to a sentence in linguistic theory serves as a basis for speech planning and execution (see Ferreira 2000: 318-320 for review). The sentence, or sentence-like unit is also the basis of Myers-Scotton's MLF. In her earlier work, Myers-Scotton took the sentence as the basis of bilingual production. More recently she has argued for the complement phrase (CP) as the structural unit in which bilingual speech is ordered (Myers-Scotton 2002; Myers-Scotton and Jake 1995: 982). (3)
Myers-Scotton observed that as a rule in bilingual speech the grammatical structure of a constituent can be exhaustively described using a language production model based on a single language, termed the matrix language, as in the model sketched above. An embedded language (EL) may be inserted within this constituent only to the extent that it follows the grammar of the matrix language (ML). Research has shown that EL elements most frequently are single lexemes, devoid of EL grammatical properties other than a category marking (like N). Larger embedded constituents with EL grammar may also occur, provided they are inserted at a position compatible with the ML.
Much support for the appropriateness of this model has come from empirical studies which have shown that bilingual constituents are as a rule governed by the grammar of a single language (see Myers-Scotton and Jake 2001 for a listing of such studies). Myers-Scotton generalizes this observation in her uniform structure principle, which states:
a given constituent type in any language has a uniform abstract structure and the requirements of well-formedness for this constituent type must be observed whenever the constituent appears. (Myers-Scotton 2002: 120) The notion of matrix language is a valuable analytic concept for an understanding of structural aspects of multilingual speech precisely because it is narrowly constrained in linguistic and psycholinguistic terms. One corollary of this tight definition, however, is that it is not available as a term for characterizing language units outside of the CP domain on which it is based, at least not without explicit emendation. The uniform structure principle pertains to constituents one at a time, not to languages. To see this, consider the following two usages of the term "matrix language." In Myers-Scotton (2000: 32-33) the matrix language is characterized thus: "... the matrix language includes slots for permissible surface-level morphemes, but is not synonymous with a fully-fleshed out linguistic variety."
Describing a case study by Kaufman and Aronoff (1991), Myers-Scotton characterizes the shift from Hebrew to English dominance, as "... a very rapid changeover from Hebrew as the matrix language of her bilingual speech to English as the matrix language" (Myers-Scotton 2002:182). Kaufman and Aronoff outline steps by which a bilingual Hebrew speaker moves from being Hebrew-dominate, to fully bilingual, to English-dominate, concentrating, however, on describing how the structure of her Hebrew changes. At the age of 2;6 she is described as being as fluent in Hebrew as her peers. When she moves to the USA, however, her Hebrew at first is fully maintained in the face of an increasing proficiency in English ("a balanced bilingual" [Kaufman and Aronoff 1991: 180], i.e., both Hebrew and English equally available as CPs) and then deteriorates in various ways. What Kaufman and Aronoff document is a change in overall language proficiency, exemplified in detail in the changing Hebrew verbal morphology of the speaker, not a step by step change in individual CPs.
Given that the term matrix language is based on the grammatical definition of CP, the term "matrix language" is not appropriate here. If matrix language = language defining the structure of a given CP, (4) then substituting "language of a given CP" or "language of a given constituent" for "matrix language" in the above quote does not give a felicitous meaning. In fact, in describing the changeover from Hebrew to English as one of a changeover in matrix languages, it is hard to interpret Myers-Scotton as using the term here in anything other than the sense of fully fleshed-out language.
Were problems relating to the use of the term "matrix language" confined to terminological matters such as this it could be left to the attentive reader to work out the discrepancy for him- or herself. There are two more serious issues involved, however. One of these is prominently associated with multilingual language situations which I term "complex," involving the use of three or more languages. The bulk of this article is dedicated to this problem, and will be introduced via an extensive example in the next section. Very briefly, the problem is as follows. If a population of speakers uses two or more MLs, and if these MLs differ in respect of the EL elements which they support, either quantitatively or qualitatively can the notion of matrix language capture these differences? A second less central problem involving language sequences across CP boundaries is introduced in Section 6.
2. A problem and an example
Nigerian Arabs in Maiduguri, Nigeria, constitute a linguistic minority of about 10% of the total city population (approx. 500,000 people). There are three languages of wider communication, Hausa, Kanuri, and English, which are spoken to a greater or lesser degree by all Maiduguri residents, including the Nigerian Arabs. In addition, Standard Arabic is learned by many as well. More details about the linguistic situation and the corpus used are given below in Sections 3 and 4.
The linguistic problem pertaining to the notion of matrix language can be illustrated on the basis of the following two excerpts from the corpus. They are from the same text, and they center on one speaker, labeled "s." There are two other interlocutors ("q," "a"). The languages appearing in the excerpts are Nigerian Arabic, Hausa (italics), and English (underlined).
(1) (about 33 minutes into conversation)
1q maa darab, shuqula maa darab shuqula hagaara ke bas
2s gal be 1-paailot be d-dayyaara waga
3a waga, ya faadi a ina?
4s an taba faadi ai
5q ke da muxxa, muxxa daayir walla shunu
6s yaa taba faaduwa
7q ina tsammaani ina tsammaani kwakwalwassa ...
8s shii ya sa baasa soo su yi tuka jirgii ai
9? banzan banzaa nee wallahi
10q shi daa paailot ne?
11s yana faailot bees faailot fi naajeerya
12q fi najeeriya
14q min daaka al-kataloo da buguulu laa shunu
1q He isn't [mad]. He isn't mad, it's just that he's rude.
2s It's said the pilot and plane crashed.
3a He crashed? Where did he crash?
4s He crashed once.
5q He's like that. Is he crazy, or what?
6s He crashed once.
7q I think, I think his brain ... ?
8s It's because of that they don't like them to fly planes.
9? He's really a useless person.
10q It's that pilot?
11s He's a pilot. The best pilot in Nigeria.
12q In Nigeria?
14q (Better) than that one the one they killed what's he called?
(2) (about 45 minutes into conversation)
1s maa buwassilan najeeriya ille baad tushuur kaza,
3s xalaas, al-baxiraat deela jan, abiyoola taarif [[fi 1-waqit].sub.1] da ya zoo basu kaamaa shi ba
5s sunaa soo su saamee shi da ezibit da komii zaasu kaamaa shi, shii ne kawai anaa nan anaa nan, sei al baaxiraat jaayaat, [[waqit da].sub.2], umaru dikko yaa yi talifoon daga can yaa ce a cenja sojoojii da suna baakin ruwaa catta qayyiruuum, al-askar al-fi qashim alme da
6a ijiibu aaxar
7s fishaan abiyoola yaa yi diil da su, shii nee sai aka ce to, [[lookaci yaa yi kusa].sub.3] shaba mbaakir bukura, hinna buwassjlan da, xataas jo xammoohum masho sabbo difeens, [[a wajan].sub.4]
1s They won't arrive in Nigeria until after so many months.
3s Okay, the ships came. As for Abiola you know at the time if he'd come they wouldn't have arrested him.
5s (Because) they want him to be caught with the exhibits so they're waiting there (to catch him). So the boats are coming just now, [at this point] Umaru Diko telehophoned from there to transfer the soldiers who are working on the coast all of them should be transferred, the military on the coast.
6a They should bring others.
7s Because Abiola, he made a deal with them. It went like this. [When the time came (for the shipment)] like tomorrow or the day after, when the ships are to arrive they came and gathered them up them and stationed (other) defense forces/n the place.
The first excerpt divides roughly into three parts as far as the MLs go. It begins with two Nigerian Arabic turns, in turns 3-5 switches either between speakers or in turn 3 within speaker between Hausa and Nigerian Arabic, it runs in Hausa from 6-10, and then moves back to Nigerian Arabic from turns 11-14. There are a few lexical insertions as well, all from English. Note that the word "pilot," in two phonetic guises, is inserted both in Nigerian Arabic and Hausa CPs (turns 2, 10, 11, the third token in bees failot is considered part of an EL island). In this first excerpt, Nigerian Arabic and Hausa matrices are essentially used sequentially, and both support insertions from English. All in all, this excerpt adheres well to the parameters which Myers-Scotton (2002:112) terms "classic codes-witching": matrix language shifts occur at CP boundaries and bilingual CPs involve insertions of single words into frames set by the matrix language. In line 2, for instance, the English word paailot is marked as expected by the Nigerian Arabic definite marker l-. However, instead of one language which sets the frame for the CP, as in classic codeswitching, two do: Hausa and Nigerian Arabic.
The second excerpt is characterized by a greater degree of switching, both lexical and constituent. Lexical insertions, again all from English, ("exhibit," "telephone," "deal," "defense") occur in both Hausa and Nigerian Arabic matrices. The constituent insertions as well occur in both Hausa and Nigerian Arabic matrices. These are set of in brackets "[ ]" and numbered consecutively in excerpt (2). For instance excerpt (2) turn 5s has a CP in the middle which can roughly be parsed as in (3),
(3) [[waqit da].sub.COMP], [umaru dikko yaa yi talifoon daga time this Umaru Diko TAM do telephone from can ]IP]CP there 'At this time, Umaru Diko telephoned from there ...'
the Arabic 'at this time,' waqit da is inserted before the subject and Hausa verb 'he telephoned,' yaa yi talifoon. On the other hand, in excerpt (2) 7s "xalaas jo xammoohum masho sabbo difeens, a wajan" the Hausa prepositional phrase a wajan 'in the place' is inserted at the end of the Nigerian Arabic CP. Within each of his three turns speaker "s" switches between Hausa and Nigerian Arabic CPs a number of times, embeds English elements in both Hausa and Nigerian Arabic CPs, and embeds Nigerian Arabic islands in Hausa matrices, and Hausa islands in Nigerian Arabic matrices.
This mode of speaking is quite typical of Arabs living in Maiduguri, Nigeria, and virtually any Arab under the age of 40 living in the city is capable of participating in exchanges of these types, subject to constraints of personal fluency.
The mixed CPs in these excerpts, both with Hausa and Nigerian Arabic MLs, are equally correct according to their respective grammars. Looking at the Hausa constituent insertion a wajan at the very end of turn 7, for example, one could substitute the Nigerian Arabic prepositional fi l-bakaan 'at the place' in the same position (see [4a]). Similarly the Nigerian Arabic constituent insertion waqit da 'at the time' could be substituted for by Hausa a lookaein nan 'at the time' (or equivalently, a lookacin or a lookaci) so that instead of (3), (4b) could also be produced with no disruption of the underlying Hausa structure.
(4) a .... sabbo difeens fi 1-bakaan b. a lookacin nan, umaru dikko yaa yi talifoon daga can
In both cases the embedded island is inserted appropriately according to the ML language. Note that the constituent insertions differ in their sentential placement. The Nigerian Arabic phrases are CP initial, the Hausa CP final. Without breaking any rules of the ML, the insertions nonetheless display different patterns of insertions, in this case in terms of position in the CP, a major point taken up further in Section 6.
The problem, however, is that there is nothing in MLF theory which allows Nigerian Arabic or Hausa to be differentiated in respect, for instance, of where embedded islands occur. It only predicts that insertions into a Hausa or Nigerian Arabic flame will be correct according to the rules of Hausa or Nigerian Arabic. (5)
The MLF does, however, implicitly provide an initial hypothesis about how insertions in the Hausa or Nigerian Arabic CPs should behave, namely that they should behave in the same way. This follows from Myers-Scotton's uniform structure principle quoted above: once a CP is planned, it will be executed using the structure of a single language. EL insertions are "blind" as to which matrix is being used. In the current data, as the two excerpts have illustrated, Nigerian Arabic and Hausa are the CPs of choice.
The current data allows the null hypothesis to be tested, through in an expanded framework of hypotheses. A comparison of the behavior of insertions into Hausa and Nigerian Arabic matrices can have three outcomes.
(5) a. The null hypothesis: insertions into Hausa and Nigerian Arabic matrices cannot be differentiated quantitatively or qualitatively; b. Arbitrary differentiation: insertions into Hausa and Nigerian Arabic behave differently; however, there is no consistent pattern of difference; c. Tendential differentiation: insertions into Hausa and Nigerian Arabic behave differently, and the differences tend in a particular direction.
The present corpus allows these outcomes to be given empirical test. In Sections 5 and 6 I will examine four types of insertions into Nigerian Arabic and Hausa (6) matrices. Two of these are insertions at the lexical level, and two at the syntactic. Before proceeding to the linguistic data I first briefly summarize the social and historical background of the population of speakers used in this study (Section 3) and basic data relating to the corpus used (Section 4).
3. Nigerian Arabs in Maiduguri
The present article describes codeswitching among native speakers of Nigerian Arabic living in Maiduguri, the capital and largest city of Borno in NE Nigeria. Arabs (as they call themselves) constitute about 10% of the population of Borno, the most northeastern state in Nigeria. They are not a new population in the area, having begun settling in the Lake Chad region in the fourteenth century. They probably form the same percentage of the total population of Maiduguri. Locally they are known as "Shuwa," an exonym. Their sizable presence in Maiduguri is about three generations old, and there is no indication that Arabic is being lost among the urban population (see Owens 1998). As noted above, the Arabs are exposed to up to three languages of wider communication (LWC): English (the effective national language of Nigeria), Hausa (the current urban LWC of northern Nigeria), and Kanuri (the older LWC of Maiduguri). Standard Arabic is also important in the community, though it is not an LWC.
Not all speakers will have equal degrees of exposure to each of these languages. English is acquired through formal schooling, as is Standard Arabic. Between these two languages, it is far easier to acquire oral fluency in English than in Standard Arabic, as the latter serves nowhere as a lingua franca, even among religiously orientated Muslims. Rather, Standard Arabic is in the first instance a written medium. However, given the rise of the so-called Islamiyya schools in Nigeria over the past fifteen years, schools' (which espouse, inter alia, Standard Arabic-based teaching) use of this variety has probably increased. All younger Maiduguri Arabs I have met speak Hausa, most with great fluency (see Notes 11 and 13 for concrete exemplification). Kanuri is spoken by many, though in urban areas in northeastern Nigeria it is losing ground as an LWC to Hausa. In rural Borno, Kanuri continues as the principle LWC. While there are many minority languages comparable in size, or smaller than Arabic, these languages are hardly learned by Maiduguri Arabs.
This, in short, is the linguistic world of Maiduguri Arabs. A crucial point is that in one context or another, any of the languages listed in the previous paragraph may be appropriate. On a comparative basis this point is an important one. Unlike western and industrialized societies, language use is dictated only to a small degree by explicit policy, and is channeled only minimally via education. English, being the de facto national language, is probably the language of highest prestige. Access to English is relatively restricted in NE Nigeria, however, as educational institutions do not have the means to promote mass literacy in this language. That official language policy has only a small role to play is witnessed by the ascendancy of Hausa as the main lingua franca in Maiduguri. Up to the second World War, Hausa was largely the language of Hausa traders in Maiduguri. Today, however, it has become the main urban LWC, in Maiduguri and elsewhere in northern Nigeria. Speakers of the old LWC, Kanuri, though still politically dominant in Borno, have been unable to prevent the ascendancy of Hausa (see Bross 2002).
The language situation among the Nigerian Arabs in Maiduguri can be described as complex, in that besides their own native language, Hausa, English, and Standard Arabic are all languages sanctioned in one way or another, while in many circles Kanuri may still be acquired as a second language in Maiduguri. This situation supports a stable multilingualism where, at the present time, no single language of wider communication is so dominant that it impinges on the generational transmission of native Nigerian Arabic within Maiduguri. Furthermore, I would suggest that because in different situations each of the languages named above is more or less appropriate, codeswitching involving any of them will be witnessed regularly, and it will not be limited to a single type (only or predominantly at CP boundaries or only or predominantly bilingual-CP; see excerpts  and  for exemplification).
It can also be noted that the situation illustrated here is not confined to Maiduguri, Nigeria. Rather it appears in similar form in many African urban areas (see Haust 1995: 95; Haust and Dittmar 1998 on Gambian cities; Goyvaerts 1997; Goyvaerts and Zembele 1992 on Bukavu; Swigart 1992 on Dakar; etc.).
4. The corpus
The data is based on a collection of ten texts, comprising just over 84,000 words, from about nine hours of recording. There are 26 active participants in these texts, by which is understood speakers who, in a given text, contribute 250 words or more. An additional eight speakers make only minimal appearances. As already noted, all are native Nigerian Arabic speakers, and most of the 26 either were born in Maiduguri, or moved to the city by the age of twelve.
It is important to have an idea about the frequency of each of the languages in the texts. I use two parameters for assessing the relative frequency of each language. The first is a discourse-based measure which gives a rough idea about the extent to which each language serves as a matrix language in adjacent CPs. In the discourse, at a speaker switch, the next speaker can choose to continue in the same language as the previous one or switch to a new one. Table 1 shows the frequency and percentages of how often the discourse continues in the same language at a speaker switch. For instance, in excerpt (1) the first two turns from 1 q to 2s are in Nigerian Arabic. Before the third turn there is a speaker-internal switch to Hausa, and at the end of his turn from 3a to 4s the conversation continues in Hausa. This is followed by two turns where the language switches, and then come four speakers' turns in which Hausa is maintained across the turn, and so on. The statistics in Table 1 summarize how often the same language is maintained between speaker turns.
Here it can be seen that generally the only two languages which support a discourse across speaker turns are Nigerian Arabic and Hausa.
Table 2 is a simple word count of the ten texts.
Here as well the languages divide into two sets. English and Standard Arabic are frequent enough to be significant lexical donors (of. the single word English insertions in  and  for examples), but not so frequent as to be significant matrix builders. Hausa and Nigerian Arabic, on the other hand, together establish the basis of connected discourse, comprising 95% of all words. In general Nigerian Arabic and Hausa may be termed discourse languages, whereas English and Standard Arabic are lexical donor languages. (7) As a shorthand, I will also refer to English and Standard Arabic as L3s.
The ratio of .57 : .43, of Nigerian Arabic to Hausa words in the corpus is important for interpreting a number of distributional observations in the following sections. I will round off the ratio at 3:2 in favor of Nigerian Arabic. In fact, if only the totals of Hausa and NA are considered (48,724 : 29,668) the ratio between these two languages is very close to 3 : 2.
Taken together, the excerpts in (1) and (2) and Tables 1 and 2 illustrate one key point, namely that the speakers in this corpus have a fluent command of both Nigerian Arabic and Hausa. (8)
5. Lexical insertions in Nigerian Arabic and Hausa matrices
At this point I turn to the description of four types of insertions into Hausa and Nigerian Arabic matrices. Each of these types focuses on a prominent insertional characteristic of a different language in the corpus, so that generalizations formulated at the end of the article encompass all of the languages. In this section I look at two types of lexical insertions, and in Section 6, at two types of embedded islands.
5.1. Degree of lexical insertion
Table 2, along with Table 1, establishes a prima facie statistical basis for a distinction I draw between discourse and lexical languages. In the rest of this section I will concentrate on differences between the two discourse languages, Hausa and Nigerian Arabic. One parameter of comparison concerns the question whether the two languages behave similarly relative to the way in which lexical and embedded constituent insertion takes place in them. It is imaginable, for example, that there is no insertion of any sort in one of the two. In this discussion, a basic proportion which needs to be kept in mind is the 3:2 Nigerian Arabic:Hausa (see above) total word ratio. This gives a rough idea about how much of the corpus is executed in Nigerian Arabic, and how much in Hausa.
Table 3 summarizes the amount of intra-CP lexical insertion which occurs in the two main languages of the texts, Nigerian Arabic and Hausa, insertions like "pilot, exhibit, deal" in excerpts (1) and (2) above. In this count, I leave out language switches which are negligibly represented, for example, the twelve lexical switches in which a Kanuri word is inserted in a Nigerian Arabic matrix. Lexical insertions of all grammatical categories except proper names are given here.
In Table 3 it is evident that Hausa and Nigerian Arabic behave differently in respect of their matrix language status. In particular, lexical insertions from Standard Arabic almost categorically presuppose a Nigerian Arabic matrix. There are 638 Standard Arabic insertions into Nigerian Arabic matrices, only thirteen into Hausa ones. Effectively, it can be said that insertion of a Standard Arabic word requires a Nigerian Arabic matrix, as in (6), taken from the same text as excerpts (1) and (2) (speaker q).
(6) humma me indu-um ittifaaqiya be naas hineeni they not at-their agreement with people here 'They don't have an agreement with the people here.'
The noun ittifaaqiya is derived from a form 8 verbal root. This derived class of verbs does not exist is Nigerian Arabic, but does, of course, in Standard Arabic, from which it is taken.
English lexical insertion, on the other hand, is only slightly below the 3:2 ratio which characterizes the total Nigerian Arabic/Hausa word allotment in the texts, namely 1121:484 = 70:30. In simple quantitative terms, English lexical insertions globally are sensitive to an Arabic or Hausa matrix to a far lesser degree than are Standard Arabic ones. As seen in excerpt (1), for instance, the English "pilot" is inserted equally in Hausa and Nigerian Arabic matrices.
What Table 3 establishes is that Hausa, like Nigerian Arabic, allows frequent lexical switching within its domain. There remains, however, an important difference between the two: whereas Nigerian Arabic is a matrix which allows insertion from any language in the repertoire of multilingual Nigerian Arabs, Hausa supports lexical insertions from English only. This difference can be diagramed as in Figure 1.
[FIGURE 1 OMITTED]
The circles represent the two main matrix languages, Nigerian Arabic and Hausa. Lexical insertions come into Nigerian Arabic from Standard Arabic, English (and Hausa, which I won't discuss here), whereas lexical insertions into Hausa effectively come from English only.
It is not surprising that Standard Arabic insertions into Nigerian Arabic matrices should predominate over those into Hausa. There are, after all, many structural similarities between Nigerian Arabic and Standard Arabic, which favors a strong convergence from Standard Arabic towards Nigerian Arabic (see Owens 2002:197-201 for some specific exemplification). What is less expected, however, is that Standard Arabic insertions in Hausa matrices should be almost nonexistent. This "gap" may be elucidated from three perspectives.
First, the functional "need" for the large number both of the English and Standard Arabic insertions is to allow a rapid discourse about institutions relating to the state and government, education, technology, and so on. Many doublets are found, with basically the same concept being expressed either in an English or a Standard Arabic lexeme. There are two tokens of ittifaaqiya in this corpus, for instance (see, e.g., ), and two of the close English equivalent, agriimen(t) (as is lugo agriiment 'they got an agreement,' see Owens [2000: 308] for doublets based on a larger corpus). The same semantic space is being filled by words from two different languages, English and Standard Arabic. This basic functional need exists in NE Nigeria independently of Nigerian Arabic and Hausa, yet only the English insertions are fairly evenly distributed between Nigerian Arabic and Hausa matrices.
Secondly, the Hausa language itself is not inherently inimical to Standard Arabic. To the contrary, historically the largest number of loanwords in Hausa has came from Standard Arabic (see Greenberg 1947; Baldi 1988; and Haugen 1950; Poplack et al. 1988; Myers-Scotton 1997; Field 2002:8 on the close relation between codeswitching and borrowing).
Thirdly, it might be supposed that Standard Arabic is not inserted into Hausa matrices because little Hausa happened to be used in those texts where competent speakers of Standard Arabic took part (see Note 8). Looking at the statistics in Table 2, however, does not confirm this. The two texts with the highest frequency of Standard Arabic, texts 5 and 7, also have a fairly high percentage of Hausa, and a high absolute amount of Hausa as well. On the other hand, text 10 with the lowest percentage of Hausa has only the fifth highest frequency of Standard Arabic.
There is thus no a priori reason why Standard Arabic should require a Nigerian Arabic matrix. The important point for present purposes is that the statistics in Table 3 indicate that the Nigerian Arabic and Hausa matrices are not equal in respect of lexical insertions from Standard Arabic.
Before moving on to the next topic, referring briefly to the three hypotheses outlined in (5) it is clear that (5a) is not appropriate in the present case: Hausa and Nigerian are not equal in the extent to which they allow lexical insertions in this universe of speakers. Nigerian Arabic is effectively the only matrix which supports insertion from Standard Arabic, and to a small degree it supports insertions from English more than does Hausa. From one example it is not possible to decide the appropriateness of (5b) vs. (5c).
5.2. Morphological complexity of inserted elements: plural contexts
Commensurate with most documented CS corpora, the large majority of the inserted lexical items are nouns. There are 926 English nouns inserted in NA matrices, 457 in Hausa ones. Proper nouns (e.g. names of countries, public institutions) are excluded from the statistic. The roughly 2 : 1 ratio in favor of insertions in Nigerian Arabic matrices observed to hold for all lexical insertions (see Table 3) appears here as well.
Concentrating on these, inserted English nouns can be divided into three classes: forms which correspond to the bare English form, those which carry English inflectional marking, and those which carry either Arabic or Hausa ML morphological marking. Examples (all textual) are found in (7)-(11) below. (7a) gives examples of Nigerian Arabic ML marking, (Tb) of Hausa. Examples of English -s marking are given in (8)-(10) and of [empty set] marking in (11).
What will be compared is the extent to which English noun insertions occur in a matrix with a plural meaning, either marked as a morphological plural or in bare form. One perspective here is that a plural context is marked, formally and semantically, as compared to the singular and that in such a marked context, matrices formed by Nigerian Arabic and Hausa might differ in respect of the insertions they support. This hypothesis is exploratory. Its components are as follows. Potentially contrastive matrices are identified, in this case Nigerian Arabic vs. Hausa plural contexts. Plural contexts are assumed to be marked, following the classic idea of Greenberg (1974). Greenberg examined the effects of markedness mainly in terms of phonology and morphology (e.g. the marked plural context will, crosslinguistically, support fewer morphological distinctions than the unmarked singular; gender distinctions, for instance, will tend to be neutralized in the plural, maintained in the singular). In the present case it will be manifested in the potentially different way insertions from the L3 English are items are sensitive to a Hausa or Nigerian Arabic matrix.
To illustrate the relative formal complexity of plural vs. singular contexts, in Nigerian Arabic a plural noun will require specific plural morphemes (distinguished for M and F) in adjectives, demonstratives, object pronouns, and subject-verb agreement. Singular nouns occur with unmarked forms of adjectives and verbs in subject agreement. In, for example, (11) below, for instance, the subject steet shows agreement with the plural subject via the FPL suffix -an. By contrast, the singular verb form, qassad 'he, it agreed' is the unmarked stem. In Hausa, plural contexts are characterized by plural subject agreement with the verb (TAM marker), and in Maiduguri Hausa, on a variable basis (see Note 11) with demonstratives.
Examples of English noun insertions are as follows.
(7) a. NA plural: talafoon-aat 'telephones' (f. pl.), tiicaa-yiin 'teachers' (m. pl.)
b. Hausa plural: teel-ooli 'tailors'
(8) al-fakt-s deela the-facts these.F 'these facts'
(9) fi n-nomadik skuul-z (9) in DEF-nomadic school-PL 'in the nomadic schools'
As the statistics in Table 4 will presently show, it is common for an English plural noun to maintain its plural form. (10) Nonetheless, that the larger NP matrix in each example is Nigerian Arabic can be seen, for instance, in the definite article al- prefixed to both nouns, the assimilation of this al- to a following alveolar consonant according to Nigerian Arabic phonological rules in (9), and the Nigerian Arabic plural demonstrative deela in (8). A comparable Hausa matrix is,
(10) wani straat-z (11) certain strategy-PL 'certain strategies'
In this corpus, the English plural is used appropriately in all tokens where it appears, namely, to represent plurality. However, in not all contexts where a plural form is expected does a plural form occur. In about 29% of the cases a bare form occurs instead, as in (11) below. Note that plurality is indicated here by the plural agreement on the modifying adjective and on the verb.
(11) steet waad-aat qassad-an state some-FPL agree-FPL 'Some states agreed.'
The statistics in Table 4 below indicate how frequently an English noun occurs inserted in a plural context.
For exemplification, in Nigerian Arabic matrices there are 22 tokens of English nouns with a Nigerian Arabic plural suffix, 78 tokens of nouns with the English plural suffix plural -s, and in 44 tokens a bare noun occurs (as in ). In all, there are 144 Nigerian Arabic plural contexts with an inserted English noun.
There are two important aspects to these statistics. On the one hand, the "accuracy" of the plural suffix is strikingly similar for the Nigerian Arabic and Hausa matrices. The plural suffix -s appears in 54% and 51% of possible contexts in the NA and Hausa matrices, respectively, for instance.
On the other hand, the insertion of an English noun (both explicitly marked for plurality, and wrongly marked in a bare form) is far more likely to occur in a Nigerian Arabic plural context than in a Hausa one. Whether measured by the overall ratio of 144: 37 (NA: Hausa), the number of times an English plural suffix actually appears (78: 19), or the number of times an NA or Hausa ML plural is suffixed (22:8), an English noun is far more likely in an Arabic matrix than in a Hausa one, and the ratio for each category is in the range or 3 : 1 to 4: 1 in favor of the NA context. Importantly, this ratio is well above the 3 : 2 ratio which defines the total ratio of Nigerian Arabic to Hausa words in the texts as a whole. It is also well above the approximately 2:1 ratio which defines the total ratio of Nigerian Arabic to Hausa lexical (Table 3) or noun insertions from English. This difference in favor of the Nigerian Arabic plural matrices is significant at the .000 level (using the figures 926 nouns in Nigerian Arabic matrix, 457 in Hausa context, and 144/37 for the total noun insertions in a Nigerian Arabic/Hausa plural context). (12)
What emerges from this discussion is a result similar to that documented in Section 5.1. English nouns can be inserted into Nigerian Arabic or Hausa plural matrices. Plurality on this English insertion may be either marked overtly with an ML plural suffix, the English -s, or it may be a bare form. Morphological marking is not the issue here (see the end of Section 7 and Note 20). The crucial point is that in the plural context an English noun is inserted and that the plural context is a marked one. The null hypothesis says that Nigerian Arabic and Hausa plural matrices should have English insertions at a rate comparable to their overall occurrence. In fact, in the plural contexts where English noun insertions appear the Nigerian Arabic matrix is chosen overproportionally.
With the two results from the behavior of lexical insertions into Nigerian Arabic and Hausa matrices an initial tendency is observable. In both cases it may be said that the Nigerian Arabic matrices support more types of L3 (shorthand for English and Standard Arabic) insertions than do the Hausa. In respect of languages or language varieties which are inserted into the matrices, a Standard Arabic insertion nearly categorically presupposes a Nigerian Arabic matrix. Given an English noun insertion in a marked plural context, it is much more likely that it will appear in a Nigerian Arabic than in Hausa matrix. (13)
6. Priming and spillovers: inter-sentential switches out of and into embedded elements
As seen in the introduction, structurally-orientated analyses of codeswitching have taken the sentence as the basic analytic unit. Indeed, Muysken (2000) in a recent introduction to codeswitching builds a fundamental tri-partite typology using, inter alia, the parameter of whether a switch occurs mid-sentence (insertions) or at the sentence edge (alternations). Within the CS tradition (e.g. Auer 1988; Sebha and Wootton 1998) the conversation-analytic approach of course operates across sentence boundaries, though geared towards a pragmatic rather than grammatical analysis of the structure of switches.
The CP (or sentence) as the analytic basis of CS has the advantage that it allows CS theory to draw on various traditions of grammatical analysis, both within the "pure" linguistics tradition, and within psycholinguistics.
It would be a mistake, however, to assume that the sentence represents the upper limit to what, in the dynamics of codeswitching, can be investigated in a structurally interesting manner. Turning to the data from the present corpus, two sorts of patterns are notable, which I term primings and spillovers. In contradistinction to the lexical data in Section 5, here one is dealing with embedded islands. They may be represented as follows. (For the terminology "sentence" in this section, see Note 3.)
(12) Priming Spillover H [A] # A H # [H] A A [H] #n A # [A] U # = sentence boundary, [L] = language embedded in left or right part of matrix
The first, priming, says, using the example from the top row, that a Hausa matrix has an embedded Arabic constituent which runs to the end of the CP. The following sentence has an Arabic matrix without EL constituent islands. Expressed differently, a CP ends in an embedded language constituent, and the following sentence continues in this language. The second, spillover, says that a Hausa matrix sentence ends, followed by a further sentence whose ML is Arabic, but which begins with a Hausa embedded constituent. (14) A CP ends and the following sentence switches its ML. However, it begins with an EL island that is the same language as the preceding sentence. Examples of the four situations are as follows. (13) and (14) are examples of priming, (15) and (16) of spillover.
(13) H[A]#A ya-na affektin diin-a [fi ?iyaal-na],# [??]iyaal-na kooliig it-TAM affects me in children-our children-our colleagues hinee-na iyaal of-us children barno ke da, hum bierfu kalaam hawsa Kanuri like they 3-know-PL language Hausa 'It affects me, [in our children]. # Our children, our colleagues, the children of Kanuri, they know Hausa.' (14) A[H]#H minista hanaa-ku indi-na deen [na shinkaafa kaza kaza]# minister of-your.PL have-us debt of rice so so ku-na da sane da shii you-TAM aware of it 'Your minister, we have a debt for rice of so and so much. # You are aware of it.'
(15) A#[A]H humma arab dool bas t-ilgaa-um fi 1-bikaan they arabs these just you-find-them in DEF-place da, # [amma humma doolaka DET but they those dool da] kun san, mutaane sun rigaa su zama those DET you know people they surpass them settle 'Those Arabs, you just find them in this place. # [But those other ones] you know, the people they settled before them.'
(16) H#[H]A K maza-n yau ba hali-n-sa kee nan wai in ji Men-of today PRT condition-of-his it is here yes I hear ba, ai dai # PRT PRT 4 [maza-n yau] di bas haal-him sei Men-of today this just situation-their really K 'Today's man, right, this is their situation (character), they say #' 4 'As for today's husbands, this is simply their condition.'
In the case of priming, the embedded language at the end of [S.sub.1] presages, or primes (in the psycholinguistic terminology), the shift to the following matrix language of [S.sub.2]. In (14) for example, the first CP is Nigerian Arabic, but at the end of it there is a Hausa embedded constituent consisting of the prepositional phrase na shinkaafa kaza kaza. This ends the first CP, and the language of the next sentence continues completely in Hausa. The Hausa embedded island can be said to prime the continued use of Hausa in the next sentence.
In spillover, the matrix language of [S.sub.1] continues on in the adjacent [S.sub.2], but as an embedded island, not as the matrix language. The language of [S.sub.1] can be said to spill over into [S.sub.2] in the form of an adjacent, embedded constituent. In (15), for instance, the first sentence is a Nigerian Arabic CP. The following sentence has a Hausa matrix CP, but it begins with an embedded Arabic topic, set off in brackets. In the Hausa matrix, the object pronoun su cross-references the topic. (3) above is another example of this type of spillover, assuming the EL Nigerian Arabic adverbial phrase at the beginning of the second sentence is in the COMP of the Hausa [S.sub.2].
Primings and spillover go in both directions in the data. Whereas in (15) the Hausa constituent at the end of the CP primes a following Hausa sentence, in (14) the embedded Nigerian Arabic prepositional phrase fi iyaalna can be said to prime the following Nigerian Arabic sentence. Similarly, (16) is exactly the opposite of (15): the first CP is in Hausa, while the following one begins with an embedded Hausa topic before moving on to the Arabic matrix.
The point which I will illustrate here is that priming and spillover, though attested in both Hausa and Arabic matrices, have differential weightings. The discussion in this section is based on the following statistics.
Beginning with a brief summary of the statistics, Table 5a summarizes all contexts of switching. The asymmetric distributions show, as it were, that there is a case to answer for. Switches across sentence boundaries either out of (priming) or into (spillover) embedded islands from Arabic to Hausa or the reverse are not randomly distributed.
Referring to the schemata in (12), while all four patterns are attested in the data (see - above), for each type of switching, two of the tendencies are significant. For priming, the following pattern is slightly dominant: A [H] # H. An embedded Hausa constituent at the end of a CP, more than an embedded Arabic constituent, favors a switch in the next sentence to a Hausa matrix. For spillover the following is heavily dominant: A # [A] H. An Arabic matrix, to a far greater degree than a Hausa one, will favor a continuation of the same language into the following matrix, as an initial embedded constituent.
The statistics in Table 5a hide an important distinction, namely, whether the priming/spillover takes place within speakers or across speaker boundaries. The numbers in Table 5b are all intra-speaker switches. One and the same speaker has a sentence boundary, and he or she continues on into the next sentence. Another situation is where the sentence switch coincides with a speaker switch. (16) above illustrates the latter case. The speaker labeled 'K' completes her turn completely in Hausa. Speaker 4 continues the conversation, beginning with an embedded Hausa topic in a larger Arabic CP.
Accounting for this additional variable, within turn (or within speaker) vs. between turn switching, the differences between priming and spillovers become more pronounced. The within-speaker data is found in Table 5b, where the same trends are found as in Table 5a, except that for the within-speaker switching the correlations of priming and spillover with the respective switch directions are more accentuated. Looking at between-speaker switching (Table 5c) the correlation is somewhat lower. The differences thus hold to a greater degree where one and the same speaker is continuing the next sentence (= Table 5b).
I will leave a broad generalization based on the observations in this section until Section 6.2. Keeping close to a descriptive summary for the moment, it has been seen again in this section that the two discourse languages, Hausa and Arabic, behave differently in respect of the bilingual structures they support. Whereas in Section 5 they differed in the types of L3 lexical material they support, this time the difference is a syntactic one in terms of the sequencing of language switches across sentence boundaries. In the next two sections I interpret these sequencing patterns, beginning with priming.
Priming describes the configuration: Lx [Ly] # Ly. The switch to an embedded [Ly] occurs among a disparate set of post-predicate elements. In the current data the [Ly] constituent is realized, inter alia, as a prepositional phrase, possessor phrase embedded in a noun phrase, nominal object of verb, noun clause object of a verb, and various time, place, manner, and modal adverbs. There seems to be no obvious, discrete, discourse-based functional explanation for the switch. (15) However, while the motivation for the switches to the EL catalogued under priming are unclear and probably of heterogeneous origin, I will suggest that the choice of language across the sentence boundary is not arbitrary.
In what is here termed priming, a switch at the end of a Nigerian Arabic CP to an embedded Hausa island will somewhat favor a continuation of a Hausa matrix in [S.sub.2]. I would like to relate this pattern to a recent psycholinguistic study on L1/L2 sequencing. Meuter and Allport (1999) (also Meuter et al. 2002) have argued that language production in a weaker L2 requires active suppression of L1.
Meuter and Allport studied bilingual reactions to numerals on a computer screen. The numbers were projected in different colors, each color eliciting a response in a certain language. A red "9," for example, required a response in French "neuf," whereas a blue "9" required a response in English "nine." The numbers were projected in various sequences, so respondents would have to continue in the same language as the previous response, or switch to a different one. Various languages and various sequences of numbers were tested, always with bilinguals with whom, it is implied, L2 was not as fluent as L1. (16) Using a common psycholinguistic criterion for distinguishing the effects of responding in L1 or L2, response or reaction times (RTs) for four sequences of answers were analyzed:
(17) a. L1 to L1 (e.g. blue "9" followed by blue "6," where blue represents color of respondents native language) b. L2 to L2 c. L1 to L2 (e.g. blue "9" followed by red "6") d. L2 to L1
L1 to L1 response times were faster than L2 to L2. What interested the authors, however, were the response times between native and non-native languages. Here they found (in three separate runs of the experiment) that respondents were consistently faster in switching from L1 to L2 than from L2 to L1 (Meuter and Allport 1999: 33). They explain this asymmetry in terms of language competence, though using the concepts (metaphors?) of suppression and potentiation:
... for language production in a weaker L2, active suppression of the competitor language (L1) is needed, as well as potentiation of L2. On a subsequent switch trial, this L2 language set thus generates powerful interference with the intended response. For production in L1, in contrast, little of any competitor language(s) may be needed. Hence the reverse asymmetry in the language switching costs. (Meuter and Allport 1999: 33)
In short, it is easy to switch into L2, but once in L2, relatively more difficult to switch out of it, back into L1. Once in an L2 mode, one will tend to stay there. It is this conclusion which I would like to relate to one of the two findings in Section 6. Admittedly, it is a long step extrapolating from a constructed psycholinguistic experiment based on computer responses to figures on a computer screen, to naturally-occurring data. Moreover, to draw the parallel, the two discourse languages in the current data, Nigerian Arabic and Hausa, have to be stipulated as L1 = Nigerian Arabic, L2 = Hausa. Note that to this point no appeal has been made to the concept of "native language."
With all these caveats, there is, however, an interesting parallel in what I termed the priming effect in Table 5. This describes the sequence: Lx [Ly] # Ly.
There is a switch from a matrix to an embedded element at the end of a CP, and the next sentence continues in the language of the embedded element. This pattern occurs to a relatively greater degree when Ly = Hausa than when Ly = Arabic (see particularly Table 5b). Drawing on Meuter and Allport's experiment I would suggest the following explanation for this pattern: once into the less accessible language, Hausa, there will be a tendency to stay in that language, in a cross-sentence context to continue the next sentence in Hausa, due to the cost of switching back into Arabic from L2. If experimental data indicates a greater effort needed in switching from L2 to L1, the data in Table 5 can be read as an avoidance strategy, the avoidance of the effort needed to get back into L1. The switch to the embedded Hausa constituent in S1 primes the continuation of Hausa in S2.
The spillover pattern can be related to a tendency observed by a number of authors working with bilingual corpora (Hasselmo 1975: 259; Nishimura 1989: 372, 1995; Nortier 1990: 135; Swigart 1992: 350; Myers Scotton et al. 1996: 27; Boumans 1998: 380), namely, that switches between topic-predicate or subject-predicate constituents will tend to occur in the order L1-L2 (L1 topic/subject, L2 predicate). Boumans relates this to the distribution of old and new information in the clause, whereby old information is encoded in what Boumans terms the community (minority) language, the new in the superimposed (L2). This sequencing also has a psycholinguistic correlate: topics tend to be high in conceptual accessibility (Levelt 1989: 261), and in general, L1 will be more accessible than L2 (see below). In short, CP-initial EL embeddings occur in a prominent position important for organizing the flow of discourse.
The statistics in Table 5b show that the spillover pattern encodes precisely this observed L1-L2 tendency: if there is a switch between topic/subject and predicate, (17) then it will tend to occur in the sequence [Nigerian Arabic]--Hausa.
I would like to highlight two aspects of this pattern, one psycholinguistic, one sociolinguistic. First, relating the spillover pattern to models of language production, it is assumed that in connected discourse, planning for S2 begins before S1 has come to an end (Levelt 1989: 258). It has also been suggested that the planning does not extend beyond the next clause. The current data shows that the spillover works strongly in favor of Nigerian Arabic. This can be related to the assumption that Nigerian Arabic is more accessible than is Hausa. Speakers will tend to continue from S1 to an embedded NA element in S2 because, as they are simultaneously producing a sentence and planning for the subsequent one, it is easiest to plan while speaking in the more accessible language.
The question of ease/difficulty in switching into and out of languages introduced in Section 6.1 could, moreover, play a role in explaining the distribution of the spillover effect as well. Nigerian Arabic speakers of Hausa are slightly constrained in their ability to move out of [H]. The marked underrepresentation of the configuration, H # [H] A can be explained by the fact that it is easier to switch at the sentence boundary, rather than carry the Hausa over into the initial position of an Arabic matrix, where, should a switch occur, it will need to be effected in a discourse-prominent position CP-internally. Moving out of Nigerian Arabic into Hausa, A # [A] H, on the other hand, should not be difficult, and hence the far larger frequency of the latter pattern.
A second factor is at this point speculative, though at least commensurate with the observed patterns of switching. I would suggest that a general effect of using L2 in the predicate position is to bestow a greater saliency to the message which it delivers. This, of course, is a well-known perspective in codeswitching (Gumperz 1982), one which, in the simple manner formulated here, provides, at best, a starting point for a more precise enquiry into the distribution of codeswitched elements into old and new information packages, and the uses to which they are put.
In short, the spillover effect documented in Table 5 can be seen as a function of at least two factors. First, the embedded Nigerian Arabic initial element is an unmarked continuation in the more accessible language. It facilitates, should the discourse necessity arise, the planning of a sentence in which there occurs a subsequent shift to Hausa in the predicate. Secondly, the direction of shift from Nigerian Arabic to Hausa is favored by the overall social status of Hausa as the language of wider communication.
This article has compared the behavior of Nigerian Arabic and Hausa among a group of Nigerian Arabic multilinguals, with respect to the types of codeswitching insertions each language supports. Four differences were discussed, which are summarized in Tables 6a and 6b.
The four insertional phenomena summarized here are not all from closely-related grammatical domains, so a key question is whether and how they can be attributed to a single factor. To begin with, I suggested above in Section 1 that appeal to the notion of matrix language alone is not sufficient. This is because the four phenomena discussed here are distinguished in Nigerian Arabic and Hausa matrices by degree alone. For the present set of speakers, both Nigerian Arabic and Hausa can be invoked to build MLs. Comparing any two well-formed Nigerian Arabic and Hausa CPs, one cannot be said to be more a matrix than the other. Given the CP-based definition of matrix it is, barring reformulation or extension of the concept of matrix language, not possible to distinguish between Hausa and Nigerian Arabic in this respect.
This assertion might appear odd in the light of Myers-Scotton's recent work on language attrition and convergence (2002). In this work convergence is an important concept, defined in terms of increasing similarity in the abstract representation of two languages at the semantico-conceptual, predicate argument, and/or morphological levels. However, there are three important differences between this work and the present corpus.
First, there are interactions not only between Nigerian Arabic and Hausa, but between these two languages and the other languages found in the community. These might exhibit convergences of certain sorts between Hausa and Nigerian Arabic (see below), but they also result in significant differences. For instance, Standard Arabic is not a variety generally available when Hausa is active.
Secondly, in Section 6 it was argued that there are cross-sentence differences relating to the sequencing of languages. The notion of matrix language, however, does not look beyond the CP. As currently formulated, it does not account for such statistically-discernible dependencies.
Thirdly, the general sociolinguistic background to the Nigerian Arabic speakers in Maiduguri is one of maintenance of their Arabic in the face of multilingualism in other dominant languages of the community. Hausa, English, and Standard Arabic are representatives of these in the present corpus. This is not a situation Myers-Scotton deals with in detail, and her comments on the linguistic effects of stable bilingualism (Myers-Scotton 2002:112), based on Dutch-Turkish bilinguals in Holland, do not adequately reflect the linguistic situation in Maiduguri. Myers-Scotton notes that in Dutch-Turkish CS in Holland switching between languages tends to be between monolingual CPs. While this sort of switching does occur in the present corpus, as exemplified in excerpts (1) and (2), as seen here there are also a large number of bilingual CPs. (18)
I should emphasize that these observations are not to be taken as a criticism of the core notion of matrix language, but rather to indicate that there are multilingual phenomena which lie outside the purview of the matrix language concept as presently formulated. I also do not think it necessary to adjust the notion of matrix language to accommodate such observations. Matrix language is a well-defined concept and useful precisely because it offers a fixed point of orientation, both for phenomena explicable within its borders (e.g. in terms of the uniform structure principle), and for recognizing phenomena lying outside its domain, such as that described here contrasting how Hausa vs. Nigerian Arabic matrices incorporate other-language insertions.
With this in mind, I will turn to a characterization of the differences summarized in Table 5. In fact, the four grammatical phenomena discussed here are taken from rather diverse multilingual phenomena, and I do not think it possible to subsume all under a single, tight explanatory category. More modestly, I will argue that the phenomena can be understand as emanating from a single source, only some of whose elements are well-defined at this time.
The data in this article compare Nigerian Arabic and Hausa as matrix languages among the same population of speakers. It examines two lexical and two syntactic phenomena, and looks at two further languages, English and Standard Arabic, which serve as lexical sources. Taking the two matrix languages as the basis, the phenomena described here can be represented in Figure 2.
[FIGURE 2 OMITTED]
Each circle represents one of the matrix languages. The top half of the circle is for lexical insertions, the bottom for syntactic embeddings, in this article discussed in terms of primings and spillovers. The boxes represent insertions into the respective MLs. Summing over the total insertions in the corpus in individual matrices, the size of the boxes indicates the degree to which a given matrix supports the relevant phenomena.
The four phenomena are parallel in each matrix language. The two types of lexical insertions are identical as far as content goes, insertions of English nouns in plural contexts or insertions of Standard Arabic lexemes. As they are unitary phenomena their behavior in the two matrix languages can be directly compared. Nigerian Arabic, as seen in Section 5, supports lexical insertions to a significantly larger degree than does the Hausa matrix, hence the larger squares.
The two syntactic embeddings are the mirror image of each other: Hausa is embedded in a Nigerian Arabic matrix and Nigerian Arabic in the Hausa. Again the size of the squares should represent the degree to which the given phenomenon is supported by each matrix language. However, note that in the case of the syntactic embeddings not only size of the box (for relative frequency), but also their location (embedded in which ML) and the nature of the phenomenon they represent (priming vs. spillover) are relevant to their interpretation. Priming and spillover are distinguished by the direction of the arrows. An arrow pointing from an embedded language box to the larger matrix circle is priming (embedded element Ly primes continuation in following matrix Ly) while an arrow pointing from the matrix circle to the embedded box is spillover (language of preceding ML spills over as an embedded constituent in following ML).
I will begin with the spillovers (Section 6.2) and come to the primings (Section 6.1) after making a first generalization. Spillovers occur to a far greater degree in Hausa matrices than in Nigerian Arabic. More importantly, these spillover island insertions are CP-initial, and in the majority of cases are discourse prominent topics or subjects. An observational generalization tying this to the lexical insertions is that it is Nigerian Arabic, which supports more types of lexical insertions from L3s (from Standard Arabic, in plural contexts), and also the language more easily inserted as an EL island in conceptually prominent position in Hausa L2 matrices. From these three cases it may be said that in some sense the L1, Nigerian Arabic, has the wider distribution, both in terms of the types of insertions which it supports (e.g. Standard Arabic) and of the contexts where insertions occur (in a more marked plural context, embedded in information-prominent CP-initial position).
Turning now to priming, it might appear that this phenomenon is the odd man out, because instances of priming (sequences of Lx [Ly] # Ly) are fairly evenly distributed in the Nigerian Arabic and Hausa matrices, though Hausa is slightly dominant in the speaker-internal statistics, and compared to spillover, proportionally strongly dominant (see Table 5b). In contrast to spillover, however, which has a specific functional profile (see Section 6.2), the reasons for the switch to the embedded language in the case of priming are not obvious. For some reason, a switch occurs at the end of a CP to the embedded language. However, given this switch, it was argued that if it is a switch to Hausa, there is a slight impetus to continue in the next CP in Hausa. The basic idea is that there is a processing cost involved, once in the L2 Hausa mode, for speakers to switch out of L2 and back into the L1 Nigerian Arabic.
There are, therefore, general psycholinguistic "explanations" which can be associated with distributions of switching summarized in Figure 2 and can all be related to the supposition that Nigerian Arabic is more accessible than Hausa. In the light of the many studies showing the processing of L1 vs. L2 (e.g. Snodgrass 1993; Kroll and Stewart 1994; von Studnitz and Green 1997; using reaction time as the diagnostic) to be more efficient and more accessible, the difference in regards to the three phenomena discussed so far is not surprising. What is of special interest, however, is that Hausa and Nigerian Arabic are not being compared, as it were, head to head in two of the four factors, but rather in the differential way in which L3s are inserted into these two languages.
I would emphasize that the criteria used to differentiate Nigerian Arabic from Hausa are to be evaluated on a case-by-case basis. For the current linguistic analysis, the differences between the two languages for this population of speakers derives from the differences in the phenomena compared, not from a predetermined concept such as native language, even if, in order to invoke analogies with psycholinguistic studies, L1 and L2 had to be stipulated. Clearly in this data set what one would intuitively call the native language (Nigerian Arabic) is the more accessible one. However, given the high degree of multilingualism in Maiduguri, it would not be surprising to find either in the contemporary youngest generation of Nigerian Arabic speakers, or the next, many multilingual speakers who will speak Arabic and identify themselves as ethnic Arabs, yet by operational criteria such as the current ones will have Hausa as a more accessible language.
Moreover, there is one facet of the present data in which Nigerian Arabic and Hausa are not differentiated. In Section 5.2 it was noted that the relative (as opposed to absolute) frequency of occurrence of the English plural -s was the same in Nigerian Arabic and Hausa matrices. In both matrices about 66% of the English insertions in plural contexts were marked by the plural -s. I believe this runs counter to what an access-based hypothesis would suppose. Briefly, accessing of inflectionally complex items has usually been shown experimentally to be slower than less complex items (e.g. Laine et al. 1999; Bybee 1995: 231; Levelt et al. 1999: 13; Portin and Laine 2001). Assuming Nigerian Arabic to be more accessible than is Hausa, it could be hypothesized that a more complex morphological insertion, such as English N + s, would be harder to effect in the L2 (Hausa) than the L1 (Nigerian Arabic) matrix. (19) This is proportionally not the case in the present data. (20) In other words, there is no automatic rule which says that insertional behavior in Nigerian Arabic and Hausa matrices will differ even if it often, if not as a rule, does. Empirical research is necessary.
This article has documented patterns of codeswitching among a minority linguistic group in a complex linguistic environment. The main theoretical construct used for this description is the matrix language frame model. This model, however, uses a CP or sentence-based analytic category, and therefore, does not readily elucidate comparative material such as presented in this article. It is impossible to differentiate the matrix status of Nigerian Arabic and Hausa matrices if they are all "correct" according to the rules of Hausa or Nigerian Arabic. Differences are present however, but of a quantitative nature. Since, as argued here, the differences tend in a consistent direction, a hierarchy of matrix languages can be extrapolated.
I would suggest that it is useful to have an independent descriptive term to describe a situation where the same population of speakers uses two (or more) languages as the basis of matrices and where further languages may be incorporated into these matrices on a lexically-determined basis.
It would be confusing to speak of one language as being "more matrix" than another (besides being inelegant English), and as seen in Section 1, using the term "matrix" other than as an entity defined in structural grammatical terms runs the risk of introducing terminological ambiguity. As an alternative, one could use the psycholinguistic term "accessibility" to describe the gradation, with the adjective form allowing one to speak of one language being more accessible than another. L3 lexical insertion and spillover reflect the greater degree of accessibility of Nigerian Arabic, and, as argued in Section 6.2, the higher degree of priming for Hausa reflects a lower degree of accessibility.
There is a general problem with this, however, not dissimilar to using the term "matrix." Accessibility as a technical term is defined within psycholinguistic theory according to the practices and methodologies of this subdiscipline. While the linguistic phenomenon described here has in a number of places been interpreted in terms of language processing, the relation to it is metaphorical, as no experimental verification is offered. Furthermore, "accessibility" falls within an on-line processing model of speech production, whereas in linguistic theory it is customary to have a vocabulary referring to static, structural entities. As an alternative, independent metaphor, I would suggest that the language domains in Figures 1 and 2 can be thought of as nodes tying together the multilingual's language behavior. The nodes of each language may differ in the way and degree to which they integrate linguistic phenomena. Comparing these, depending on the phenomenon involved, one language may be said to be more nodal, comparably nodal, etc., with another. According to the four measures described here, Nigerian Arabic is more nodal than Hausa.
(1.) This research has been supported by the German Research Council (DFG) under the auspices of two research grants, "Hausa-Arabic Codeswitching in Maiduguri" and the SFB-FK 560, project A1, "Language Vitality in West African Cities." I would like to thank Michael Bross, Jidda Hassan, and Mohammad Munkaila for providing extra data and for checking on the Nigerian Arabic and Hausa translations. I would like to thank two reviewers for critical comments on an earlier version of this paper. Correspondence address: The Center for Advanced Study of Language, University of Maryland, PO Box 25, College Park, Maryland, 20742, USA. E-mail: email@example.com.
Abbreviations are as follows: CP "complement phrase," CS "codeswitching," H "Hausa," LWC "language of wider communication," ML "matrix language," MLF "Matrix Language Frame," NA "Nigerian Arabic," PP "prepositional phrase," PRT "particle," SA "Standard Arabic," Lx [Ly] "language y embedded in language x," "," in linguistic examples = pause. In the examples and translations, Nigerian Arabic is in normal font, Hausa in italics, and English (in one case Standard Arabic) is underlined.
(2.) Levelt and Myers-Scotton appear to differ in the role of the formulator. For Levelt, the formulator mediates between the conceptual and lemma level, whereas in Myers-Scotton lemmas send directions to the formulator, which is responsible for the various aspects of final message linearization. I have not found detailed discussion of the differences, and they do not impinge on the present research. The present paragraph summarizes Levelt 1989.
(3.) In practice authors often define their structural units in formal terms (like "CP"), but in their descriptions will use more commonplace terminology, should it differ from the technical designations (e.g. "sentence," see, e.g., Ferreira 2000).
As the present article is cast broadly in the Matrix Language Frame model tradition, I try to adhere to Myers-Scotton grammatical terminology, in particular using the construct "CP." In most instances this is unproblematic. There is, however, a theoretical issue in the use of the term CP itself. The construct CP is neutral as to whether it pertains to what traditionally are known as dependent sentences, or independent ones. Myers-Scotton's (1998: 92) term "inter-CP" switching does not adequately disambiguate the situation, since this term would describe both a language switch between two independent sentences and a switch between main clause and a dependent adverbial clause, for instance. A terminology is needed to distinguish the two cases. One might speak of a maximal CP, for instance, in the case of independent sentence. Not wanting to delve into an issue of theoretical syntax, I use the following pragmatic terminology. CP is used for the unit defined as the building block of the MLF model. "Sentence" is used for sequences of independent (maximal) CPs. Such sequences are the main focus of Section 6.
(4.) In more recent work, Myers-Scotton would probably talk in terms of "abstract structure underlying the language of a given CP," rather than "language" as such. I am using "language" as a shorthand to this more abstract characterization.
(5.) The problem introduced in this article can be subsumed under a more general perspective, which can be termed "differential structure of embedded insertions": the character of EL insertions is dependent on an independent grouping variable. Most frequently the grouping variable has been identified variously as one of education, or degree of language competence. For example, Myers-Scotton (2000: 31; Finlayson et al. 1998; see also Haust 1995:277 280; Budzhak-Jones and Poplack 1997; Bentahila and Davies 1998) determined that different speakers of Zulu and Sotho have different patterns of insertions correlating with education: the more educated have a higher degree of embedded EL insertions, and the less educated more lexical insertion. The Zulu and Sotho matrices are equally "correct" for both classes of speakers, but the resulting individual matrices (bilingual CPs) are different.
In the present article the grouping variable is the identity of the two languages, Nigerian Arabic vs. Hausa and the form of embeddings in them.
(6.) The nature of the L2 Hausa of Maiduguri Arabs is a problem requiring independent research. One point to be kept in mind is that any assessment of their Hausa requires as a baseline comparison norms for L1 Hausa in Maiduguri, not Standard Hausa, which may be different from L1 Maiduguri Hausa (see Note 11 for one exemplification of this point, see also Bross 2001).
(7.) As is implicit in the statistics in Table 1, where there are 31 instances of English being maintained as the ML across speaker turns, not all of the words listed in the English column in Table 2 are lexical insertions. In text 9, for instance, there are a total of 490 English words. Of these, 153 tokens are in one, two, three or four word lexical switches In Note 14 is described what is meant by a multiword lexical switch. Of these, 118 are single-word English, 32 are two-word, 2 are three-word, and 1 is four-word (prarneri skuul maanymen boot 'primary school management board' classified as a compound lexical insertion, not as an embedded island). This amounts to a total of 192 words which are considered English lexical insertions in text 9. The remaining 308 English words in this text occur mainly in embedded English EL islands, such as lee inkomin siviliyan administre[??][??]n 'for the incoming civilian administration' and laas yia nee 'it was last year,' where the underlined words are considered embedded phrases (the former as object of a Nigerian Arabic preposition, the latter as predicate to Hausa nee).
(8.) Defining degrees of fluency in any of the languages for the individual speakers is a task well beyond the limits of the present study. The following variables would need to be dealt with systematically. First, there is the practical problem of testing for degree of fluency. Written literacy tests are problematic since the speakers are not used to them, and they would have to be administered on a one-on-one basis. This would involve tracking down many conversation participants who were recruited on a voluntary basis, and may have moved away or are otherwise difficult to find.
Secondly, there is the question of what variety should be tested. Standard Hausa is not the same even as L1 Hausa of Maiduguri (of native Hausa speakers, see Notes 6, 11, and 13). Similarly, there are no studies describing the English of Maiduguri, though it may be assumed that the English used in the texts reflects local norms to some degree. Strictly speaking, before representative language tests could even be written for the various languages and repertoires in this corpus, individual studies of Hausa, English, and Standard Arabic in Maiduguri would need to be made. Thirdly, many codeswitching studies (e.g. Poplack and Meechan 1998: 128) relativize the efficacy of non-corpus-based methods in ascertaining the structure of bilingual language behavior. Many CS studies are based exclusively on CS corpora, and this paper follows this tradition. One reason a sample of texts is used is precisely to even out influence from single speakers. That this method can be effective in capturing a community-wide practice I think is evident in the distribution of Standard Arabic insertions. As can be seen in Table 3 below, some speakers make frequent use of Standard Arabic insertions (texts 3, 5, 7, 9-11 to a degree) while others have almost none at all. The participants in texts 2, 4, and 6, where little Standard Arabic is used, in fact are those who have never studied Standard Arabic formally or even memorized the Classical Arabic of the Koran extensively. The participants in the other texts, on the other hand, have studied Standard Arabic up to secondary school, some even at the university level. The point of the community-wide norm is that all speakers who insert Standard Arabic extensively do so in the same manner, namely in a matrix formed by Nigerian Arabic only. Relating this point to speaker competence, it may be said that if the speakers have a certain knowledge of Standard Arabic, they will use it in a uniform manner, namely, by filling out matrices formed by Nigerian Arabic, and not by filling out matrices formed by Hausa.
(9) The collocation nomadik skuulz is coded as a compound noun not as an adjective + noun (in which case it would count as an embedded island rather than as a single lexical insertion). This is justified on three grounds. First, the collocation represents a specific semantic concept. A nomadic school is not simply a "school for nomads." Rather, it is an entity called into existence by the Nigerian ministry of education in the 1980s to cater for the educational needs of nomads. It implies a specific history and administrative structure. Secondly, the entire collocation is marked by the definite article al-, in the same way a single noun is marked for definiteness in a monolingual Nigerian Arabic NP. Thirdly, the word "nomadic" occurs only in fixed collocations, and does not function as a singly-occurring adjective.
(10.) In many CS corpora EL plurals are rare or nonexistent, and in fact a strict reading of the MLF predicts that plural morphemes, being (early indirectly elected) system morphemes, are supplied by the ML. However, there are corpora where EL plurals are statistically well-represented. Backus (2003: 94) in a Dutch-Turkish corpus reports that 11 of 27 Dutch EL plural nouns (type count) are marked by a Dutch plural (the remainder with an ML Turkish plural). Haust (1995: 129, 131) in English-Mandinka-Wolof switching also reports nearly 40 tokens of English EL plurals in -s, 31 with double plural marking from Mandinka or Wolof.
In passing, it can be noted that in contrast to the Gambia data there are no examples of double plural marking (e.g. an English plural -s followed by an Arabic plural suffix like -iin).
(11.) In Standard Hausa a plural noun would occur with a plural demonstrative, wasu straatijiiz 'certain strategies.' As pointed out in Note 6 above, Maiduguri Hausa deviates in may respects from Standard Hausa. In Maiduguri Hausa, for instance, plural marking is sometimes neutralized in favor of a singular form.
To make this point concrete, I can cite data from my colleague Michael Bross. He has collected an extensive corpus of L1 Maiduguri Hausa and L2 Hausa of Maiduguri Arabs. The corpus consists of about 200,000 words, divided 60%: 40% between Hausa L1 and Hausa L2 speakers. Of the 19 Arab Hausa L2 speakers, 8 also are part of the present CS corpus. He reports that the tokens for "certain" in plural contexts are as follows: wasu/wadansu "certain PL" + plural noun: L1 Hausa 28/Nigerian Arab L2 Hausa 22; wani "certain SG" + PL noun (i.e. number in morpheme for "certain" neutralized) 15/10. The number distinction is neutralized in one third of the tokens (25/75), and there is, proportional to the total number of words in the corpus, little difference between the Hausa L1 and Nigerian Arab Hausa L2 speakers in respect of neutralization. In fact, the Nigerian Arabs use the plural form wasu to a slightly higher degree.
(12.) It can also be noted in passing that the overproportional presence of English insertions in plural contexts generally holds not only for the totals based on all texts, as in Table 4, but also in terms of texts and individual speakers in the texts. In the text from which excerpts (1) and (2) are taken, for instance, the ratio of English lexical insertions in plural NA/Hausa contexts is 16:7. The total number of English noun insertions in all contexts (singular/plural, etc.) is 107:82. In this text, 15% of the English noun insertions are in plural Nigerian Arabic contexts, vs. only 8.5% of the Hausa.
(13.) There is one variable which was not controlled for. It could be that overall in the texts, Hausa plural contexts are relatively rare and that the English insertions in Hausa matrices are tracking this overall pattern. I have not coded monolingual CPs grammatically so this cannot be controlled for quantitatively at the moment. It may be remarked, however, that in Michael Bross' monolingual corpora Nigerian Arabic L2 Hausa actually has a slightly higher percentage of overt plural marking on Hausa nouns than does Maiduguri L1 Hausa. The figures are as follows, bearing in mind that in this corpus 60% of the texts (word count) are by Hausa L1 speakers, 40% Nigerian Arab Hausa L2. The total number of common nouns used by Hausa L1 speakers is 21,501, by Nigerian Arab Hausa L2 speakers 13,087. This divides as 62%: 38%. The total number of overtly-marked plurals is 6,184 for the Hausa L1 group, 4,075 for the Nigerian Arab Hausa L2, a 60%:40% division.
(14.) I should note that the terms "priming" and "spillover" applied to cross-sentence language sequence are my own. I know of no psycholinguistic studies in which cross-sentence language dependencies are investigated in detail. Even monolingually-based, cross-sentence dependency studies, for instance that studied under the rubric of syntactic persistence, are not numerous. In a recent study, Smith and Wheeldon (2001) do show cross-sentence priming effects for coordinated NPs, indicating that cross-sentence priming is plausible as a general phenomenon.
(15.) Furthermore, in this pattern the switch to an embedded Hausa island is overproportionally flagged by a pause. Whereas all Hausa EL constituent islands in Nigerian Arabic matrices are preceded about 33% of the time by a pause, in this priming pattern they are set off by pause 42% of the time. On the other hand, the Nigerian Arabic embedded islands in this pattern are relatively rarely (about 12%) preceded by a pause. It appears that in some sense the CP-final switches in the priming pattern are less integrated into the conceptual structure of the CP of the ML than those in the CP-initial switches in the spillover pattern.
(16.) No independent measures of respondent competence in L2 were offered.
(17.) 93, or about 66% of the spillover tokens in Table 5a represent switches between either [subject]--predicate or [[topic]--I]cp. This is prima facie evidence that topicality correlates with the spillover pattern identified in Table 5. The remaining spillover constituents are mainly different types of adverbs, as in example (3) above.
(18.) There are few detailed CS studies where the same group of speakers make heavy use of two matrix languages, and utilize further languages for lexical support. As noted above, the situation is not uncommon in Africa. However, finding data sets for direct comparison is difficult. Haust (1995) (also Haust and Dittmar 1998) has many similarities, with Wolof and Mandinka as well-represented MLs, English as a heavily-used lexical donor. It appears in her study, however, that the statistics largely summarize the configurations L1 Wolof + English or L1 Mandinka + English. There are indications (Haust 1995: 269) that Wolof speakers in her corpus use Mandinka L2 to a degree, which would correspond to the repertoire structure in the current study of L1 Nigerian Arabic + L2 Hausa. However, Haust does not describe the linguistic outcomes of this L2 Mandinka and what sort of CS it supports.
(19.) This incidentally would be one line of explanation for the proportionally higher frequency of insertion of English nouns in Nigerian Arabic as opposed to Hausa plural contexts discussed in Section 5.2. The more marked plural context requires a greater degree of processing power, which is more readily available in a Nigerian Arabic matrix.
(20.) Descriptively it could be said that a sort of convergence is observable here in that Hausa and Nigerian Arabic matrices handle English plurals morphologically in the same way. I do not know if this is peculiar to the Nigerian Arabic LI speakers, or whether English plurals are treated in a similar way by Hausa L1 speakers in Maiduguri.
Auer, Peter (1988). A conversation analytical approach to code-switching and transfer. In Codeswitching: Anthropological and Sociolinguistic Perspectives, M. Heller (ed.), 187-213. Berlin and New York: Mouton de Gruyter.
Backus, Ad (2003). Units in code switching: evidence for multimorphemic elements in the lexicon. Linguistics 41(1), 83-132.
Baldi, Sergio (1988). A First Ethnolinguistic Comparison of Arabic Loanwords Common to Hausa and Swahili. Naples: Istituto Universiterio Orientale.
Bentahila, Abdelali; and Davies, Eirlys (1998). Codeswitching, an unequal partnership. In Codeswitching Worldwide, Rodolpho Jacobson (ed.), 25-49. Berlin and New York: Mouton de Gruyter.
Boumans, Louis (1998). The Syntax of Codeswitching. Analysing Moroccan Arabic/Dutch Conversation. Tilburg: Tilburg University Press.
Bross, Michael (2001). L2 speakers, koineization and the spread of language norms: Hausa in Maiduguri, Nigeria. Paper presented at the conference Language Migration and the City, Bayreuth University.
--(2002). State and perspective of languages in Maiduguri. In From Bulamari to Yerwa to Metropolitan Maiduguri, Rupert Kawka (ed.), 103-116. Cologne: Koppe.
Budzhak-Jones, Svitlana; and Poplack, Shana (1997). Two generations, two strategies: the fate of bare English-origin nouns in Ukranian. Journal of Sociolinguistics 1,225-258.
Bybee, Joan (1995). Diachronic and typological properties of morphology and their implications for representation. In Morphological Aspects of Language Processing, Laurie Feldman (ed.), 225-246. Hillsdale, N J: Lawrence Erlbaum.
Ferreira, Fernanda (2000). Syntax in language production: an approach using tree-adjunction grammar. In Aspects of Language Production, Linda Wheeldon (ed.), 291-330. Hove: Taylor and Francis.
Field, Frederic (2002). Linguistic Borrowing in Bilingual Contexts. Amsterdam: John Benjamins.
Finlayson, Rosalie; Calteaux, Karen; and Myers-Scotton, Carol (1998). Orderly mixing and accommodation in South African codeswitching. Journal of Sociolinguistics 2, 395-420.
Goyvaerts, Didier (1997). Power, ethnicity and the remarkable rise of Lingala in Bukavu, Eastern Zaire. International Journal of the Sociology of Language 127, 25-46.
--; and Zembele, Tembe (1992). Codeswitching in Bukavu. Journal of Multilingual and Multicultural Development 13, 71-81.
Greenberg, Joseph (1947). Arabic loanwords in Hausa. Word 3, 85-97. --(1974). Language Typology: A Historical and Analytical Overview. The Hague: Mouton.
Gumperz, John (1982). Discourse Strategies. Cambridge: Cambridge University Press.
Hasselmo, Nils (1975). Bilingualism as an emerging sociolinguistic system. In The Nordic Languages and Modern Linguistics 2, Karl Dahlstadt (ed.), 245-266. Stockholm: Almqvist and Wiksell.
Haugen, Einar (1950). The analysis of linguistic borrowing. Language 26, 210-231.
Haust, Delia (1995). Codeswitching in Gambia. Cologne: Koppe.
--; and Dittmar, Norbert (1998). Taxonomic or functional models in the description of codeswitching: evidence from Mandinka and Wolof in African contact situations. In Codeswitching Worldwide, Rodolpho Jacobson (ed.), 79-90. Berlin and New York: Mouton de Gruyter.
Kaufman, Dorit; and Aronoff, Mark (1991). Morphological disintegration and reconstruction in first language attrition. In First Language Attrition, Herbert Seliger and Robert Vago (eds.), 175-188. Cambridge: Cambridge University Press.
Kroll, Judith; and Stewart, Erika (1994). Category interference in translation and picture naming: evidence for asymmetric connections between bilingual memory representations. Journal of Language and Memory 33, 149-174.
Laine, Matti; Vainio, Seppo; and Hyona, Jukka (1999). Lexical access routes to nouns in a morphologically rich language. Journal of Memory and Language 40, 109-135.
Levelt, Willem (1989). Speaking: From Intention to Articulation. Cambridge: Cambridge University Press.
--; Roelofs, Ardi; and Meyer, Antje (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences 22, 1-75.
Meuter, Renata; and Allport, Alan (1999). Bilingual language switching in naming: asymmetrical costs of language selection. Journal of Memory and Language 40, 2540.
--; Humphreys, Glyn; and Rumiati, Raffaela (2002). Bilingual language switching and the frontal lobes: modulatory control in language selection. International Journal of Bilingualism 7, 109-124.
Muysken, Pieter (2000). Bilingual Speech." A Typology of Code-Mixing. Cambridge: Cambridge University Press.
Myers-Scotton, Carol (1997). Duelling Languages: Grammatical Structure in Codeswitching, 2nd ed. Oxford: Clarendon.
--(1998). Structural uniformities vs. community differences in codeswitching. In Codes-witching Worldwide, Rodolpho Jacobson (ed.), 91-108. Berlin and New York: Mouton de Gruyter.
--(2000). The matrix language frame model: developments and responses. In Codeswitching Worldwide II, Rodolfo Jacobson (ed.), 23-58. Berlin and New York: Mouton de Gruyter.
--(2002). Contact Linguistics: Bilingual Encounters and Grammatical Outcomes. Oxford: Oxford University Press.
--; and Jake, Janice (1995). Matching lemmas in a bilingual language competence and production model: evidence from intrasentential code switching. Linguistics 33, 981-1024.
--; and Jake, Janice (2000). Four types of morpheme: evidence from aphasia, codeswitching and second language acquisition. Linguistics 38, 1053-1100.
--; and Jake, Janice (2001). Explaining aspects of code-switching and their implications. In One Mind, Two Languages, Janet Nicol (ed.), 84-116. Oxford: Blackwell.
--; Jake, Janice; and Okasha, Maha (1996). Arabic and constraints on codeswitching. In Perspectives on Arabic Linguistics IX, Mushira Eid and Dilworth Parkinson (eds.), 9-43. Amsterdam: John Benjamins.
Nishimura, Miwa (1989). The topic-comment construction in Japanese-English codeswitching. Worm Englishes 8, 365-377.
--(1995). A functional analysis of Japanese-English code-switching. Journal of Pragmatics 23, 157-181.
Nortier, Jacomine (1990). Dutch and Moroccan Arabic in Contact." Code Switching among Moroccans in the Netherlands. Foris: Dordrecht.
Owens, Jonathan (1998). Neighborhood and Ancestry: Variation in the Spoken Arabic of Maiduguri, Nigeria. Amsterdam: John Benjamins.
--(2000). Loanwords in Nigerian Arabic: a quantitative approach. In Arabic as a Minority Language, Contributions to Sociolinguistics 83, J. Owens (ed.), 259-346. Berlin and New York: Mouton de Gruyter.
--(2002). Processing the world piece by piece: iconicity, lexical insertion and possessives in Nigerian Arabic codeswitching. Language Variation and Change 14, 173 207.
Poplack, Shana; Sankoff, David; and Miller, Christopher (1988). The social correlates and linguistic processes of lexical borrowing and assimilation. Linguistics 26, 47-104.
--; and Meechan, Marjory (1998). How languages fit together in codemixing. IJB 2, 127-138.
Portin, Marja; and Laine, Matti (2001). Processing cost associated with inflectional morphology in bilingual speakers. Bilingualism: Language and Cognition 4, 55-62.
Roehlfs, Ardi (2000). Weaver++ and other computational models of lemma retrieval and word-form encoding. In Aspects of Language Production, Linda Wheeldon (ed.), 71-114. Hove: Taylor and Francis.
Sebha, Mark; and Wootton, Tony (1998). We, they and identity: sequential versus identity-related explanations in code-switching. In Code-switching in Conversation, Peter Auer (ed.), 262-289. London: Routledge.
Smith, Mark; and Wheeldon, Linda (2001). Syntactic priming in spoken sentence production--an online study. Cognition 78, 123-164.
Snodgrass, Joan (1993). Translating versus picture naming: similarities and differences. In The Bilingual Lexicon, Robert Schreuder and Bert Weltens (eds.), 83-113. Amsterdam: John Benjamins.
Swigart, Leigh (1992). Two codes or one? The insiders' view and the description of codeswitching in Dakar. Journal of Multilingual and Multicultural Development 13, 83-102.
von Studnitz, Roswitha; and Green, David (1997). Lexical decision and language switching. International Journal of Bilingualism 1, 3-24.
University of Maryland
Received 3 December 2002 Revised version received 20 January 2004
Table 1. Continuation in same language between speaker turn Frequency Percentage % continue Arabic 4076 67 continue Hausa 1961 33 continue English 31 .5 continue Kanuri 9 .1 continue Standard Arabic 0 0 Table 2. Number of words per language in ten texts Text Nigerian Hausa English Standard Kanuri Arabic Arabic 2 3,591 4,556 156 0 0 3 6,682 1,811 815 271 3 4 2,775 1,121 213 3 0 5 6,165 2,894 333 150 1 6 3,424 4,974 285 7 0 7 6,950 3,949 231 256 4 9 4,639 3,212 490 65 0 10 4,749 1,418 31 57 0 11 3,783 4,959 22 63 0 12 5,966 774 945 36 99 Total 48,724 29,668 3,521 908 107 word ratio of Nigerian Arabic to Hausa (total words): 57%:43%; 62%:38% (only Hausa+ NA words) Table 3. Lexical insertions Embedded language English Hausa SA Matrix language Nigerian Arabic 1121 139 638 Hausa 484 13 Table 4. English inserted noun in plural contexts Number of % of contexts contexts Arabic matrix: NA plural 22 15 -s 78 54 [empty set] 44 31 Total pl. contexts 144 Hausa matrix: Hausa plural 8 22 -s 19 51 [empty set] 10 27 Total pl. contexts 37 Table 5. Priming and spillover Priming Spillover a. Language switch (total) Arabic to Hausa 52 (A [H] # H) 24 (H # [H] A) Hausa to Arabic 49 (H [A] # A) 111 (A # [A] H) Pearson chi-square = 32, df = 1, p < .000 (all probabilities are 2-sided) b. Within speaker Arabic to Hausa 23 5 Hausa to Arabic 17 59 Pearson chi-square = 29, df = 1, p < .000 c. Between speaker Arabic to Hausa 29 19 Hausa to Arabic 32 52 Pearson chi-square = 7.6, df = 1, p < 0.006 Table 6a. Lexical insertions Nigerian Arabic matrix Hausa matrix supports Standard Arabic insertions supports few Standard Arabic insertions supports English lexical insertions in supports proportionally fewer English lexical plural contexts insertions in plural contexts Table 6b. Embedded syntactic insertions Hausa: overproportional priming in context A [H] # H Nigerian Arabic: overproportional spillover into CP-initial position: A # [A] H