Codeswitching as Projection of Bilingual Lemmas in Contact

Unlike most previous studies of Codeswitching (CS) focused on describing surface configurations of switched items (i.e., where CS is structurally possible) or the switched items (i.e., what items from another language can be switched), this paper explores formulation processes of bilingual speech and the nature of the bilingual mental lexicon and its activity in CS. More specifically, it applies the Bilingual Lemma Activation Model (Wei, 2002, 2006b) to the data drawn from various naturally occurring CS instances. It claims that the mental lexicon does not simply contain lexemes and their meanings, but also lemmas, which are abstract entries in the mental lexicon that support the surface realization of actual lexemes. Lemmas are abstract in that they contain phonological, morphological, semantic, syntactic and pragmatic information about lexemes. It further claims that lemmas in the bilingual mental lexicon are language-specific and are in contact during a discourse involving CS at three levels of abstract lexical structure: lexical-conceptual structure, predicate-argument structure, and morphological realization patterns. The CS instances described and analyzed in this paper provide evidence that the bilingual speaker’s two linguistic systems are unequally activated in CS, and CS is an outcome of bilingual lemmas in contact.

theta role of PERCEPT (i.e., what is known), and these elements appear in a particular order. As evidenced in Richards (1976), Faerch and Kasper (1984), Nation (1990), Ringbom (1987), Pienemann (1999 and others, lemmas also contain information about the word 's spelling, pronunciation, its register, the kind of discourse it typically enters into, and it is pragmatic function. Thus, the mental lexicon can be defined as the speaker's internal representation of language specific knowledge about the surface forms, that is, lemmas. Sharing the view of Kaplan and Bresnan (1982), Jackendoff (1983), Kempen and Hoenkamp (1987), Levelt posits that the lexical informat ion stored with the entry for a part icular verb has to co mmunicate with the other constituents in the sentence as driven by the verb itself and exp licates the lemma for give: conceptual specification: CA USE (X, (GOposs (Y, (FROM/TO (X, Z))))), conceptual arguments: (X, Y, Z), syntactic category: V, grammatical functions: (SUBJ, DO, IO) (198 9, p. 191). As exp licated, give requires three conceptual arguments, each of which is assigned a specific thematic ro le, and they are mapped onto the grammatical functions : (Pienemann, 1999, p. 63) Thus, the "mental lexicon" differs fro m the "lexicon" in that it is not just an indiv idual speaker's collection of words but deals with how those words are stored, activated, processed, and retrieved by each speaker. The activation of lemmas in the mental lexicon p lays a central ro le in speech production.
That is, lemma activation of [articu lar lexical items in the ment al lexicon med iate between conceptualizat ion and speech formulation as a necessary level of speech p roduction.
One of the most crucial assumptions underlying this study is that lemmas are language -specific for lexicalizat ion patterns of a particular language. For the notion of language-specific lemmas, one of the frequently cited examp les is provided by Talmy (1985, p. 69): (English) The bottle floated into the case vs. (Spanish) La botella entró a la Cueva flotando (The bottle moved-into the cave floating). While in English, motion with manner or condit ion of movement can be conflated into a single lemma (i.e., FLOAT), in Spanish, a different pattern is required to exp ress the notion of FLOATING periphrastically using the gerund.
In Levelt's model of speech production (1989), semantic and syntactic information constitute the lemma of the lexical information wh ile mo rphological and phonological information constitute the form o f the lexical item. Levelt's model is designed for describing the major co mponents and processes of monolingual speech production, and it must be adapted to account for bilingual speech behavior such as CS. Myers-Scotton and Jake (1995) claim that lemmas are activated by language-specific semantic/pragmat ic feature bundles that come fro m the CONCEPTUA LIZER. Following the above lines of thin king, th is study proposes that it is the preverbal message or the speaker's commun icative intention that activates language-specific lemmas in the bilingual mental lexicon. In other word s, it is the semantic/pragmatic feature bundles selected by the CONCEPTUA LIZERE at the conceptual level that trigger the appropriate lemmas into activity before the FORM ULATOR has access to the relevant lexical items in the mental lexicon. Adapted from Levelt (1989), Jake (2000a, 2000b) and Wei (2002), below is the simp lified model of lemma activation in the bilingual mental lexicon.
(>: "before") Conceptual Level: activation of speaker's preverbal message/intention → semantic/pragmat ic feature bundles selected by the CONCEPTUA LIZER > Lemma Level: activation of language-specific lemmas > Functional Level: act ivation of the FORM ULATOR → projection of language -specific procedures > Position Level: p rojection of surface forms → mo rphological/phonological realizat ion patterns.
One of the key questions being asked about the nature of the bilingual mental lexicon is whether the bilingual's two lexicons are linked to a shared conceptual store or separated in two conceptual stores (Keatley, 1992). So me recent bilingual processing and representation models assume that in the bilingual mental lexicon while phonological and morphosyntactic forms can be rather easily differentiated, lexical meanings or concepts are largely shared (Kroll and Ste wart, 1994;Kro ll and Sunderman, 2003;Costa, 2005). Accordingly, most studies of conceptual representation have focused on the links between word forms and mean ings and factors that affect the bilingual conceptual processing, but not the nature of conceptual representation itself (Kroll & de Groot, 1997;de Groot, 2002;Kroll & To kowicz, 2005). Other studies have focused on cross -linguistic differences in lexical/conceptual representation (Pavlenko, 2009;Jarvis, 2009;Altarriba & Basnight-Bro wn, 2009;Jiang, 1999Jiang, , 2000.
Influenced by the above linguistic and psycholinguistic models of the bilingual lexical/conceptual representation and based on the BLA Model, this paper claims that lemmas in the bilingual mental lexicon are language-specific. This is because while the monolingual "mental lexicon represents a complex self-organizing system," the "bilingual mental lexicon, as opposed to the monolingual one, integrates the units of two linguistic systems and, therefore, ensures the processes of speech perception and production in two languages" (Leshchenko, Dotsenko, & Ostapenko, 2015, p. 1040. It further claims that language-specific lemmas in contact during a discourse involving CS. As assumed, CS is described and explained in terms of bilingual lemma activation at any level of abstract lexical structure. Based on some naturally occurring and commonly observed CS instances, this paper presents three arguments: (1) Lemmas are unequally activated in either monolingual or b ilingual speech production. It is the unequal activation of lemmas fro m the bilingual mental lexicon which motivates and constraints the speaker's choice of morphemes in CS (Wei, 2002).
(2) It is the structural procedures dictated by the Matrix Language (ML) (similar to the notion of "base" or "host" language) which set the sentential frame for sentences containing switched items the Embedded Language (EL) (similar to the notion of "guest" language) (Myers-Scotton, 1993 them as the EL, but the activated lemmas fro m the EL must be sufficiently congruent with the ML counterparts at each of the three levels of abstract lexical structure: lexical -conceptual structure, predicate-argument structure, and morphological realization patterns. Thus, the fundamental assumption is that CS itself is a linguistic system and is governed and constrained by a set of morphosyntactic principles and rules governing any other linguistic system, and CS is an outcome of bilingual lemmas in contact.

Codes witching as Bilingual Lemma Acti vation
Adopting Levelt's (1989) speech production model, Myers -Scotton and Jake's (1995) bilingual language competence and production model and Wei's (2002) bilingual speech production model, Wei (2006b) proposes the BLA Model to describe and explain CS in terms of the nature and activity of the bilingual mental lexicon. This model also draws on some contemporary proposals and assumptions about the nature and activity of bilingual language modes in speech production. For examp le, Faerch and Kasper (1986) and Grosjean and Soares (1986) assume that the bilingual's language systems can be kept separate because they may be activated at different levels, depending on whether they are used at the mo ment. Green (1986) and Paradis (1981Paradis ( , 1997 propose that the language system of a bilingual is organized in two subsets, one for each language, that can be selectively activated, activated simu ltaneously to various degrees, or deactivated independently for one another. Thus, as assumed , when the bilingual is in a monolingual mode, the other language must be deactivated or inhib ited. Green (1986) postulates that bilingual speakers who wish to select a particular language for the current communicat ion must ensure that its activation exceeds that of the competing language(s) in their procession. According to Soares and Grosjean (1984), Green (1986) and Grosjean (1997), the deactivation of bilingual speakers' other language(s) is rarely total. Th is is clearly evidenced in the interference effects in bilinguals' production of the speaker-specific deviations fro m the target language being spoken due to the influence of the deactivated language. Paradis (1997) assumes that when bilingual speakers select one language rather than the other, the activation threshold of the nonselected language must be raised; however, the language not being selected for use at the mo ment is never totally deactivated. Paradis (1989) and Berg and Schade (1992) even claim tha t deactivation of one language may be so difficu lt that abundance involuntary mixing or hybrid ization may occur. According to Paradis (1997), deactivation or inhib ition of one of a b ilingual's languages is a matter of degree.
However, "this deactivation has led to much theorizing and much controversy around the notion of a language switch or a mon itor system," but "what is certain, however, is that bilinguals rarely deactivate the other language totally. This is seen in various types of production interference-the involuntary influence of one language on the other…" (Grosjean & Spares, 1986, p. 146). According to Grosjean (1997), if the bilingual is in the monolingual mode the guest language becomes deactivated as least as possible and as a consequence, the deactivated guest language does not act upon the base language often; if the bilingual is in the bilingual mode, the guest language becomes activated to a relatively high level but less so than the base language, and thus the activated guest language intrudes upon the base language very often. Bilinguals find themselves at various points along the language mode continuum, which corresponds to different levels of activation of the two languages, and the base language is always more strongly activated than the guest language (Grosjean, 1985(Grosjean, , 1989(Grosjean, , 1994(Grosjean, , 1997. Following these lines of thinking, Wei (2000aWei ( , 2000bWei ( , 2002Wei ( , 2006b assumes that when bilinguals are in the bilingual mode to produce CS, both of their languages are activated, but their ML is more strongly activated than their EL. According to Wei (2020), it is the language mode chosen and the relative degree of act ivation of the ML and the EL that cause the amount of CS and the morphosyntactic principles governing CS.
Sotillo's investigation of morphosyntactic features and sociolinguistic functions of CS in face-to-face and short message service conversations among Spanish -English and Tagalog-English bilinguals finds that the bilingual lemma activation "facilitates the interpretation of bilingual speech behavior and switches between languages as cognitively based strategies at an abstract level" (2016, p. 21). This finding supports Wei's proposal (2006b ) that CS can be better exp lained at an abstract level in the sense that lemma act ivation of particular lexical items in the bilingual mental lexicon must mediate between the conceptual level, where specific semantic-prag matic feature bundles are selected as desired for communicat ion, and the functional level, where language-specific morphosyntactic procedures are projected before surface forms are realized at the positional level. The fundamental assumption underlying the BLA Model is that bilingual lemmas are never equally activated in CS, and the unequal activation of bilingual lemmas is the driving force for CS.

Codes witching as Unequal Acti vation of Bilingual Lemmas
As introduced earlier, the BLA Model assumes that lemmas are abstract entries in the mental lexicon which support the surface realization of actual lexemes because lemmas contain phonological, morphological, semantic, syntactic, and pragmatic information about lexemes stored in the mental lexicon. That is why lemmas in the mental lexicon are defined as the speaker's internal representation of knowledge about surface forms. Different fro m any other models of the mental lexicon, the BLA Model further proposes that lemmas in the bilingual mental lexicon are language-specific, and CS is an outcome of bilingual lemmas in contact. Different fro m other models of CS, the BLA Model claims that CS ju xtapositions which may surface do not have much to do with surface linear or typological correspondences between the participating languages but originate with d irections contained in lemmas (cf. Myers-Scotton & Jake, 1995;Wei, 2001aWei, , 2001b.
Regarding the bilingual mental lexicon, Green (1986Green ( , 1993 and de Bot and Schreuder (1993) propose that the lexical items belonging to different languages must be organized in subsets which can be either fully activated or deactivated. Myers -Scotton (1993) proposes that there is a universal set of semantic and pragmatic feature bundles availab le for the lexical-conceptual structure of lemmas. Wei (2002Wei ( , 2006b proposes that lemmas in the bilingual mental lexicon are language -specific and lexicalization patterns across languages reflect that semantic and pragmatic feature bundles across related lemmas in different languages are presented in different configurations. Adopting these proposals, the BLA Model claims that it is cross -linguistic d ifferences in how informat ion is organized at the level of lexical-conceptual structure and at the level of predicate-argu ment structure that affect code choices and structures which are predicted to occur in CS. Such a claim imp licates that it is the unequal activation of bilingual lemmas that drives CS and constrains its lexical and structural configurations.

Morphemes Unequally Activated in Codeswitching
One of the major assumptions underlying the BLA Model is that lemmas, in addit ion to other abstract entries about particular lexemes, contain semantic and prag mat ic feature bundles which encode the lexical-conceptual structure representing the speaker's co mmunicative intention as preverbal message generated by the CONCEPTUA LIZER (Levelt, 1989;Poulisse & Bongaerts, 1994). Th is is because at the level of lexical-conceptual structure the speaker seeks appropriate linguistic material for h is/her communicat ive intention, and it is at this level the speaker conceptualizes and act ivates the appropriate concepts. The activated concepts will then activate the corresponding lemmas in the mental lexicon.
According to Roelofs (1992), Levelt (1995), and Wei (2002), it is sufficiently activated lemmas that activates the associated lexeme. According to Myers -Scotton (1993), there is a universal set of semantic and pragmatic features availab le for the lexical-conceptual structuring of lemmas. Wei (2001b) claims that the presence and conflation of universally available semantic and prag matic features may vary cross-linguistically. The BLA Model proposes that the relative importance of cross -linguistic lexical-conceptual differences in lemmas in the bilingual mental lexicon and the consequences of CS should be considered as evidence of variation in semantic-prag matic feature bundles.
As introduced above, lemmas in the bilingual mental lexicon are language -specific. In other words, bilingual lemmas are tagged for specific languages and support the realization of actual lexemes. The naturally occurring CS instances to be discussed indicate that it is content morphemes, rather than system morphemes, wh ich encode the speaker's intended meaning. Wei (2001bWei ( , 2002 posits that the major reason for content and system morphemes to be accessed differently lies in the fact that only content morphe mes contain semantic/prag matic feature bundles, but system mo rphemes only p lay their abstract grammatical notions. At a certain point during a discourse, the speaker may switch to certain EL content morphemes to convey his/her intended or desired meanings as generated in his/her preverbal message. It is in this sense that certain language-specific lemmas are conceptually projected fro m the bilingual mental lexicon in a discourse involving CS.
Below are some CS instances involving various language pairs which show that EL content morphemes can be easily switched because they are projected fro m the EL lemmas sufficiently congruent with those of the ML. [1] I co mmand you to do the nokum.
You that-CL article yet not fin ish PART/AFFIRM -QUE "You haven't fin ished that article yet?" (Ch inese/English; Wei, 2002, p. 696) [10] ii desu keredo mo tuition ga totemo expensive desu. persuade is from English, the EL, but the complementizer la is a system morpheme fro m Marathi, the ML. In [8] terras is fro m Dutch, the EL, but it receives locative case from Turkish, the ML. In [9] article and finish are content morphemes fro m English, the EL, but nei (equivalent to "that" in English) and pian (a Chinese noun classifier) going together as a determiner is fro m Chinese, the ML. Also, there is no perfect aspect marking on the English verb finish, since Ch inese does not have verb morphology of any sort for th is and other grammatical purposes. In [10] tuition and expensive are fro m English, the EL, but totemo (very) modifying expensive is a system morpheme fro m Japanese, the ML.
The above examp les provide the ev idence that in CS, b ilingual lemmas are in contact and are not equally activated. EL content morphemes can be freely activated to be switched for the speaker's communicat ive intention at a certain point during a discourse, but EL system morphemes cannot. The evidence clearly indicates that it is the ML which provides all system mo rphemes.

Morphosyntactic Procedures Unequally Activated in Codeswitching
In addition to the assumption that morphemes are unequally activated in CS, the other crucial assumption underlying the BLA Model is that the language pairs involved in CS do not participate equally, one p laying a more central ro le than the other in controlling mo rphosyntactic procedures. In other words, bilingual lemmas fo r mo rphosyntactic procedures are unequally act ivated in CS (Myers-Scotton, 1993). In a bilingual mode, although both language networks are activated, one is more activated than the other (Green, 1986;Grosjean, 1997). One of the main argu ments of the MLF Model is that CS occurs within the constraints of a sentential frame wh ich must be set by the morphosyntactic procedures dictated by the ML wo rd order and syntactically relevant system morphemes (i.e., relational or functional elements) (Myers-Scotton, 1993). As assumed, at the conceptual level the discourse mode is chosen with one of the participating languages as the ML and then corresponding language-specific lemmas are activated at the lemma level to realize the speaker's preverbal message, resulting in CS morphosyntactically controlled by the ML. Grosjean (1989Grosjean ( , 1997 claims that the amount of language mixing depends on the language mode the bilingual speaker is cu rrently in, whether monolingual, b ilingual, or anywhere else on the language-mode continuum between these two modes. Wei (2015) fu rther claims that it is particular levels of activation of the ML and the EL which determine the bilingual speaker's adoption of a particular position on the language-mode continuum. de Bot and Schreuder (1993) and Wei (2009a) claim that bilingual speakers are able to separate the language systems they know and to mix them in a bilingual mode. Th is is because that bilingual speakers know that the ML and the EL play unequal roles in CS, and CS is not simply a so-called "mixed" speech but is governed by a set of structural principles, such as those proposed in the MLF Model (Myers -Scotton, 1993Myers -Scotton & Jake, 1995;Wei, 2001b) and the BLA Model (2006b). Such a notion of language separation becomes crucial in explaining the structural principles governing CS and in exp loring the nature and activity of the bilingual mental lexicon in a discourse involving CS.
Below are so me CS instances involving various language pairs which provide strong eviden ce that it is the ML which p rovides all the morphosyntactic procedures for CS.
c.9-t ime DEM-c.9 c.9-non-PAST-depend with c.10-cert ificate c.10-your z-a Ø-shule c.10-A SSOC The representative CS instances involving various language pairs discussed in the above sections reveal that bilingual speakers can activate any language known to them as the M L, and the M L is mo re activated than the EL. As assumed in the M LF Model and the BLA Model, it is the ML whose lemmas are fully activated fo r both content and system morphemes and morphosyntactic procedures, and the EL only supplies certain content morphemes to form M L+EL constituents. The BLA Model emphasizes that such an activation of deactivation occurs at the conceptual level of the speech production process.
If at the conceptual level, bilingual speakers decide to choose the bilingual mode, they will then choose intersentential or intrasentential codeswitching, and if the latter is chosen, they must choose one of the languages as the ML. If the bilingual mode and intrasentential codeswitching (CS) are chosen, language-specific semantic/prag matic feature bundles will activate language-specific lemmas in the bilingual mental lexicon at the lemma level. Such activated lemmas will send directions to the FORMULATER at the functional level for project ion of language-specific mo rphosyntactic procedures to be realized at the position level, where the activated EL content morphemes are inserted into the grammatical frame p rovided by the ML.

Lemma Congruence Checking as an Organizing Princi ple Governing Codes witching
Lemma congruence is defined as "a match between the ML an d the EL at the lemma level with respect to linguistically relevant features" (Myers -Scotton and Jake, 1995, p. 985 predicate-argument structure, thematic structure is mapped onto grammat ical relations. At the level of morphological realization patterns, word order, agreement mo rphology, case marking, tense/aspect marking, and phonological forms are realized. Lemma congruence must be checked at each of these levels in the bilingual mental lexicon, wh ich determines the way an EL item may be integrated into an ML frame. Relevant to the study presented in this paper is lemma congruence at the first two levels of abstract lexical structure.
The BLA Model views the nature of lemma congruence relevant to CS as mo re comp lex because several different levels or subsystems must be checked, and articulates the relation of lemma congruence checking to a model of b ilingual speech production. The BLA Model assumes that CS ju xtapositions which may surface do not have much to do with superficial linear or typological correspondences between the participating languages. It argues that since lemmas are abstra ct entries in the mental lexicon (Levelt, 1989), CS ju xtaposition must orig inate with directions in the speaker's mental lexicon (Myers-Scotton & Jake, 1995). According to Wei (2001bWei ( , 2002Wei ( , 2006b, lexicalization patterns across languages reflect the fact that there are different configurations of features across related lemmas in different languages. Hypotheses about cross -linguistic differences in how informat ion is organized at the level of lexical-conceptual structure and at the level of predicate-argument structure, whether pragmatic or semantic, or semantic with morphological consequences, affect the structures which will appear in CS. That is why the BLA Model regards lemma congruence between the languages as an organizing princip le in CS production.

Lemma Congruence in Lexical-Conceptual Structure
One obvious reason for bilingual speakers to engage in CS is that they select individual EL content morphemes to encode their communicat ive intentions. That is, at the conceptual level bilingual speakers do not produce surface level morphemes but rather make appropriate choices about the semantic/pragmat ic feature bundles as desired. The semantic/prag matic feature bundles chosen at this abstract level activate the lemmas in the bilingual mental lexicon to support surface level morphemes.
However, the EL lemmas activated by the semantic/pragmatic feature bundles at the conceptual level must be sufficiently congruent with the counterparts of the ML in order fo r CS to occur. In other words, sufficient congruence between the EL and ML lemmas support the existing lexemes in both languages.
de Bot and Schreuder (1993) posit that because different languages may lexicalize in d ifferent ways, the language to be used in L2 production must be specified befo re chunking takes p lace. Relevant to CS is the imp licat ion that the language pairs involved may differ in semantic/prag matic feature bundles encoded in lexical-conceptual structure. Grosjean (1982) reports that some codeswitches are motivated by the lack of a particu lar word in one of the languages or by the greater availability of a word in the other language. Wei (2006b) claims that it is language-specific lemma d ifferences at the level of lexical-conceptual structure which motivate CS. In terms of lemma congruence between the languages involved in CS, there is still sufficient cross -linguistic congruence if such lemma differences are only partial. The BLA Model posits a partial lemma difference is one of the major reasons for a particular EL content morpheme to be activated and switched.
Below are some CS instances which illustrate the notion of "lemma congruence" checking in lexical-conceptual structure.
I have two-CLA SSIF paper to morrow must turn in but I at the mo ment one-CLASSIF yet not fin ish PARTIC/AFFIRM "I have two papers [which] I must turn in to morro w, but at the mo ment I haven't finished one yet." [21] zhu zai zheli hen fanbian, meitian you school bus.
live PREP/ LOC here very convenient everyday have school bus "It"s very convenient to live here. There is a school bus every day." [22] wo xiawu qu jian wode advisor. wo bu neg he ni yiq i qu mall le.
I afternoon go see my advisor I not can with you together go mall PARTIC/AFFIRM "I'm going to see my advisor this afternoon. I can't go to the mall with you." (Ch inese/English; Wei, 2001b, p. 159) In [20] paper in English may mean any written piece of word, such as an article, a report, an essay or a composition, but the Chinese equivalent noun zhi (paper) itself on ly means a p iece o f paper to wrap things up in or to write something on. In [21] a school bus in English means a bus main ly for transporting students to and from a school, but the Ch inese equivalent noun phrase xiaoche (school bus) usually means a bus for transporting a school's sports or performance team or equip ment. In Ch ina, most schools even do not have xiaoche. In [22] an English advisor is expected to assume mo re responsibilit ies than a Chinese advisor. In the academic setting, an English advisor is an instructor or professor who offers advice or counsel to students regarding their academic weakness, improvement or progress, sequential course requirements, research projects, thesis or dissertation writ ing, and so on.
Most English advisors are also those who recommend their students to the job market or professional agencies. Chinese does have the equivalent noun daoshi (advisor), but a daoshi does not necessarily assume the same responsibilit ies as an English advisor. In Ch ina, only a g raduate student may have a daoshi, whose expected responsibility is to supervise a student in writ ing his/her thesis or dissertation.
These CS instances reveal that there exist semantic d ifferences in the semantic/prag matic feature bundles of conceptually-related lexemes. Although Chinese possesses conceptually similar lexemes, in these CS instances, the related English lexemes are selected as more appropriate to convey the speakers' intended meanings. (Japanese/English; Wei, 2006b, p. 170) In [23] the speaker switches to registration for the possible reason that in Japanese universities/colleges, though students must register for the courses to take they are not free to select the courses which they are interested in taking. The speaker may choose the English word to convey his/her in tended meaning more accurately. In [24] the speaker switches to bedroom for the possible reason that the concept of "bedroom" is relatively new to Japanese. A tradit ional Japanese room is often used not only for sleeping but also for eating, studying, entertaining guests, or for other daily family activit ies. In other words, the concept of "bedroom" in English is not the same as that in Japanese. In [25] the concept of a community force may not only be A merican, but the general expression "commun ity fo rce" in the American context may include "neighborhood crime watch", "drug free zone", and so on. The Japanese expression similar to "co mmun ity force" is chouka (neighborhood association), but such an association is mainly for organizing local social and cu ltural activit ies, overseeing environ mental sanitation, taking care of the old, mediat ing a dispute, and so on. The speaker switches to community force probably to mean something beyond Japanese chouka.
As commonly observed in naturally occurring CS instances, bilingual speakers may switch to particular lexical items of another language at a certain point during a discourse. This is most probably because of cross-linguistic differences in language-specific lemmas underlying particular lexemes. In terms of abstract lexical-conceptual structure, languages may possess similar lexical items, but language cues may have different values (Li, 1996;Nishimura, 1997;Wei, 2001bWei, , 2002. When the language cue specifies a particular language at a certain point of b ilingual speech production, its lexical item receives activation and is thus easier to be selected. As suggested by Green (1986), lemmas are tagged with a language label. Poulisse and Bongaerts (1994) propose the lexical access model to explain how and why bilinguals differentiate between the lexical items of two or more language systems. The implications of their model for explaining CS are that EL lemmas may receiv e more act ivation than the corresponding ML lemmas when the speaker's preverbal message contains the specification [+EL] for some reason or other. The speaker may switch to an EL lexical item because the ML has a similar but nonequivalent lexical item or the ML has not lexicalized a part icular concept. The EL lexical items in the above CS instances indicate that their lemmas receive the most activation and are selected as appropriate thereby allowing access to the corresponding EL lexemes.
The above instances show how cross-linguistic differences in semantic/pragmat ic feature bundles may affect lexical selections in mixed constituents. Checking for lemma congruence across ML and EL elements becomes necessary, and the result of this checking has consequences for how the EL lexemes supported by the EL lemmas in question will appear in CS.

Lemma Congruence in Predicate-Argument Structure
As assumed in the M LF Model and the BLA Model, the lexical items stored in the mental lexicon contain syntactic, morphological, and phonological informat ion about them in addition to the specification of their lexical content (i.e., semantic features). The activation of the appropriate lexical items frees the syntactic information about them, which will activate syntactic procedures. For grammatical encoding and phonological encoding (Levelt, 1989), the FORMULATOR must have access to the mental lexicon. Since lexical items fro m both languages involved in CS can be activated, the question becomes which of the two languages controls the syntactic procedures .
The MLF Model (Myers-Scotton, 1993 claims that whichever language activated as the ML provides the sentential frame into which the EL content morphemes are switched. Thus, it is the ML which controls the predicate-argu ment structure and supplies system morphemes, a subcategorization frame for the verb, and the morpheme order. It should be noted that there is an inseparable connection between the level of lexical-conceptual structure and the level of predicate-argu ment structure with regard to lemma selection fro m the mental lexicon. As stated by Kroll and de Groot, "language specific lemmas fro m the interconnection between the lexical-conceptual mappings to and from syntax" (1997, p. 190). The BLA Model (Wei, 2006b(Wei, , 2009a(Wei, , 2015(Wei, , 2020 emphasizes that lemma congruence checking at the level of lexical-conceptual alone is not sufficient enough for CS to occur. Lemma congruence checking at the level of p redicate-argu ment structure must also come into play. This is because in order for the supported EL lexeme to be morphosyntactically integrated into the sentential frame set by the ML, its morphosyntactic pattern must be congruent with that of the M L lexeme whose lemma is sending the morphosyntactic directions to the FORMULATOR. Wei (2009aWei ( , 2009b finds that one of the frequently occurring instances of Chinese/English CS is that Chinese bilinguals tend to switch to English verbs/verb phrases as well as English nouns/ noun phrases. [26] ni dei xiang banfa make money.
you must think way make money "You must think of ways to make money." [27] ta gong dao, ta dei xue dri ve.
he just arrive he must learn drive "He just arrived, and he must learn how to drive." [28] wo de che you give me trouble le.
UT-POSS building PREP/in su mmer all will demo lish PA RTIC/AFFIRM "All the buildings at UT (University Terrace) will be demo lished in summer." (Ch inese/English; Wei, 2009b, p. 332) In [26] the speaker switches to make money, which fits the Chinese morphosyntactic frame for the V-O order. In [27] the infinit ive verb drive is switched into the Chinese verb phrase as its direct object, which is allo wed in the Chinese morphosyntactic frame. In [28] give me trouble is switched into the Chinese subcategorization frame for the V-O-O order. In [29] complain is switched into the subject position, which is congruent with the Chinese sentential frame, excep t that Chinese does not possess the infinitive marker to to introduce the verb or the du mmy pronoun it to balance the sentence as in

Lemma Incongruence as a Dri vi ng Force for Codes witching
As commonly observed, there are cross -linguistic differences at any level of abstract lexical structure (i.e., at the level of lexical-conceptual structure, predicate-argument structure or morphological realization patterns). Consequently, when an EL lemma is selected, but it does not match that of the ML, some compro mising strategies must be taken in order for CS to occur. One of the compro mising strategies is for the speaker to produce EL islands (Jake & Myers-Scotton, 1997;Wei, 2001bWei, , 2002. An EL island is a constituent in which an EL content morpheme occurs with only other EL morphemes, including EL system mo rphemes. EL islands may be produced at the level of lexical -conceptual structure or at the level of predicate-argument structure when lemma incongruence occurs but the speaker still wants to be engaged in CS. Such a compro mising strategy makes CS possible because in producing an EL island, it is the EL which directs the FORMULATOR to activate only the EL mo rphosyntactic procedures. According to Levelt (1989), in speech production, it is the FORMULATOR wh ich gives language-specific directions for the grammatical and phonological encoding. For possible CS, if the language pairs cannot possibly be encoded by means of the same morphosyntactic procedures, the speaker may stop the encoding of one of them and continue with the other so as to solve the problem of lemma incongruence. Different fro m Levelt 's model of speech production, the MLF Model specifies that in CS the choice of one EL procedure versus another is determined by the larger ML frame. The BLA Mo del claims that lemma incongruence between the Language pairs involved in CS in regard to lexical -conceptual structure and predicate-argument structure is one of the major reasons for EL islands to be produced if the EL lemmas are selected for the speaker's co mmunicat ive intentions. The BLA Model regards lemma incongruence as a driving force for CS.

Lemma Incongruence in Lexical-Conceptual Structure
As commonly observed, languages do not lexicalize concepts in the same way due to the fact that lexical representations are language-specific. As assumed in the BLA Model, if the bilingual mode is chosen at the conceptual level, speakers do not produce surface morphemes but rather make appropriate choices about the semantic/pragmatic in formation that they in tend to convey. If the EL lemmas for some particu lar semantic/prag mat ic feature bundles are activated but do not match the ML counterparts, speakers must take the co mpro mising strategy by producing EL islands for possible CS configurations.
then you tomorrow call me "Then you call me to morrow." [32] ni neng-bu-neng give me a ri de? you can-not-can give me a ride "Can you give me a ride?" [33] na wo yid ian come to pick you up.
so I one o'clock co me to pick you up "So, I'll co me to pick you up at one o'clock." (Ch inese/English; Wei, 2001b, p. 162) In [31] the English semantic features of "co mmunicate with by telephone" are conflated in the verb call, but the Chinese equivalent to call me is da dianhua gei wo ("make phone to me "). Since the speaker chooses the EL lemma wh ich activates the EL lexical-conceptual structure, the whole verb phrase "call me " is activated and produced as an EL island. In [32] the verb phrase give me a ride is incongruent with the ML counterpart song wo yixia ("send me one time"). The incongruence between these two verb phrases lies in their individual lexical-conceptual structure. While in the EL lexical-conceptual structure of the means of transportation is conflated in the noun ride as the direct object of the verb give, in the M L lexical-conceptual structure it is conflated in the verb song, wh ich itself may not contain the means of transportation at all. The speaker chooses the EL exp ression most probably to make his intended meaning exp licit and specific. Thus, when the EL lemma underly ing "ride" is activated, the whole EL verb phrase is accessed and produced as an EL island. In [33] pick you up is accessed as an EL island, in wh ich the pronominal object you is placed between the verb pick and the particle satellite up. The speaker chooses pick up most probably because the EL phrasal verb contains the mean ing of "to take on as a passenger", but the ML counterpart jie usually does not. In Ch inese, the verb jie means "me et" (e.g., to go to the bus/train station or airport to meet somebody"), wh ich does not necessarily involve providing personal transportation. The speaker chooses the English phrasal verb pick up to realize his communicat ion intention more accurately. Also, in [33] co me is accessed with the infinitive maker to, an EL system morpheme, to introduce the phrasal verb pick you up. This is because in English the infin itive marker beco mes obligatory if two successive verbs are activated and selected simu ltaneously. Thus, come to pick you up is switched as a larger EL island into the ML sentential frame.
Cross-linguistic d ifferences in semantic/pragmat ic feature bundles are also found in bilingual speakers' choice of certain fixed/idio matic expressions from the EL. If the mean ing or intention contained in the EL lemma is not sufficiently congruent with that in the ML lemma, and if b ilingual speakers choose the one in the EL lemma, they will p roduce an EL island in CS.
you do-not-go Kmart hear have many things on sale "Are you going to Kmart? (I've heard there're many things on sale." [35] nali you wutai jiqi dan san-tai si out of order .
there have five-CLASSIF machine but three-CLA SSIF COP/be out of order "There were five machines there, but three of them were out of order.
[36] wo bu neng baozheng arrive your ho me on time but I surely co me.
I not can guarantee arrive your home on time but I surely co me "I can't guarantee (that I) will arrive at you're your ho me on time, but I'll surely co me." (Ch inese/English; Wei, 2001b, p. 163) In are not. Since Such EL lemmas are activated in the bilingual speakers' mental lexicon, the output of the expressions follows the EL lexical-conceptual structure and morphosyntactic procedures.
It seems that one of the major reasons for EL islands to be switched is that in the case of non-idio mat ic expressions, the speaker's intended meaning at a certain point during a discourse cannot be realized in the ML because of the insufficient congruence between the ML semantic/pragmatic feature bundles and those of the EL. The other major reason is that in the case of fixed/id io matic expressions, a complete EL island is activated as a single lexical unit fo r the speaker's preferred EL lexical-conceptual structure.
In either case, the compro mising strategy is to produce EL islands. In ot her words, a part icular EL semantic/pragmat ic concept is accessed as a single unit observing the specific EL morphosyntactic procedures.

Lemma Incongruence in Predicate-Argument Structure
As specified in the M LF Model, it is the ML which controls the mo rphosyntactic structure of the sentence containing the switched items fro m the EL by supplying system morphemes, subcategorizat ion frames for verbs, and mo rpheme order. As claimed in the BLA Model, morphosyntactic procedures are realized by the FORM UALTO R at the functional level, but before they are sent to the FORM ULATOR, lemmas fro m both languages can be activated at a certain point during a discourse. Thus, in addition to lemma congruence checking at the level of lexical -conceptual structure, lemma congruence checking at the level of predicate-argu ment structure must also come into play for possible CS. The BLA Model posits that even if the lexical-conceptual structures between the two languages are sufficiently congruent, the ML predicate-argument structure will reject the mapping if a particular EL predicate-argu ment structure is not sufficiently congruent with that of the ML. If this happens, but the speaker prefers the EL concept as realized in the EL predicate-argument structure, a compro mising strategy much be taken for possible CS. That is, the speaker must produce an EL island.
I can wait for you till t wo o'clock "I can wait for you till t wo o'clock." (Ch inese/English; Wei, 2001b, p. 166) [38] ta jingchang fails students in exams.
she often fails students in exams "She often fails students in exams." [39] ni b iye hou keyi teach English to nonnati ve s peakers .
you graduate CONJ/after can teach English to nonnative speakers "After you graduate, you can teach English to nonnative speakers." (Ch inese/English; Wei, 2001b, p. 168) [40] wo meit ian dei hel p her with her homework.
I everyday have to help her with her ho mework "Everyday I have to help her with her ho mework." (Ch inese/English: Wei, 2005Wei, , p. 2346 In [37] wait for is accessed as a single phrasal verb., an EL island. While in English the object you, the THEM E, is introduced by the preposition, in Chinese the same object is introduced by a single verb deng (wait). In [38] the verb phrase headed by fail is an EL island. While in English fail can be used as a causative verb and this takes the grammat ical subject as the AGENT who makes the failu re happen, in Chinese the equivalent verb shibai means "be defeated in…" and is used only as a noncausative verb with the grammatical subject as the EXPERIENCER. In [39] nonnative speakers, the RECIPIENT, is introduced in the English indirect object dative construction headed by preposition to. By contrast, the equivalent Chinese verb jiao (teach) only permits the double object construction (e.g., jiao ta English (teach him Eng lish)). In [40] her homework , the THEM E, is introduced by the preposition with. By contrast, in Chinese the THEM E is always introduced by a specific verb such as zhuo (do).
Such instances of CS reveal that if speakers select certain particu lar EL verbs at the level of lexical-conceptual structure, but the EL and the ML are incongruent at the level of predicate-argument structure, they may activate the whole verb phrases in the EL and produce them as EL islands .

Conclusion
This study explains CS at some abstract levels of the bilingual speech production process with a focus on bilingual lemma activation as a crucial interface between speaker intention and code activation. It claims that lemmas in the bilingual mental lexicon are language-specific, language-specific lemmas are in contact during a discourse involving CS, and such a contact occurs at three distinct but related levels of abstract lexical structure: lexical-conceptual structure, predicate-argument structure, and morphological realizat ion patterns. Thus, this study regards CS as projection of bilingual lemmas in contact. Based on the linguistic analysis of some naturally occurring CS instances involving various language pairs, this study reaches several conclus ions reading the nature and activity of the bilingual mental lexicon as observed in the bilingual speech productio n process.
(1) While the bilingual speaker's languages are turned "on" during a discourse involving CS, they are not equally activated at the same t ime. Whichever language that the bilingual speaker chooses as the ML is mo re strongly activated than the EL. It is only the ML wh ich provides the sentential frame, controls morphosyntactic procedures, and provides all system morphemes as well as most content morphemes.
(2) The bilingual mental lexicon contains language-specific lemmas, which are in contact in CS. If the speaker chooses the bilingual mode to engage in CS, he/she may activate certain language -specific lemmas as desired for h is/her intended meaning. The EL only supplies content morphemes as desired by the speaker to be switched into the ML sentential frame. This is because only conceptually activated EL lemmas can appear in CS to express the speaker's intended meaning.
(3) The intentional use of the EL system often manifests itself as CS. So me switches are motivated by the lack of particular words in the ML fo r the speaker's co mmunicat ive intentions. The speaker may take co mpensatory strategies to solve lexical problems caused by the lexical gap b etween the languages involved. When the speaker's co mmun icative intention contained in the preverbal me ssage call for a particular EL content morpheme, this selection activates the EL lemma supporting that morpheme.
(4) For a possible CS realization, the activated EL lemmas must be sufficiently congruent with the counterparts of the ML at the three levels of abstract lexical structure or some comb ination of these levels: lexical-conceptual structure, predicate-argu ment structure, and morphological realization patterns. If lemma incongruence or insufficient congruence occurs between the language pairs at any of these levels, but the speaker does not want to g ive up CS, a radical comp ro mising strategy, such as the production of EL islands, must be taken in order for CS to be possible.
As proposed in the MLF Model and the BLA Model, the unequal roles played by the ML and the EL, different aspects of abstract lexical structure, and bilingual lemmas in contact affect CS. This study posits that CS, like any natural speech behavior, is a ru le governed bilingual behavior and regards CS as projection of bilingual lemmas in contact.