2.1 Some features of Distributed Morphology

A note on lexical entries is in order here. Since the implementation of the Minimalist Program (MP) (Chomsky 1995), lexical entries are taken to be an array of features constructed in the generative lexicon and inserted “on-line” in the course of the derivation. In Chomsky’s terms “[a] typical lexical entry consists of phonological matrix and other features, among them the categorial features N, V, and so on; and in the case of Ns, Case and agreement features (person, number, gender), henceforth ϕ-features” (Chomsky 1995: 33). In other words, in a lexicalist approach to word formation as envisioned in the MP, it is the inherent features that allow lexical entries to induce relationships with other lexical entries, and ϕ-features are taken to be part of these presyntactic lexical entries, which are in turn treated as the set of ϕ-features in the syntax.

In line with the standard Y-model of grammar, DM offers an alternative view to the nature of lexical entries. In particular, there are two core components that are crucial for the analysis of bilingual data compatible with the tenets of DM: (i) a lexical entry (or word) is decomposed into a category-neutral root (represented with ‘√’ in this paper) devoid of any grammatical information and a categorizing head (i.e., n, v), that is, words are built in the syntax; (ii) Late Insertion is one of the core tenets of DM that distinguishes it from the lexicalist approach (Harley & Noyer 1999; Arad 2005; Embrick & Noyer 2001; Embick & Marantz 2008; Harley 2014, among many others). In a Late Insertion approach, the properties of traditional lexical entries are distributed across separate components of the grammar divided into three lists, each of which relevant to only a subset of the functions of the lexicon in a lexicalist approach to grammar. One list (the so-called List 1) contains a list of formatives which enter the syntactic computation including grammatical categories (i.e., n, v), functional features (i.e., ϕ-features, [+NEG], [+PAST], [DEF]), and category-neutral roots (√s). These feature bundles are the atomic elements of syntactic computation and are assembled onto hierarchical structures via successive applications of the operation Merge. The resulting syntactic representations, which are fully specified with syntactico-semantic features, are then sent off to the interfaces via the operation Spell-Out. Crucially, the emerging terminal nodes represent a ‘position of exponence’ and must receive some phonological content at PF and semantic information at LF (Harley 2014).

On the PF component of the Y-model, a second list (the so-called List 2) stores Vocabulary Items (phonological matrices). Vocabulary items (henceforth VIs) are correspondence rules for relating syntactic representations (terminal nodes) to phonological representation in context, that is, the pairing of a phonological exponent with grammatical context. VIs are accessed in an operation known as Vocabulary Insertion. Since VIs may be conditionalized to the input’s local environment (i.e., grammatical context), they compete for insertion according to the Subset Principle (Halle 1997; Embick & Noyer 2007), which holds that a VI is inserted in the ‘position of exponence’ if all or a subset of the specified features on the terminal node match the VI’s features.

On the LF side, a third list (the so-called List 3) is accessed. This is the Encyclopedia. Similar to VIs at PF, the Encyclopedia stores a set of correspondence rules for mapping syntactic feature bundles to semantic representations (i.e., information for “special” meanings) in context (Harley 2014). Correspondence rules in the Encyclopedia may be subject to the Subset Principle as with Vocabulary Insertion, but the nature of the Encyclopedia is still unsettled in the literature (see Harley 2014).

2.2 Gender assignment in linguistic theory

The prevailing assumption in linguistic theory is that gender assignment is a lexical representation in the sense that lexical entries (nouns) are assigned gender features via lexical rules in the lexicon (Roca 1989; Harris 1991; Carstens 2000). In particular, Harris developed an account of gender in Spanish and argued that “the formal representation of grammatical gender involves one privative (non-binary) gender mark, … f(eminine)” (Harris 1991: 29). Harris formulated lexical rules to enact the f-feature to nouns. First, he pointed out that Spanish has a declension class system and exhibits clear tendencies between its genders and its declension classes. However, Harris stressed that gender and declension class are two distinct phenomena. AlthoughHarris’s (1991) analysis of gender in Spanish is often cited in the literature on gender assignment in codeswitched speech, there are no formal analyses that have directly applied his lexical rules to the codeswitching facts. On the other hand, Harris’s distinction of gender and declension class has been directly applied in formal analyses to gender in codeswitched speech, although in a misguided way as I explain in Section 5 of this paper.

The structural approach, which traces back toPicallo’s (1991) analysis of gender in Catalan, offers an alternative to the lexicalist approach to gender assignment. In a structural approach, gender features are assigned to nouns on the basis of linguistic properties such as biological sex, animacy, humanness or phonological and morphological correlates (Kramer 2015; 2020). Kramer (2015; 2016) showed howPicallo’s (1991) proposal that gender features project their own phrase in the syntax is unsustainable. As an alternative, inKramer (2015), she pursued the structural approach to gender assignment and developed a comprehensive approach to gender within a DM perspective. In particular, Kramer (2015) proposed that gender features are syntactically located on the nominalizer head n, which is involved in nominalization. It follows that, in a structural approach to gender assignment, gender information is constructed via syntactic merge of a head n with gender features and a category-neutral √ as illustrated in (3).

Moreover, Kramer (2015) delimited a clear distinction between biological gender (i.e., sex) and grammatical gender (arbitrary gender). In particular, she proposed that interpretable syntactic gender features [+/–FEM] capture biological gender whereas an uninterpretable version of the same [+FEM] feature represents grammatical gender in the syntax for a language like Spanish. In her analysis, only interpretable gender features “trigger some kind of effect at LF (e.g., insertion of a denotation) such that their presence/absence (or change in +/– value) changes the interpretation of a linguistic value” (2015: 58). In other words, a nominalized √ [n_[+/–FEM], √] as in (3) is interpreted as male/female-denoting at LF, and semantic licensing conditions can be used to restrict interpretable gender (see Kramer 2015 for more details). The uninterpretable version of the [+FEM] feature, on the other hand, is invisible to LF and, consequently, does not contribute denotations or licenses alternations at LF. In short, the property of (un)interpretability is used as the driving force to distinguish biological gender from grammatical gender inKramer’s (2015) comprehensive analysis of gender assignment. In this paper, I follow Kramer’s proposal of gender features in the grammar and show how the codeswitching facts provide further support for a structural approach to gender assignment.

2.3 Formal approaches to codeswitching

In this section, I provide a brief summary of formal analyses of codeswitching with particular emphasis toMacSwan’s (2000) model of the bilingual grammar. I conclude that a compartmentalized view of the bilingual grammar as impelled by MacSwan’s model is empirical untenable and theoretically undesirable (see also López 2020 for similar criticism).

Formal approaches to codeswitching can be divided into two types. First, there are constraint-based models which invoke specific constraints or rules unique to codeswitching (i.e., Poplack 1980; Myers-Scotton 1993; Belazi, Rubin & Toribio 1994). On the other hand, MacSwan instantiated a research agenda of codeswitching in which “all grammatical relations and operations which are relevant to monolingual language are relevant to bilingual language, and only these” (MacSwan 2000: 43). That is, MacSwan’s model dissented from previous codeswitching analyses which invoked rules and specific constraints. MacSwan’s analysis is grounded in the tenets of the MP (Chomsky 1995) and has inspired important minimalist analyses of gender assignment in codeswitched speech as I explain in Section 2.3.2.

Since the inception of the DM architecture of the grammar (Halle & Marantz 1993), scholars have also pursued a Late Insertion approach to codeswitching. This trend of research examines various bilingual corpora involving not only Spanish and English but also German, Basque, and Norwegian, among other languages (González-Vilbazo & López 2011; Alexiadou et al. 2015; Grimstad et al. 2015; Lillo-Martin et al. 2016; Vergara & López 2017; Grimstad et al. 2018; Burkholder 2018; Riksem 2018; López 2020). The analysis presented in this paper is in line with the DM trend of research. But first I scrutinize the theoretical assumptions of MacSwan’s model of the bilingual architecture and discuss the predictions it makes in terms of gender assignment in codeswitched speech.

4.1 Research questions

In line with the structural approach to gender assignment (Kramer 2015; 2020), the working hypothesis in the present study is that gender features are assigned on the basis of linguistic properties such as biological sex, analogical gender, and/or default genders, but not lexically specified in the bilingual lexicon. Building on the previous findings reported in Section 3, this study addresses the following research questions (RQs):

4.3 Discussion

The findings of the present study can be summarized as follows. With respect to RQ1, the CESA corpus provides support for biological gender as a reliable predictor for gender assignment in codeswitched speech; this is consistent with all previously reported evidence (Poplack et al. 1982; Otheguy & Lapidus 2003; Clegg & Waltermire 2009) but not in line with the few studies that have found negative evidence for biological gender as a reliable predictor in other bilingual communities (Balam 2016; Valdés Kroff 2016). As for RQ2, we found no strong evidence for analogical gender, contrary to the few studies that found some evidence for analogical gender in other bilingual communities (Poplack et al. 1982; Jake et al. 2002). Only a handful of English inanimate nouns that have feminine Spanish equivalents whose phonemic make up strongly correlates with feminine gender are feminine in the CESA corpus; this finding corroboratesDelgado’s (2018) experimental study, which demonstrated that Spanish heritage speakers preferred the singular feminine article la as opposed to the singular masculine article el with some highly frequent English nouns whose Spanish equivalents are feminine. Instead, masculine gender is a prevailing default gender in the CESA corpus; and this is in line with previously reported data across bilingual communities (Dubord 2004; Aaron 2015; Balam 2016; Valdés Kroff 2016; Królikowska et al. 2019, cited in Beatty-Martínez & Dussias 2019). As for RQ3, the distribution of gender assignment in the CESA corpus can best be accounted for by a Late Insertion approach to the grammar. In the next section, I elaborate on this answer and present a theoretical account that best fit the CESA evidence.

5.1. The nature of roots (√) in the bilingual architecture

The nature of √s in the bilingual lexicon requires a careful analysis.⁶ Here, and in line withLópez (2020), I followHarley’s (2014) proposal that √s have no phonological or semantic content in List 1 but must be individuated from one another in the syntax. In other words, syntactic √s are pure units of structural computation free of semantic and phonological content in the syntax. To distinguish one √ from another, I followAcquaviva (2008),Pfau (2009), andHarley (2014) in assuming that √s are identified by non-phonological “index notations” (i.e., √₇₇) that serve “as the linkage between a set of instructions for phonological realization in context and a set of instructions for semantic interpretation in context” (Harley 2014: 226).

In particular, I propose that √s have no language specific information in List 1, and each terminal node emerging from the syntax represents a position of exponence that must receive phonological and semantic content post-syntactically at PF and LF, respectively. More specifically, I suggest that formatives stored in List 1 which enter the syntactic computation—that is, feature bundles including grammatical categories (i.e., n, v), functional features (i.e., [+FEM], [+NEG], [+past], [DEF]), and category-neutral √s—are not duplicated in the bilingual lexicon. Instead, a single √ like [√₇₇] from List 1 has two Vocabulary Items for insertion at PF in the bilingual architecture: [√₇₇] ↔ <mes-a, table>. Consequently, VIs stored in List 2 can compete to realize particular terminal nodes emerging from the syntactic derivation as explained Section 5.3. If √s are devoid of phonological content, it follows that the bilingual architecture does not consist of distinct phonologies as proposed inMacSwan (2000). Rather, there is a single PF module in the bilingual architecture (see also López 2020), and the bilingual speaker has more VIs in her List 2 repertoire to realize a particular emerging terminal from List 1.

5.2 Gender features in the bilingual lexicon

With the exception of several nouns (i.e., persona ‘person.F’, víctima ‘victim.F’, etc.), human nouns are sexed in Spanish, andHarris (1991) demonstrated that biological gender categorically determines gender assignment whereby nouns denoting male referents are masculine and those denoting female referents are feminine. Moreover, as discussed in Section 2.2,Kramer (2015) postulated that interpretable [+/–FEM] gender features encode biological gender in the grammar. Here, I suggest that biological gender (interpreatble gender) is the driving force of gender assignment in Spanish–English codeswitched speech. I explain how.

In Sections 3 of this paper, I showed that there is ample evidence indicating that biological gender almost categorically determines gender assignment with human-denoting nouns across bilingual communities, including the CESA corpus examined in Section 4.⁷ In addition, there is some evidence indicating that bilingual speakers also associate Spanish adjectives with English human nouns, and the grammatical gender of the Spanish adjective is determined on the basis of the biological gender of the noun’s referent as illustrated in (10).

In (10b), the Spanish adjective terca ‘stubborn-F’ refers to a female referent identified in the discourse inPfaff’s (1979) study, and the English noun man triggers masculine gender in (10a). Furthermore, McConnell-Ginet (2013) rightly pointed out that English has a handful of human nouns which are morphologically marked for feminine gender: actress, countess, princess, duchess, empress, abbess, and poetess.

In the case of the Spanish–English bilingual lexicon, I propose that there is only one gender feature, namely [FEM], whose (un)interpretability becomes crucial for interpretation at the semantics system and the assignment mechanism invoked in codeswitched speech. In particular, I follow Kramer’s proposal that only interpretable gender features “trigger some kind of effect at LF [Logical Form, semantics system] (e.g., insertion of a denotation) such that their presence/absence (or change in +/– value) changes the interpretation of a linguistic value” (2015: 58). Similarly, I suggest that an uninterpretable version of the same [+FEM] feature represents feminine grammatical gender in codeswitched speech for agreement purposes only because, unlike interpretable gender, uninterpretable gender is irrelevant for interpretation at the semantics. Importantly, (un)interpretable gender features form part of List 1 in the bilingual architecture, meaning that they are not language specific and/or transferred from one lexicon to the other as it is often proposed in the literature on codeswitching. Although [FEM] differs in interpretability, interpretable [+FEM] and uninterpretable [+FEM] are not independent of each other since they are the same kind of feature in the grammar (Kramer 2015; 2020), but only interpretable gender is relevant for meaning and a reliable predictor for gender assignment.

With these assumptions in place, I propose that the affix -ess of a noun like actress is the exponent of a n_[+FEM] bundle in the context of certain English √s as shown in (11).

In (11), a n combined with interpretable [+FEM] represents the femaleness of certain English √s for purposes of interpretation at LF. According to the Vocabulary Insertion rule in (11), the n_[+FEM] bundle is morphologically exponed as /ess/ at PF with the English √s listed in (11). The logical question at this point is whether interpretable gender represented via the syntactic [+/–FEM] feature is visible to the head D in an English DP, (i.e., the girl). I sustain that it is for the following reasons. First, the syntactic [+/–FEM] feature triggers agreement with pronouns in English, which in turn suggests that English has grammatical gender if pronouns count as elements that agree. Secondly, the syntactic [+/–FEM] feature is invoked in the spell-out form of the Det in Spanish Det–English noun switches with human-denoting nouns as demonstrated throughout this paper. Crucially, note that interpretable gender does not pose a threat to the interfaces because it is legible at LF.

Gender assignment with English nouns that lack biological gender occurring in codeswitched speech raises intriguing questions about the nature of gender features and the assignment mechanism. Analogical gender as discussed in Section 3 is a possible explanation for gender assignment with inanimate nouns in codeswitched speech. However, the results of the CESA corpus combined with previous studies indicate that analogical gender is not a viable predictor for gender assignment in codeswitched speech since feminine gender is limited to a handful of English inanimate nouns, and masculine gender seems to be the default gender rather than the result of analogical gender. Here, I propose that the uninterpretable version of the [+FEM] feature represents feminine analogical gender in codeswitched speech for purpose of syntactic agreement. Importantly, note that the presence of uninterpretable [+FEM] in codeswitched speech does not imply “transferring” feminine gender from the gendered lexicon as proposed inLiceras et al. (2008). In my view, it is an epistemic matter whether a speaker invokes the gender of the Spanish equivalent. If so, I suggest that the concept of analogical gender can be attributed to the Encyclopedia as I explain in the next section—where I also address masculine default gender in codeswitched speech.

5.3 The analysis

Table 7 illustrates the distribution of gender with human-denoting nouns in the Spanish–English bilingual lexicon. First, note that both Spanish and English have different-root nouns with lexico-semantic gender and epicene nouns whose gender is context-dependent. However, only Spanish has post-stem vowels that alternate according to the sex of the noun’s referent: -a for female referents and -o for male referents. The analysis of gender assignment in codeswitched speech proposed in this paper is primarily based on the distribution of gender in Table 7.

Let us first see how we can combine human-denoting nouns that have lexico-semantic gender with those that exhibit vowel alternation in Table 7. Following the premises of the single-lexicon approach outlined in Sections 5.1 and 5.2, I propose that one and the same √ (i.e., [√₇₅]) has two phonological matrices stored in List 2 of the bilingual architecture, whereby √s are paired with a specific type of n in the syntax. Consider the examples in (12).

The examples in (12) illustrate that a √ merged with n_[+FEM] is interpreted as female- and one merged with n_[-FEM] is interpreted as male-denoting, respectively, at LF. Consequently, the emerging √ combined with a type of n represents a position of exponence and must receive phonological content at PF via contextual restrictions.

According to the Vocabulary Insertion rules in (12), phonological matrices are inserted in the position of a syntactic terminal in the context of a bundled n. Given the VIs available for exponence in (12), I must say something about how language exponents from List 2 can compete for insertion at PF when codeswitching. I suggest that morphophonological information guides the bilingual speaker to the phonological realization of a target terminal emerging from the syntax. This information must be part of the linguistic competence of the bilingual speaker since Spanish and English exhibit different morphophonological properties. For instance, Oltra-Massuet & Arregi (2005) showed that Spanish has a morphological well-formedness condition which holds that all syntactic functional heads require a theme node (TH) position at the morphological module in the derivation. In the case of nouns, a TH node is post-syntactically adjoined to the nominalizer n and corresponds to nominal declension classes (I, II, III) whose realization depends on three theme vowels (-o, -a, -e/Ø).

When codeswitching, I propose that a Spanish phonological matrix like /niñ-/ in (12) is inserted in the context of a n hosting a TH position as illustrated in (13).

In (13), the TH node is purely morphological and hosts class features (I, II, III) which are in turn morphologically realized as theme vowels via Vocabulary Insertion (see Oltra-Massuet & Arregi 2005; Bermúdez-Otero 2013; Kramer 2015). But how can the TH position guide the competition of phonological matrices from List 2 when codeswitching?

The TH position is of crucial relevance for the computation of stress in Spanish: it determines stress placement with verbs and non-verbal environments (Oltra-Massuet & Arregi 2005). In other words, the TH position adjoined to n in (13) requires that the phonological matrix inserted in such context obeys the phonotactics of Spanish in order for a Spanish word to be constructed. The phonological matrices /girl/ and /boy/ in (12) cannot be inserted in (13) because they do not obey the requirements imposed by the TH position adjoined to n. Again, the reference to the internal syntactic structure of nouns as illustrated in (13) is part of the bilingual’s linguistic competence. Later, in Section 5.4, I will provide further support on the crucial relevance of a TH position in the bilingual architecture.

Adopting a probe-goal relation for agreement (Chomsky 2000; 2001), I assume that the head D enters the derivation with an unvalued gender feature ([GEN:_]) in codeswitched speech. The unvalued feature probes (dotted line) in (14) and finds a valued (and interpretable) gender feature on the head n in both instances in (14); gender and number are independent probes in (14).

Valuation occurs via Agree: the probe acquires the [+FEM] feature in (14a) and the [–FEM] in (14b) (solid line). It follows that a valued [+/–FEM] feature is acquired by the bundle head D for morphological realization at PF. An interpretable [+/–FEM] gender feature does not pose a threat at the interfaces, and I speculate that the unvalued [GEN:_] feature of the head D is invisible when gender is irrelevant for syntactic agreement (i.e., in the context of English mode).

Vocabulary Items for the Det system in the Spanish–English bilingual lexicon are presented in Table 8. The Spanish Det system has a contrastive distinction for definiteness, number, and gender. I assume that the bilingual speaker has access to these contrastive features.

In the analysis in (14), an interpretable [+/–FEM] feature is acquired by the head D as the result of a probe-goal relation. Now consider the VIs of the Det system in (15).

Vocabulary Insertion in (15) obeys the Subset Principle (Halle 1997; also Embick & Noyer 2007), which holds that given the set of VIs competing for insertion into a given terminal with a set of morphosyntactic features, only the one with more features in common with the target node without having any more will win the competition and be inserted. Note that the VI for the feminine Det la in (15a) contains a [+FEM] morphosyntactic feature, and the head D emerging from the syntax in (14a) has the same feature inducing the VI in (15a). The head D with a [–FEM] feature is realized with el in (15b) because la in (15a) has a non-matching feature; in other words, the article el is the underspecified form (see also Riksem 2018 for the application of the Subset Principle to language mixing).

Although inanimate nouns lack biological gender, English inanimate nouns are also assigned grammatical gender in codeswitched speech as illustrated in (16).

The analysis of the CESA corpus reported in Section 4 revealed that English inanimate nouns are overwhelmingly masculine (16b) in codeswitched speech. We also saw that only a handful (n = 29) of inanimate nouns whose Spanish equivalents are feminine (16a) and exhibit phonemic make up that strongly correlates with feminine gender are feminine in the CESA corpus.

As pointed out in Section 5.2, I propose that one and the same feature, namely [FEM], accounts for gender assignment with both human nouns and unsexed entities in the Spanish–English bilingual lexicon. In particular, I suggest that an uninterpretable version of the [+FEM] feature captures feminine grammatical gender with inanimate nouns. An important question at this point is how do we prevent interpretable gender from merging with non-human roots. In DM, Encyclopedia entries are pairings of √s and categorizing heads (Arad 2005). This opens the possibility that some licensing condition invoked at the Encyclopedia ensures that human-denoting √s are only interpretable in the context of a n with interpretable gender. Again, this is possible because √s combine with different flavors of ns, so that √s denoting unsexed entities are prohibited with interpretable gender at the Encyclopedia because interpretable [+/–FEM] gender features are incapable of finding a referent in the world for interpretation. Similarly, the Encyclopedia also ensures that the √ for a human noun like mother is incompatible with a n with interpretable [–FEM]. This analysis captures the fact that grammatical feminine gender is restricted to a handful of English nouns in codeswitched speech, presumably because gender agreement is easier with frequent English nouns that have feminine equivalents and there is no need to adopt a default gender. Future work is needed in this direction, but attributing the (un)interpretability of [+FEM] to a licensing condition applied at the Encyclopedia is suitable with the single lexicon approach of the bilingual architecture.

A a probe-goal relation as in (14) also applies in the context of n with uninterpretable [+FEM], and the bundle D hosting the English noun in (16) spells out as la with inanimate nouns like la leash according to the rules in (15a). In other words, the same morphological realization applies for the (un)interpretable [+FEM] feature.

English inanimate nouns are overwhelmingly masculine in the CESA corpus. In fact, masculine gender is a prevailing default gender across Spanish–English bilingual communities as reported in Section 3. Masculine default gender must be amenable in the bilingual architecture. In particular, I suggest that these inanimate nouns lack gender features altogether. In other words, their √s combine with a n lacking gender features, a plain n. In this case, the unvalued [GEN:_] of the head D in (14a) does not find its valued counterpart because the potential goal (plain n) lacks gender features. I assume thatPreminger’s (2009) failed Agree applies in this case: the agreement operation applies as normal, but D’s unvalued feature is ‘reset’ to a default value as part of sending the syntactic output to the interfaces. It follows that the bundle D arrives to PF with an unvalued feature (failed probe), which is in turn incompatible with la in (15a), which has a valued [+FEM] feature. Therefore, the underspecified form el of the Det system in (15b) is inserted with a failed probe.

A note on previous accounts to gender assignment in codeswitched speech framed within Late Insertion is in order here. At first glance, the analysis in (13) seems to resembleBurkholder’s (2018) proposal of gender assignment in Spanish Det–English noun switches, but it is fundamentally different. Burkholder (2018) argued that a Spanish–English bilingual speaker has two List 1s (and two List 2s) in the (duplicated) bilingual lexicon, and that a binary uninterpretable [±f] feature represents gender in codeswitched speech, namely because her analysis is based on grammatical gender in Spanish, that is, biological gender is not considered. Burkholder further proposed that a bilingual speaker can merge a √ like ‘√house’ in la house with a Spanish n hosting uninterpretable [+f], or uninterpretable [–f] for ‘√book’ in el book. It follows that a Spanish n_[+f] taken from the Spanish lexicon combined with ‘√house’ can trigger a theme node (TH) at the morphological module, but such TH position must be realized with a null element /Ø/ at PF with a √ like ‘√house’ because English lacks theme vowels. In other words, Burkholder collapsed gender and declension class in the sense that the presence of the [±f] feature (the result of gender transfer) triggers a TH position at the morphological module, imposing gender and declension class information onto English nouns in codeswitched speech.

López (2020) analyzed gender assignment as a key feature of his integrated model of the bilingual architecture. He postulated a [±f] gender feature in the bilingual lexicon. Assuming that √s are devoid of any grammatical information and phonological content as proposed in this paper, López (2020) suggested that a single √ represents different phonological matrices in List 2. As for gender assignment in codeswitched speech as in la key, López suggested that the English noun key can spell out the complement of n_[+f] in analogy to its Spanish feminine counterpart llave. In other words, the Spanish–English bilingual speaker has two phonological matrixes for spelling out of the concept key, {/key/, /llav-/}, as the complement of a single n_[+F]. López also proposed that word markers (i.e., /a/, /e/, /Ø/, etc.) spell out n_[+F], so that /e/ spells out n_[+F] in la llave and the null element /Ø/ spells out n_[+F] for the codeswitch la key.

Although they take different positions on the nature of a theme (TH) position, Burkholder (2018) andLópez (2020) proposed a binary [±f] feature to account for gender assignment in codeswitched speech.⁸ However, their analyses do not capture the distribution of gender assignment in codeswitched speech; that is, the empirical fact that biological gender almost categorically determines gender assignment, whereas feminine grammatical gender is restricted in codeswitched speech as demonstrated in this paper, and first pointed out inDelgado (2018). Moreover, if the bilingual speaker has two phonological matrixes for a single √ as proposed inLópez (2020), it is unclear how the insertion of language exponents is accomplished at PF when codeswitching, and Burkholder’s proposal that an English noun occurring in codeswitched speech triggers a TH position is unwarranted.

In this paper, I outlined an alternative analysis to gender assignment in bilingual speech. The analysis can be summarized as follows: (i) gender information is constructed from combining a category-neutral √ and the nominalizer n in the syntax, whereby gender features are located on n; (ii) the bilingual architecture consists of a single List 1, a single List 2, and a single List 3 and the Encyclopedia can handle issues of feature interpretability; (iii) the single (un)interpretable feature [FEM] captures gender assignment with human and non-human referent nouns, and biological gender (interpretable gender) plays a crucial role in the assignment mechanism, in line with typological data demonstrating that biological gender and grammatical gender are not independent of each across the world languages (Kramer 2020); (iv) the competition of phonological matrices from List 2 when codeswitching makes crucial reference to the internal syntactic structure of words, namely the TH position; (v) gender and declension class are distinct phenomena, but they are in close interaction as first noted inHarris (1991). In the next section, I say more about this last point.

5.4 Further support for the crucial role of the theme node position at PF

I believe there is further evidence indicating that the analysis to gender assignment in codeswitched speech developed in this paper is on the right track. More specifically, there is a handful of novel nouns that are fully integrated in the bilingual lexicon obeying the morphosyntax of Spanish. They are illustrated in Table 9 (see Barkin 1980; Poplack et al. 1982; Smead 2000). The novel nouns troca and breca in Table 9, whose etymology traces to English truck and break, respectively, are well-cited examples that serve as a way to illustrate the role of the TH position in the incorporation of such novel nouns in the bilingual lexicon.

Note that the nouns in Table 9 are feminine and end in the post-stem vowel -a, the marked form of the inner core inHarris’s (1991) analysis of word classes in Spanish. In the analysis developed in Section 5.3, I proposed that a √ representing an English noun like table and its Spanish equivalent mesa can be merged with an uninterpretable [+FEM] feature in the syntax for syntactic agreement purposes in codeswitched speech. I also suggested that a TH node is adjoined to the nominalizer n at the morphological module when Spanish phonotactics are required at PF, namely for computing stress in Spanish. The data in Table 9 aim to demonstrate the crucial role the TH node, and the morphological module more generally, plays in the bilingual architecture.

In particular, I speculate that the nouns in Table 9 underwent a phonological change in order to be incorporated into the bilingual lexicon so that they can function as independent phonological matrices in List 2 of the bilingual architecture. This is evident by the fact that the presence of the TH node with these nouns induced stress alignment so that these nouns can obey the phonotactics of Spanish as schematized in (17).

In (17), we can see that truck and break underwent a vocalic change, and the presence of the TH node, which is in turn spelled-out with the word marker /a/, shifted the stress pattern to the vowel preceding the TH node: /trók-a/ and /brék-a/. Again, the TH position plays a crucial role for computing stress in Spanish and for the competition of language exponents when codeswitching. In this way, the phonological matrix <trók-> competes for insertion with its counterpart <trʌk>, the former being inserted in the context of a n bearing a TH node.⁹

Note again that the novel nouns in Table 9 exhibit the marked form ofHarris’s (1991) inner core of word markers in Spanish, namely -a. Moreover, the Spanish equivalents of break and truck are actually masculine, and so analogical gender does not apply in the case breca ‘break’ and troca ‘truck’. The data in Table 9 then suggest a tight relation between the marked form -a of the inner core and the gender feature [+FEM] in the bilingual grammar in the sense that all the feminine nouns in Table 9 exhibit the word marker -a. As far as I am aware of, there is no evidence indicating that English nouns incorporated in the bilingual lexicon trigger the default word marker -o even when these nouns are overwhelmingly defaulted to masculine gender (i.e., el club, el bar, el swing, etc.).¹⁰ In short, the analysis in (17) points to the tight relation between the syntax and the morphological module of the bilingual grammar as predicted by the syntactic analysis to gender assignment outlined in this paper. I hope future studies will explore such relation of the modules in connection to the codeswitching facts (see, for example, Grimstad 2017 for a Late Insertion approach to English-origin verbs in American Norwegian). In this paper, I attempted to provide an analysis to gender assignment in codeswitched speech within a DM perspective, but gender is only one of many linguistic phenomena that are currently being debated in the literature (see López 2020 for a relevant discussion).