1 Introduction

This paper is an attempt at analyzing the puzzling exponence of Gender in a class of broken plurals in Omani Mehri, a Modern South Arabian language spoken in Western Oman.1 In typical Semitic fashion, the nouns of Mehri are either masculine or feminine. While the gender of Mehri nouns can be unambiguously assessed on the basis of the agreement patterns they trigger, Gender exponence on the nouns themselves constitutes a serious analytical challenge.

Omani Mehri exhibits a rich variety of types of plural nouns (Johnstone 1975; Lonnet 1993; Simeone-Senelle 1997; Rubin 2010; Watson 2012). The challenge taken up in this paper involves the analysis of one particular class, the class exemplified in (1). The plurals in (1) are broken plurals, i.e. the edges of the root match the edges of the word. A specific feature of the class of plurals in (1) is the presence of an extraneous glide y or w (boldface in (1c, d)) either between the second and the third radical consonant (1i, ii) or between the first and the second radical consonant (1iii, iv). No such glide appears in the corresponding singulars. In (1), each plural appears in two versions. The version in (1c) is a rough phonetic representation, the form under which words are entered in Johnstone’s Mehri Lexicon (1987). The version in (1d), by contrast, is more abstract. It does not include the effects of two regular processes, viz. the lengthening of /a/ into [ṓ] under stress and the mapping of diphthong /əy/ into long [ī] (see Bendjaballah & Ségéral 2017 for discussion). For the sake of clarity, the discussion will be mostly conducted on the basis of the more abstract forms in (1d).

(1)   a. Root b. Singular c. Plural
[phonetic]
d. Plural
/abstract/
e. Gender Gloss
  i. √rħb [rəħbḗt] [rəħṓyəb] /rəħayəb/ Feminine ‘town’
  ii. √χtm [χṓtəm] [χətṓwəm] /χətawəm/ Masculine ‘ring’
  iii. √ħrs [ħársət] [ħəwṓrəs] /ħəwarəs/ Feminine ‘camel not needing water’
  iv. √brk [bárk] [bīrṓk] /bəyrak/ Masculine ‘knee’

In (1e) I have indicated the gender of the nouns. The source of that information is the type of agreement each noun triggers. As an example, the adjective for ‘small’ comes in four different forms: k’ənnə́t (feminine singular), k’ənnə́tən (feminine plural), k’ənnáwn (masculine singular), k’ənyáwn (masculine plural). In (2), I show how the gender of the nouns for ‘√rħb, town’ (1i) and ‘√χtm, ring’ (1ii) – both in the singular and the plural – can be identified based on the form of the adjective they select.

    1. (2)
    1. a.
    1. rəħbḗt
    2. √rħb, town
    1. k’ənnə́t
    2. small.FEM.SG
    1. ‘small town’
    1.  
    1. b.
    1. rəħṓyəb
    2. √rħb, town
    1. k’ənnə́tən
    2. small.FEM.PL
    1. ‘small towns’
    1.  
    1. c.
    1. χṓtəm
    2. √χtm, ring
    1. k’ənnáwn
    2. small.MASC.SG
    1. ‘small ring’
    1.  
    1. c.
    1. χətṓwəm
    2. √χtm, ring
    1. k’ənyáwn
    2. small.MASC.PL
    1. ‘small rings’

Because plural nouns in (1) are either masculine or feminine and the glide can be located in one of two different places, the four subclasses in (3) can be defined, each of which is exemplified by one item in (1d).2 In (3), I indicate the number of items documenting each subclass in Johnstone’s Mehri Lexicon.3

(3) a. Feminine plural nouns with glide y between radicals C2 and C3 25
  b. Masculine plural nouns with glide w between radicals C2 and C3 50
  c. Feminine plural nouns with glide w between radicals C1 and C2 10
  d. Masculine plural nouns with glide y between radicals C1 and C2 7

Obviously, both extraneous glides are involved in feminine and masculine plurals of the type shown in (1). The main goal of this paper is consequently to unravel the identity of those extraneous glides and the principles governing their distribution.

In a second section, I identify the analytical challenges inherent in the data just reviewed. In a third section, I discuss the theoretical tools that make the assessment of the evidence possible. In section 4, I show why the sites and the segmental substance involved in Gender exponence appear to be variable. This is followed by a proposal in which the canonical site and the genuine exponents of Gender are defined. On that basis, I reject polarity as a phenomenon involved in Gender exponence in the plurals under study. In section 5, a residual pattern that seems not to conform to any of the generalizations laid out throughout the paper is discussed. Finally, I wrap up with concluding remarks.

2 Assessing the behavior of glides

2.1 What the extraneous glide stands for

Because the extra glides in the plural nouns reviewed in section 1 are conspicuously absent from the corresponding singulars, a possible conjecture would be that they expone Plural. For that conjecture to be the null hypothesis, it would have to be the case that the plurals in (1) are derivationally related to the corresponding singulars. A classic scenario could then be contemplated: plural formation takes the singular as its input, and realizes Plural in the form of the glide(s) in (1d). This is the familiar reasoning which leads to the conclusion that s is a plural exponent in English cats. Typically in such cases, the singular can be identified as a subset of the plural, [[cat]s].

However, a comparable scenario is highly implausible in the case of the Mehri data in (1), as I argue directly. The argument is more fully developed in Fathi (2017) and can only be sketched out here. The set of almost 100 singular nouns, masculine as well as feminine, of which the plural is of the type in (1d) is a highly heterogeneous group as regards the number of vowels they involve, the respective distribution of the full and the reduced vowels, the nature of full vowels themselves, the presence or absence of affixes, etc. The sample in (1b) only weakly reflects that variety. Nevertheless, it is significant that none of the four items in (1b) displays the same profile: CəCCḗt, CṓCəC, CáCC, CáCCət. No straightforward morphological process can cause members of such a motley group to converge into the unique and rigid pattern of their corresponding plurals. Instead, I submit, the plural pattern in (1) is an independent morphological formation devoted to the expression of nominal plurality. Its only connection with a singular is the root it shares with it. The view that a plural noun can be derived independently of a matching singular noun is nothing new. Indeed, such cases are typical of Afroasiatic (see e.g. Ségéral 1995).

On the view that the plurals in (1) are derivationally related to their singulars, it makes perfect sense to suppose that the extra glide expones the result of the derivation, viz. Plural. On the other hand, if the pattern in (1) is endowed with an architecture of its own, it can no longer be viewed as the null hypothesis that its glides are strict Plural exponents. Indeed, if these plurals owe nothing to their corresponding singulars, their glides are just as likely to be the exponents of any one, or several, of the morphosyntactic features carried by those plurals. It could well be that glides expone Gender, Plural or syncretically both Gender and Plural.

In the remainder of this paper, I will defend the view that the glides exclusively expone Gender and I will offer an alternative proposal for the realization of Plural.

There is a compelling reason to pursue that hypothesis. If Plural was indifferently marked by w or y, a logical possibility, the choice of which glides sits in any particular item should be a lexical, arbitrary feature of that item: as long as the item is plural, it should be as likely to include w or y. To put it differently, it would not be expected that the presence of one glide as opposed to the other correlates in systematic fashion with another property of the item. Yet, this is precisely what the evidence shows: the glide co-varies with Gender.

To see this, the two subcases in (4) must be distinguished depending on the location of the glide.

(4) a. The glide appears in the third position from the left (C1C2GC3)
  b. The glide appears in the second position from the left (C1GC2C3)

In the first subcase, all items displaying w are masculine and all items displaying y are feminine, e.g. masculine χətawəm vs. feminine rəħayəb.4 Such a regular correlation with Gender is unexpected on the view that the glide merely marks Plural. A similar regularity holds in the second subcase, albeit in reverse fashion: all items displaying y in the second position from the left are masculine while all items displaying w in the same place are feminine, e.g. masculine bəyrak vs. feminine ħəwarəs. Both times, a correlation can be established between the nature of the glide and the gender of the noun. And, in addition, those two generalizations are not independent of each other. Rather, each subcase can be defined by reference to, indeed as the mirror image of, the other. The question, therefore, is not whether w and y are involved in Gender exponence or not, but how? This is the first of the two main issues which form the concern of this paper.

The second issue is of a more theoretical nature. I assume that Gender is represented once in the morphosyntactic architecture of the class of plural nouns under scrutiny. Why then is its exponence distributed over the two different sites in (4)? That question links up with a second one.

We just saw that exponence in one of the two sites in (4) is the mirror image of the other. The situation is recapitulated in (5).

(5) C1C2GC3 C1GC2C3
  w realizes Masculine y realizes Masculine
  y realizes Feminine w realizes Feminine

Two views of (5) are possible. On one view, polarity is the organizing principle of bilocational exponence and both sites are reciprocally the mirror image of each other. Indeed, the situation depicted in (5) appears to correspond to the definition of polarity given in Wunderlich (2012: 160): “Polarity is a type of syncretism where the syncretic items are distributed along the diagonals rather than the rows of a paradigm.”5

I will argue for another view according to which polarity is spurious, and an asymmetrical relationship underlies the mirror image effect. Only one of the sites hosts legitimate exponence, the other is an artefactual reflection of the former. In that case, the status of each of the two sites will have to be determined.

In order to provide perspective on exponent stability, both as regards segmental substance and location within the representation, I propose to review three cases relevant to the forthcoming discussion.

2.2 Three cases of exponence

The most straightforward case of exponence can probably be described in terms of the following two properties. First, the segmental material involved in exponence is uniquely defined. That is, its possible allophonic variations can be traced back to a unique source. Second, its location in the overall phonological representation of the item exactly matches the position in which the relevant morphosyntactic position has been linearized. The regular plural of English nouns illustrates that case. On the view that its morphosyntactic position is as in (6a) and that it is linearized as in (6b),6 /z/ the designated phonological object realizing the English nominal plural will regularly appear in exactly the position specified in (6b), i.e. to the right of the site where the noun’s phonological matrix is realized; to wit the examples in (6c).

(6) a. b. c.
  N-PL /kæt/-/z/   [khæts] ‘cats’
    /rōd/-/z/   [roʊdz] ‘roads’
    /buʃ/-/z/   [bʊʃız] ‘bushes’

However, as is well known, the phonetic realization of syntactic features does not always match their morphosyntactic position. Consider in this respect a sample of Chaha second person feminine singular imperatives (see Lowenstamm 2000a for discussion). In (7a), I have indicated where the subject agreement markers are linearized with respect to the verbal stem. Several examples of second person masculine singular imperatives (not audibly marked for Person, Number, or Gender) are given in (7c) for control. The feminine forms themselves appear in (7d). While all instances of second person feminine singular marking in (7d) can be traced back to a high front vocoïd, their location and mode of realization vary dramatically; from top to bottom in (7d): as palatalization of the third radical consonant, as a vowel between the second and third radical consonants, as palatalization of the first radical consonant, as a vowel between the first and second radical consonants (along with attendant centralization of the low vowel in the latter two cases), and finally as palatalization of both instances of the second radical consonant.

(7) a. Subject agreement markers7 b. Root8 c. Masculine imperatives d. Feminine imperatives9
  Stem-Person-Num-Gen √kft kıft ‘open!’ kıfty
    √Arb ärıβ ‘milk!’ äriβ
    √gfA gıfa ‘push!’ gyıfä
    √smA sıma ‘listen!’ simä
    √sd sıdıd ‘drive the cattle!’ sıdyıdy

Evidently, the identity of the segmental material exponing second person feminine singular is unambiguous, but the site and mode of its ultimate realization is highly variable.

More frequent in Afroasiatic (including Chaha) is the converse of the configuration just reviewed: the location of the exponent is entirely stable (unlike Chaha second person feminine singular), but its segmental manifestation – again, unlike Chaha – cannot be uniquely defined. The illustrative data comes from Classical Arabic. Consider the verbs in (8b) and their corresponding causatives in (8c).

(8) a. Root b. Underived c. Causative
  √brz baraza ‘he stood out’ barraza ‘he caused X to stand out’
  √bṭl baṭala ‘it.MASC became null’ baṭṭala ‘he nullified X’
  √ʒmd ʒamada ‘it.MASC froze’ ʒammada ‘he froze X’
  √md madda ‘he stretched’ maddada ‘he caused X to stretch’

Following Guerssel & Lowenstamm (1990), I view the output of causative formation as delivering a structure where a causative position is linearized as in (9) inside the verbal template.

(9) [V CV [Caus CV] CV CV]

In a first step, the root consonants are associated to the positions other than causative. This is illustrated in (10a) with the example of baraza ‘he stood out’. Then, causative formation proceeds as indicated by the leftward arrow in (10b).

(10) a. b.
 

Suppose we now ask what the exponent of causative is. Clearly, no particular consonant can be singled out. Rather, the backward spreading of any consonant initially associated to the position in boldface in (10b) will contribute to the formation of a possible causative verb in Classical Arabic, r in the case of √brz, in the case of √bṭl, etc. That is, mere phonological identification of a dedicated position in the manner just described counts as exponence.10 In the next section, the deployment of the entire root will be interpreted in light of that proposal.

Gender exponence in the class of Mehri plural nouns in (1) differs significantly from the stable type exemplified by English regular plural. Indeed, it appears to combine features of the two other types just reviewed: non-unique location as in Chaha and non-unique segmental substance as in Arabic. Beyond these observations, the evidence in (1) remains rather opaque as to Gender exponence. I propose to deal with it in roundabout fashion, namely by focusing on the morphosyntactic structure presumably expressed by such nouns. In the next section, I speculate on the functional structure of the class of plurals under discussion. Then, I show how those tools make an assessment of the evidence possible and how in consequence my hypotheses can be refined.

3 The architecture of plural nouns and a preliminary assessment of the evidence

Consider a Mehri feminine noun such as bərkḗt /bərkē-t.FEM/ ‘√brk, pool, cistern’ and its corresponding plural bəráktən /bərak-t.FEM-ən.PL/. In these nouns, Gender and Number features are presumably located outside the stem, i.e. exponed in concatenative fashion.11 This is in contrast with a feminine plural noun such as rəħṓyəb /rəħayəb/. Indeed, the characteristic feature of the profile of broken plurals to which rəħayəb belongs is the alignment of the edges of the root with the edges of the word. That is, the entire word is comprised between the first and the last root consonants with no discernible affixes as represented in (11).

(11) [ω             ]
  [√C….….C]

However, broken plurals do carry Gender and Number features just like concatenative plurals. Indeed, both rəħayəb and bəraktən similarly trigger feminine plural agreement on an adjective, (12).

    1. (12)
    1. a.
    1. rəħṓyəb
    2. √rħb, town
    1. k’ənnə́tən
    2. small.FEM.PL
    1. ‘small towns’
    1.  
    1. b.
    1. bəráktən
    2. √brk, cistern
    1. k’ənnə́tən
    2. small.FEM.PL
    1. ‘small cisterns’

The question therefore arises as to where the Gender and Number features are located (and ultimately realized) in the case of broken plurals.

I will address this issue on the basis of the hierarchy in (13), a working hypothesis as to what the relevant ingredients are and how UG likely organizes them.

(13) Num ≥ Gen ≥ n > √

In light of a proposal due to van Riemsdijk (1990), and developed in Borer (2005) and work in Distributed Morphology (e.g. Harley & Noyer 1999), I assume without further discussion that the lexicon exclusively involves a list of roots (√). Roots enter the syntactic component upon selection by a categorial head n, a, v (for noun, adjective, verb, respectively). Consequently, the relationship between the two rightmost objects in (13) is unambiguously one of domination of the root by categorial head n.

By contrast, there is considerable discussion as to the nature and location of the representation of Gender (Gen) with respect to both the categorial head (n) and the representation of Number (Num). Some have proposed that Gen is part of NumP (Ritter 1991; 1993); others maintain that Gen is part of nP12 (Alexiadou 2004; Lowenstamm 2008; Kramer 2015; 2016; Fathi & Lowenstamm 2016); finally, it has been proposed that Gen is independent of both nP and NumP and occupies an intermediate position such that it unambiguously dominates the former and is unambiguously dominated by the latter (Picallo 1991; 2008; Bernstein 1993).

The hard core of these proposals is reflected in the formulation in (13) – consistent with Greenberg’s universals 36, 37, possibly 30 (Greenberg 1963) – whereby Gen occupies a position not lower than n and not higher than Num.

Given (13), I suggest considering the Spanish feminine plural noun, cerezas/cerez-a.FEM-s.PL/ ‘cherries’. The structure of cerezas can be construed as involving the following ingredients: an uncategorized root √CEREZ, a categorial head n, a representation of Gender, and finally a representation of Number. As shown in (14),13 these ingredients are organized in a manner consistent with (13).

    1. (14)

As a result of the decomposition illustrated in (14), the root has been singled out as the bottommost terminal. It appears associated with the richly articulated representation of a feminine plural noun. When a Semitic root is similarly factored out, the associated structure is usually taken to be the template in (15b).

My claim is that the information present in the boxed portion of (15a) – the categorial identity, the Gender and Number features – must be packed in the boxed portion of the representation of (15b), i.e. in the template.

(15) a. cerezas b. rəħayəb
 

Here, my proposal meets a program independently developed in recent work (Lowenstamm 2005; Arbaoui 2010a; b; Faust 2011; Bendjaballah 2012; 2014). The main tenet of that program is that the characteristic templates so typical of Semitic and other Afroasiatic languages are not primitive grammatical objects. Their presence in certain languages and their absence in others is not an irreducible typological feature. Rather, they are special modes of realization of universal structures. I side with these authors and I will show how this helps in construing the Mehri evidence.

For purposes of the ongoing discussion, I go a step further and represent Gender as heading its own projection (16a).

Upon head Movement and Left-adjunction,14 the complex head in (16b) is derived.

(16) a. b.
 

Under the hypothesis that each of these functional heads spells out as a light CV syllable (Lowenstamm 2005; Scheer 2012) as shown in (17), the template in (18) is eventually derived. Its rigidity follows from the further hypothesis that all ingredients in (16) are obligatory.15

    1. (17)
(18) [ CV] [n CV] [Gen CV] [Num CV]

Under the proposal just laid out, the evidence can now be construed with some precision on three counts.

First, on the view that the template is as in (18), repeated in (19), the Plural exponent is designated as the identification of the relevant site by a root consonant (much as was the case with causative exponence in Classical Arabic in (10)).

Because the extra glides do not appear in that position, I conclude that they play no role in the exponence of Plural.

    1. (19)

The second point on which some light is shed has to do with the two places where the glide can appear in a plural noun, i.e. between the first and second radical consonants (ħəwarəs/bəyrak) or between the second and third radical consonants (rəħayəb/χətawəm).

In my proposal, the two positions where glides appear have quite a different status: the position between the second and third radical is the designated position for the expression of Gender as per its label in (18). I dub it the canonical position. In consequence, the position between the first and second radical is ruled non-canonical. All four cases are represented in (20) and (21) in which vowel a, to the best of my knowledge, is nothing but the “vowel of the word.”

(20) a. Masculine χətawəm
(Canonical position)
b. Feminine rəħayəb
(Canonical position)
 
(21) a. Feminine ħəwarəs
(Non-canonical position)
b. Masculine beyrak
(Non-canonical position)
 

Thirdly, it follows that w in masculine χətawəm and y in feminine rəħayəb being located in the canonical position ([Gen CV]) must be viewed as manifesting the canonical form of the Masculine and Feminine exponents, respectively.

My identification of canonical exponence, both as regard its locus and substance, receives independent support from two quarters.

The first source is comparative evidence: throughout Semitic, u/w is associated with Masculine exponence and i/y with Feminine exponence. This can be seen from the sample in (22).16

(22)     ‘he’ ‘she’
  a. Classical Arabic huwa hiya
  b. Hebrew
  c. Ge’ez wıʔıtu yıʔıti
  d. Akkadian ʃū ʃī

The second source is the imbalance recorded in (3) and mentioned in footnote 3: the numerical superiority of plurals displaying a glide in third position is consistent with the idea that such is the canonical position for Gender exponence.

A final element militates in favor of further exploring the consequences of my proposal, namely the prospect of elucidating the nature of non-canonical exponence, possibly exposing it as being artefactual, and eventually proposing an account in which polarity plays no role.

This is carried out in the next section, which breaks down into three parts. First, I focus on the content of canonical Gender exponence (subsection 4.1). Then, I ask why Gender exponence is apparently possible in more than one place (subsection 4.2). I finally offer my view of what non-canonical exponence is and how it arises (subsection 4.3).

4 The proposal

4.1 The content of canonical Gender exponence

All Mehri nouns are capable of causing a target of agreement to vary in gender and number in the manner illustrated earlier in (2). This is true whether or not the noun itself openly manifests its gender. Consider, in this respect, pairs of nouns such as ħṓkəm ‘√ħkm, ruler’/bṓkər ‘√bkr, young she-camel’, ħáyr ‘√ħyr, donkey’/báyt ‘√byt, house’, hḗχər ‘√hχr, old man’/gḗzəl ‘√gzl, boulder’. The former member of each pair triggers masculine (singular) agreement while the latter triggers feminine (singular) agreement.17 The members of each pair have identical phonetic profile (CṓCəC, CáCC, CḗCəC, respectively), which means that neither audibly displays its gender in distinctive fashion.18

The distinction therefore must be that all six nouns in (23) – and all Mehri nouns, for that matter – bear a morphosyntactic Gender feature (call it F, for “Feminine”) differently valued, +F for Feminine and –F for Masculine.19 Once more, that feature is not realized in audible fashion in any item in (23). That is, the presence of F does not entail that it will necessarily materialize in the form of an identifiable portion of the signal.

(23) a. Feminine b. Masculine
  bṓkər +F ħṓkəm –F
  báyt +F ħáyr –F
  gḗzəl +F hḗχər –F

While F is not manifested (or not readily or unambiguously detectible) in the nouns in (23), it is audibly realized in the broken plurals which are the topic of this paper. In that class, as we know, F realizes as y when positively valued, and as w when negatively valued (feminine rəħayəb/masculine χətawəm). In view of this, I now lay out the ingredients of my proposal as regards the phonology of F.

I propose that morphosyntactic feature F spells out as in (24a). Depending on the value of F, it realizes as (24b, c).

(24) a. αF ➔ [–αROUND, –αBACK]
  b. +F ➔ [–ROUND, –BACK]
  c. –F ➔ [+ROUND, +BACK]

In the plural pattern discussed in this paper, F is realized as a glide. But in other plural patterns, F materializes in the form of a mid-vowel, back and round in the case of masculine nouns, e.g. plural ðəbṓl /ðəb{a+u}l/ ‘sides’, and front and unrounded in the case of feminine nouns, e.g. plural gəzḗr /gəz{a+i}r/ ‘matchets’.20 Consequently, I propose that [HIGH] and its value in such vocalic and consonantal realizations is not an intrinsic property of F. Rather, F strictly involves [αROUND, αBACK]. [HIGH] is contingent upon the position, C or V, onto which a given pattern will cause F to be realized. Of exclusive interest in the context of the present discussion is the realization of F on a consonantal position, a characteristic feature of the pattern under scrutiny. In (25), I implement my proposal: while [αROUND, αBACK] is the spellout of F, [HIGH] and its positive value are redundantly supplied by the C position itself.

    1. (25)

The substance of that proposal is that, inasmuch as HIGH is not part of the phonological features of F, F is associated with an incomplete (or partially specified) phonological matrix. A corollary of the proposal is that the spellout of F will remain unpronounced until redundant insertion of HIGH kicks in (+HIGH in the plural pattern under study).

This opens a new perspective on silent Gender exponence of the type illustrated in (23). Silent exponence can now be construed in two ways, (26).

(26) a. Silent exponence arises because F does not spell out.
  b. F spells out, but silent exponence arises because HIGH (and its value) is not inserted for some reason.

In (27), I illustrate the two possibilities by means of the masculine singular noun, ħṓkəm ‘ruler’ and the feminine singular noun bṓkər ‘young she-camel’, both arguably cases of silent exponence.

(27) a. b. c. d.
  ħṓkəm ħṓkəm bṓkər bṓkər
  –F –F +F +F
    +ROUND   –ROUND
    +BACK   –BACK

Configurations (27a, c) correspond to option (26a): F does not spell out; while configurations (27b, d) correspond to (26b): F does spell out, but HIGH is still not involved.

While both scenarios are compatible with silent exponence, they do not have identical content. (27a, c) merely record phonetic silence. By contrast, (27b, d) incorporate an empirical hypothesis which can be tested, verified, or refuted.21 At this point, and subject to what the investigation of further nominal and adjectival formations may bring to light regarding the grammar of Gender in Mehri, I opt for the stronger and richer of the two options in (26), viz. (26b).

In the next two subsections I will illustrate a scenario involving insertion of [HIGH] and culminating in actual pronunciation. I will also illustrate a situation where [HIGH] is not inserted and yet the consequently silent features of F make their presence felt, nevertheless.

I recapitulate the ingredients of my proposal, again deliberately formulated in the strongest possible version.

(28) a. F spells out at all times.
  b. F is realized at the Gen position and nowhere else.
  c. F is uniquely realized as in (24).

Given the claims in (28), a question arises as to why F may be realized in a non-canonical fashion (i.e. on [n CV] as shown earlier in (21)). This is the topic of the next subsection.

4.2 Why are there apparently two sites for Gender exponence?

I submit that the reason why Gender may be exponed in more than one site in the plural pattern discussed in this paper is to be found in the manner in which the root deploys over the template. Given token root √fʕl, a four place template, and edge-in association of root consonants as put forth in Yip (1988a), exactly two possibilities arise (29). In both cases, the first and last root consonants link up to the first and last templatic positions, respectively. Then, the medial consonant links up to one of the two medial templatic positions, leaving the other bare (boldface in (29)). The two configurations in (29a, b) arise.

(29) a. b.
 

Importantly, in the broken plural pattern under discussion, roots do not spread freely beyond what is shown in (29). That is, further spreading of the type indicated with arrows in (30) is just not observed.22

(30) a. b.
 
  c. d.
 

This being the case, I now demonstrate how exponence arises.

4.3 A polarity-free system of Gender exponence

Three events culminate in Gender exponence in the plurals under discussion (i) the deployment of the root, (ii) the valuation of F, and (iii) the involvement (or not) of feature HIGH. The root deploys according to one of the two options in (29). In either case, F will be valued +F or –F.

Option (29a) gives rise to canonical exponence, whereas option (29b) gives rise to non-canonical exponence.23

Option (29a) is the case where the deployment of the root has left [Gen CV] bare. The two possibilities regarding the valuation of F are represented in (31): +F in (31a) and –F in (31b).

(31) a. Feminine rəħayəb
(Canonical exponence)
b. Masculine χətawəm
(Canonical exponence)
 

Spellout of F – now illustrated with the example of the feminine plural noun rəħayəb ‘towns’ – proceeds as shown in (32). The Gen position having been skipped by the root, its onset is left bare. The spellout of F provides it with phonological material (32a) and +HIGH is redundantly inserted (32b). The canonical form of the Gender exponent arises (glide y in this case).

(32) a. b.
 

How does option (29b) culminate in non-canonical exponence? This time, the second consonantal position from the left has been left empty by the deployment of the root. At the relevant stage, the representations are as in (33), with a feminine noun in (33a) and a masculine noun in (33b).

(33) a. Feminine ħəwarəs
(Non-canonical exponence)
b. Masculine bəyrak
(Non-canonical exponence)
 

The root provides the onset of [Gen CV] with phonological material. As for F, it spells out as specified, +BACK, +ROUND for Masculine and –BACK, –ROUND for Feminine. F spells out at the Gen position and nowhere else, as indicated earlier. Because the consonantal position of [Gen CV] has been identified by root material, +HIGH is not inserted. For that reason, the features of F remain silent. Their (silent) presence will receive support shortly.

But, the salient feature of the configurations in (33) is that [n CV] has been skipped by the root. As such, they challenge well-formedness.

Several possibilities may be imagined as to how well-formedness might be attained, drawing of course on local resources.24 Local resources in the case at hand are of two kinds. One is the root. The other is the bundle of features resulting from the spellout of F. The root will be of little help since any attempt at further spreading one of its consonants beyond the initial arrangement would run counter to the generalization discussed earlier in subsection 4.2. Consequently, imaginable (but unattested) outcomes such as *ħəħarəs, *ħərarəs, *bəbrak, *bərrak, will not be further discussed.

The one remaining local resource is the bundle of features spelled out by F, namely [αROUND, αBACK]. The reassociation (or spreading) of Gender features to [n CV] would be a conceivable solution. Such a hypothetical scenario is shown by means of the arrows in (34).

(34) a. Feminine ħəwarəs b. Masculine bəyrak
 

But reassociation (or spreading) would only result in moving to the non-canonical position the rightful exponents of the canonical position (hence *ħəyarəs and *bəwrak).

While reassociation and spreading must be dismissed, it remains very clear that the Gender features are just the local resource being tapped. This is clear from the fact that the features of the glide in both ħəwarəs and bəyrak are not randomly assigned; rather, they are the exact opposite of what they would be under canonical exponence. The empirical generalization is as in (35).

(35) Non-canonical exponence is computed from canonical exponence.

My proposal crucially involves such a calculus. I propose that Gender features are copied to the left of their site of attachment (copy in boldface in (36)). Note that the (phonological) features of F are copied, not (morphosyntactic feature) F itself.

(36) a. Feminine ħəwarəs b. Masculine bəyrak
 

But now, the resulting configurations directly contravene the Obligatory Contour Principle. Two strategies are available in response. One is fusion of the two offending adjacent matrices and subsequent spreading. But with (34), we already saw that spreading does not yield the correct output. The only alternative is then to invert the values of the features of the copy, as is done in (37).

(37) a. Feminine ħəwarəs b. Masculine bəyrak
 

Well-formedness considerations will then ensure that the copied and modified matrix is realized as the onset of [n CV] in (37). This will directly trigger insertion of +HIGH and a glide will again surface (38).

(38) a. Feminine ħəwarəs b. Masculine bəyrak
 

The copy-and-dissimilate scheme I have proposed is formally similar to numerous treatments of reduplication with ensuing dissimilation of reduplicated material, be it infixed as in Cohn’s treatment of Sundanese (Cohn 1992) or prefixed as in Steriade’s discussion of Sanskrit (Steriade 1988). It falls well within the purview of the Obligatory Contour Principle as characterized by Yip (1988b: 73): “all rules involving identity of target and trigger with an output in which they are no longer identical and adjacent are OCP-triggered rules.” The outstanding feature of my proposal is that the trigger of the OCP effect remains silent.

In light of what precedes, the two main questions of this paper receive an answer: is there both canonical and non-canonical exponence? And, is polarity involved in the system? Under the analysis I have proposed, there is no such thing as non-canonical exponence as it might initially have been supposed. The Gender feature F uniquely realizes at [Gen CV] and nowhere else. Moreover, its spellout is unambiguous at all times: +ROUND, +BACK for Masculine and –ROUND, –BACK for Feminine. A parameter fully independent of the Gender system, namely the dual mode of deployment of the root is responsible for giving the impression that a dual system of Gender exponence (canonical and non-canonical) is implemented. Under one mode of deployment, [n CV] is identified by the root and [Gen CV] is left bare. The features of F (with the contribution of +HIGH) participate in the identification of the C position of [Gen CV]. This will be the only genuine case of exponence. Under the other mode of deployment, [n CV] is skipped and [Gen CV] is identified by the root. The features of F are silently present in their usual place and in their usual form. While they will not be “heard”, their presence is manifested in the way they feed a repair strategy resulting in their mirror-image.25

Now and before wrapping up the discussion, I want to shed light on a residual plural subpattern. This is the topic of the next section.

5 The fəʕyṓl plurals

The aim of this section is to introduce the residual subpattern in (39). These plural nouns are characterized by an extraneous glide y (absent in the corresponding singulars)26 and are of type fəʕyṓl /fəʕyal/. The plurals in (39d) are of interest because the front glide y they display in canonical position (between C2 and C3) would lead to the expectation that they will be feminine. Yet, they are all masculine.

(39)   a. Root b. Singular c. Plural
[phonetic]
d. Plural
/abstract/
Gloss
  i. √k’hf [k’əhṓf] [k’əhyṓf] /k’əhyaf/ ‘stony ground’
  ii. √fk’ɬ’ [fək’ṓɬ’, fák’ɬ’] [fək’yṓɬ’] /fək’yaɬ’/ ‘coconut meat’
  iii. √k’ħf [k’əħáyf] [k’əħyṓf] /k’əħyaf/ ‘coconut shell used as a cup, dish’
  iv. √bʕr [bəʕáyr] [bəʕyṓr] /bəʕyar/ ‘male camel’
  v. √fχð [əfχā́ð] [fəχyṓð] /fəχyað/ ‘flesh and bone from knee to hip’
  vi. √ɬk’f [ɬək’ā́f] [ɬək’yṓf]27 /ɬək’yaf/ ‘collar-bone; ledge’
  vii. √wʕl [wɛ̄́l] [wəʕyṓl] /wəʕyal/ ‘waʕl, mountain goat’

The challenge is that the plurals in (39d) seem to fit nowhere in the generalizations I have proposed up to this point. Indeed, a masculine plural noun such as k’əhyaf ‘stony grounds’ (superior line of (39d)) should be either k’əyhaf with y in the non-canonical position or k’əhawəf with w in the canonical position, in no case k’əhyaf (40).

    1. (40)

The data in (39) represents the entirety of the examples documenting that subpattern in Johnstone’s Mehri Lexicon (1987). Obviously, all the items in (39) involve a root whose medial consonant is a guttural, an ejective or uvular χ.28 On the other hand, when masculine plural nouns of type fəyʕṓl are examined (all are given in (41)),29none of them is found to involve a root whose medial consonant is a guttural, an ejective, or χ.

(41)   a. Root b. Singular c. Plural
[phonetic]
d. Plural
/abstract/
Gloss
  i. √brk [bárk] [bīrṓk] /bəyrak/ ‘knee’
  ii. √frɬ’ [fárɬ’] [fīrṓɬ’] /fəyraɬ’/ ‘best dates; religious duty’
  iii. √k’dr [k’ā́dər] [k’aydṓr] /k’īdar/ < /k’əydar/ ‘pot’
  iv. √ħtm [ħṓtəm] [ħaytṓm] /ħītam/ < /ħəytam/ ‘coffee pot spout’
  v. √ʕɬr30 [ʔā́ɬər] [ʔayɬṓr, ʔēɬṓr, ʔāyɬṓr]31 /ʕīɬar/ < /ʕəyɬar/ ‘male friend’
  vi. √θ’rb [θ’árb] [θ’ērṓb, θ’əyrṓb]32 /θ’īrab/ < /θ’əyrab/ ‘small piece of wood; (poet.) stature’
  vii. √k’wʕ [k’ḗʔi, k’ḗʕi] [k’ēyṓy] /k’īyay/ < /k’əyyay/ < /k’əywaʕ/33 ‘spirit, ghost’

The outstanding behavior of fəʕyṓl plurals, therefore, takes its place as one facet of the perfect complementary distribution in (42).

(42) a. The root-medial consonant of all masculine plural nouns of type fəʕyṓl is always a guttural, an ejective, or χ (fə{Gu, Ej, χ}yṓl)
  b. The root-medial consonant of all masculine plural nouns of type fəyʕṓl is never a guttural, an ejective, or χ (fəy{*Gu, *Ej, *χ}ṓl)

I conclude that the process of metathesis in (43) is at work.

(43) [n –F C1əyC2ṓC3] → [n –F C1əC2yṓC3] where C2 is a guttural, an ejective, or χ

Recognizing fəʕyṓl plurals as phonologically conditioned variants of the fəyʕṓl pattern raises the number of cases of “non-canonical exponence” to 14, 7 of which surface as fəyʕṓl (as previously documented in (3)) and 7 as fəʕyṓl.

6 Conclusion

My objective in this paper has been twofold. On the one hand, with my analysis of the exponence of Gender in a class of Omani Mehri plural nouns, I have attempted to contribute to a growing literature on the study of Gender. On the other hand, I have sought to provide more evidence for the view that templates are nothing but the output of morphosyntactic structure. The challenges were (i) accounting for the fact that Gender could apparently be marked in two different places, (ii) understanding whether the resulting pattern documents genuine polarity or not. I have argued that the problem lay outside the system of Gender exponence, indeed stemmed from the peculiar way in which the root deploys in the relevant set of plural nouns. The two major challenges apparently inherent in the evidence can then be reduced. On the one hand, I showed that Gender is exponed in one place only. On the other hand, I showed that polarity was a mere phonological effect.