The left edge of the word in the Berber derivational morphology

In many Berber varieties, causative and reciprocal verbs are built by means of monoconsonantal prefixes attached to a stem. These prefixes are realized as single or geminated depending on the properties of the stem. In this paper, it is argued that an initial templatic site is responsible for the length variation of the prefixes. Under specific licensing conditions, the initial site hosts the causative and the reciprocal prefixes by means of two distinct operations, namely movement and spreading. Moreover, complex combinations of those prefixes (causative + reciprocal, reciprocal + causative) feed apparently unrelated phenomena of selective harmony and dissimilation. They are argued to follow directly from the use of the initial site as part of the verb domain. Handled in syntactic structure, the initial site further allows accounting for the cooccurrence restrictions that the causative and the imperfective markers undergo: it is proposed that the causative takes precedence over the imperfective because it is generated lower in the structure under the vP. The same reasoning holds for the incompatibility of imperfective gemination with the reciprocal marker. It is precisely this type of restrictions that strictly phonological analyses fail to address.


Introduction
In Berber, the causative and the reciprocal prefixes are realized as single or geminated depending on the properties of the verb. The examples in (1) illustrate the phenomenon in Tashlhiyt Berber: ( The reader will have noticed that the vowels of the base form remain the same in the causative, whereas in the reciprocal they can be replaced by an invariant vocalic melody inserted after the first and before the last root consonants (/i…i/ in the aorist, /a…a/ Glossa general linguistics a journal of Lahrouchi, Mohamed. 2018. The left edge of the word in the Berber derivational morphology. Glossa: a journal of general linguistics 3(1): 30. 1-25, DOI: https://doi.org/10. 5334/gjgl.250 in the preterit). 1 In addition, the distribution of vowels in the verb appears to play an important role in the choice of the prefix variant: verbs beginning with a CV sequence select the simplex variant of the prefix, while those that begin with two consonants take the geminated variant. Complex combinations of these prefixes result in forms where only the inner prefix varies in length. As shown in the examples in (2), the causative prefix remains simplex, while the reciprocal prefix surfaces as single or geminated.
(2) Verb Causative + Reciprocal ʕawn 'help' s-m-ʕiwin 'make (them) help each other' ħada 'be beside' s-m-ħidi 'make close to each other' knu 'lean' s-mm-knu 'make lean mutually' rwi 'soil' s-mm-rwi 'make soil each other' Interestingly enough, these verbs are restricted in the way they form their imperfective. While base verbs use either gemination or tt-prefixation to mark the imperfective, causative and reciprocal verbs invariably resort to vowel insertion. Previous accounts of the length variation of the causative and the reciprocal prefixes focus on the phonological properties of the verb. Boukous (1987), Jebbour (1999) and Dell and Elmedlaoui (2002), among others, discuss prosodic constraints on the input or the output of the formation, but none of them has explicitly addressed complex forms of the types in (2) and (3). The analysis provided in this paper diverges from that trend. I claim that an initial templatic site is responsible for the length variation of the causative and the reciprocal prefixes. Moreover, complex combinations of these prefixes are argued to follow from the implementation of the initial site as part of the verb domain. The cooccurence restrictions the causative, the reciprocal and the imperfective markers display will be analysed at the interface between phonology and syntax, claiming that theses markers compete for the same templatic position. It will be argued that the identification of this position obeys a syntactic hierarchy in which the causative and the reciprocal take precedence over the imperfective because they are generated lower in the structure under vP. This proposal might appear to steer against the standard analyses, which generally agree on the fact that competition does not necessarily hold within syntactic structures, and that there is no reason for some syntactic categories to take precedence over other categories. It will be shown that templatic constraints are necessary in order to account for the cooccurrence restrictions of certain markers and the overt realisation of some morphosyntactic categories. The paper is organized as follows: section 2 discusses earlier accounts of the length variation that the causative and the reciprocal prefixes display. Section 3 provides some theoretical background necessary for the understanding of the proposed analysis. A templatic account for the length variation of the causative and the reciprocal prefixes is presented in sections 4-6. Section 7 turns to the cooccurrence restrictions underlying the distribution of the causative, the reciprocal and the imperfective markers. Section 8 concludes the paper.

Earlier accounts of the quantity variation of the causative and the reciprocal prefixes
Most accounts of the length variation of the causative and the reciprocal prefixes in Berber are phonological as they rest on syllable and prosodic structure of the verb (cf. Boukous 1987;Iazzi 1991;Lasri 1991;Jebbour 1999;Lahrouchi 2001;Dell & Elmedlaoui 2002;Saa 2010). They state that the quantity variation of these prefixes depends on the phonological properties of the verb. Others such as Chaker (1973;, Chami (1979) and Cadi (1987) examine the segmental environment of the prefixes, claiming that the non-geminated variants attach to verbs beginning with a consonant, while the geminated variants select verbs that begin with a vowel, including the short epenthetic [ə]. This section outlines the syllable-based analyses put forth in Jebbour (1999), Dell and Elmedlaoui (2002) and Saa (2010), as opposed to the templatic approach proposed in Guerssel (1992).

Syllable-based analyses
According to Saa (2010: 124) Although it is not explicitly claimed, the author suggests that the prefix is underlyingly simplex. The same reasoning underlies the analysis put forth in Dell & Elmedlaoui (2002: 128). The quantity variation of the prefix depends, the authors argue, on the number of syllables the output contains. That is, the prefix geminates when the output form is left with less than three syllables. Otherwise, the prefix remains ungeminated. An alternative is provided in Jebbour (1999: 13). The author argues that the difference between verbs like [mun] and those like [frs] lies in their moraic rather than syllabic weight. The first select the simplex variant of the causative prefix because they are bimoraic, while the second, monomoraic, require a geminated prefix. The distribution of the variants of the causative prefix ultimately lies on the distinction between monomoraic and polymoraic bases, regardless of the number of syllables: [
The form in (4a) constitutes, according to the author, an "open expression" which is grammatically unspecified. It is the abstract initial morpheme (in bold) that assigns this expression the feature [+verb]. It also allows the causative prefix to geminate by spreading into the empty initial onset (4b).
The analysis I will be developing in this paper differs in many respects from the one advocated in the case of Aït Seghroushen Tamazight. While I endorse the idea of an empty templatic site, I will argue that its activation is not automatic. Rather, it depends on the phonological properties of the verb stem, and more particularly on the ability of the following vowel to license the empty site. This will be shown necessary in order to account for the length variation of the causative morpheme in the Berber varieties studied here, as opposed to the Aït Seghroushen variety in which this morpheme is always geminated. The licensing status of the empty site will also allow to explain how complex combinations of the causative and the reciprocal morphemes result in forms where only the inner prefix varies in length while the outer one remains simplex (e.g. Tashlhiyt [s-mm-knu] 'make lean mutually' / [s-m-ħidi] 'make close to each other'), contrary to Aït Seghroushen Tamazight where it is the inner prefix that remains constantly simplex (e.g. [mm-zraj] 'leave each other' / [ss-m-zraj] 'cause to leave each other', [mm-xmaz] 'scratch each other' / [ss-m-xmaz] 'cause to scratch each other'). Moreover, it will be argued that both causative and reciprocal prefixes have access to the empty initial site, unlike the analysis proposed by Guerssel which makes the site available only to the causative prefix, considered to be acategorial, as opposed to the reciprocal morpheme.
In the remainder of this paper, I will try to describe more explicitly the templatic mechanism responsible for the length variation of the causative and the reciprocal morphemes in many Berber varieties, including Tashlhiyt, Kabyle and Tarifit. I will argue that an item-by-item computation is necessary to determine the way the empty initial site is activated in these varieties. To do so, I need to outline the main assumptions about the representation of the skeletal tier and the syllable structure in the framework adopted here.

CVCV model
As an offshoot of Government Phonology ), the CVCV model (Lowenstamm 1996) holds that syllable structure universally reduces to strict alternations of non-branching constituents, viz. onsets (C positions) and nuclei (V positions), which interact laterally to derive various syllable types (see also Scheer 2004). Skeletal positions that have no phonetic content are said to be licensed to remain empty by virtue of the government relation that they share with the neighbouring segments. Proper Government is one such relation which allows a vocalic position to remain empty when followed by a vowel. In addition, it is assumed following Kaye (1990: 314) that domain-final V positions are parametrically licensed to remain empty despite being ungoverned. This proves particularly interesting in analysing the vowel/zero alternations found in Berber varieties other than Tashlhiyt. The examples in (5) illustrate the situation in Tamazight Berber: In (5a), domain-final V 3 remains empty. V 2 , properly ungoverned surfaces as [ə] and therefore governs V 1 , leading to the form [kʃəm]. This contrasts with [kəʃmat] in (5b) where V 2, properly governed by the vowel /a/, remains empty while V 1 surfaces as schwa.
The reader is referred to Bendjaballah (2001: 188) and Ben Si Said (2014) Tashlhiyt). To illustrate this, let us consider the verbs [kʃəm] and [kəʃmat] just discussed in Tamazight Berber. In Tashlhiyt, no epenthetic vowel appears in the surface form of these verbs. Accordingly, in [kʃm] (6a) the consonant /m/ fills the V position to its left and then governs the empty V between /k/ and /ʃ/, whereas in [kʃmat] (6b) it is the consonant /ʃ/ that branches into the preceding ungoverned V position.
The left-branching representation is supported by the distribution of epenthetic vowels in Berber varieties where only vocalic elements appear in the nucleus position, including the aforementioned Tamazight variety. Indeed, the careful reader will have noticed that the syllabic consonants in the Tashlhiyt forms in (6) are exactly the ones preceded by an epenthetic schwa in Tamazight (5). Further evidence is found in languages such as German (Clark & Yallop 1995: 68), Dutch (Oostendorp 2004), and Coptic ( Before examining how syllabic consonants are involved in the computation of the length of the causative and the reciprocal prefixes in Tashlhiyt Berber, let us outline another theoretical device which proves necessary to understanding the whole analysis advocated in this paper. The next section presents the initial CV hypothesis, as stated in Lowenstamm (1999).

The initial CV hypothesis
In an attempt to rationalize the asymmetry between languages with only sonority-rising clusters at the beginning of the word (e.g. English: blue, true; French: travail 'work', plateau 'tray'; Italian: primo 'first', blusa 'blouse'), and languages where initial clusters are made sonority-free (e.g.  Lowenstamm (1999) argues that each word of a major category is preceded by an empty CV site. A word such as kit is thus represented as in (7b), where the initial site appears in bold: The postulation of this site, filled by means of phonological and morphological operations involving licensing condition, pits two types of languages against each other: -Type I languages such as English and French where the initial site is always licensed by the following vowel. -Type II languages such as Maghrebi Arabic and Hebrew where it is not always licensed.
The licensing status implies that the initial site hosts proclitics in type I languages but not in type II. The examples in (8) In these examples, the initial CV is licensed by the vowel /a/. In (8b), the obstruent and the liquid form a closed domain (in brackets) inaccessible to proper government, which allows the following vowel to license the initial site. 2 As a consequence, the definite articles le, les and l' attach directly to the initial CV. This is illustrated in the forms on the right side in (8). Biblical Hebrew differs from French in that it has surface sequences of obstruents and liquids at the beginning of the word and their mirror-image (e.g.
[lxaðim] (< /lkadim/) 'captures'). These sequences, which result from syncope of the initial vowel of the singular form according to Lowenstamm (1999: 159), allow considering the initial site in Biblical Hebrew as not always licensed. It is expected to host proclitics in words like [klaβim] but not words like [rqahim]. In fact, the initial site is used in both types of words, the author argues. In the first type, the vowel of the definite article /ha/ spreads into the V position of the initial site, while in the other it is the stem-initial consonant that spreads. This is shown in (9): According to Lowenstamm (1999: 164), a uniformity convention allows the initial CV to remain unlicensed throughout the language, even in words beginning with sonority-rising clusters. Vowel lengthening in [haːrqaħim] and consonant gemination in [hakklaβim] are viewed as the result of the identification of a templatic site which may not remain silent inside the phonological domain of the word. It should be noted, however, that lengthening of the definite article's vowel occurs only when consonant gemination fails, that is when the initial consonant belongs to the class of consonants which resist gemination. 3 An attempt is made in this paper to pinpoint a third type of languages (Berber), in which the initial site is not always licensed and its identification operates item by item. 4 I will argue that the causative and the reciprocal morphemes either spread or move into the initial site, depending on its licensing status. Moreover, complex combinations of these prefixes and apparently unrelated phenomena of selective (dis)harmony will be shown to follow directly from the presence of the initial site.

Causative forms
In many Berber varieties, the causative prefix is realized single or geminated depending on the properties of the stem. Generally-speaking, the prefix is single when attached to a stem that begins with a consonant-vowel sequence. It is geminated when attached to a stem beginning with a CC cluster. The examples in (10) illustrate the situation in Tashlhiyt Berber. 5 The same situation holds in Tarifit Berber and Kabyle, as shown in the examples in (11) and (12).  Dell & Elmedlaoui 1985;2002;Ridouane 2008). This feature will be shown to be of paramount importance in determining the status of the initial site, and therefore the size of the causative and reciprocal prefixes in verbs with no peripheral vowels. Before turning to this issue in section 4.3, let us show how the size variation of the causative prefix obtains in Tarifit and Kabyle. This variation follows directly, I argue, from the status of the initial CV with regard to licensing. In Berber, the initial site may be activated by means of two operations, namely movement and spreading. This is in contrast to French and Hebrew where only one of the two operations may be implemented. Movement occurs when the initial site is licensed by the following vowel, as opposed to spreading which is viewed as the result of the identification of a templatic site which may not remain empty inside the phonological domain of the word (see Hebrew in (9)).
The quantity variation of the causative prefix in the examples in (11) and (12) follows directly from the use of one or the other of the operations just mentioned. According to the present analysis, the causative prefix is underlyingly simplex. Its quantity variation depends on the way the empty initial site is activated: (13) (i) /s/ moves into the initial site when it is licensed, that is properly governed by the following vowel.
(ii) It geminates by spreading into the empty site when it is unlicensed.
The causative forms represented in (14) and (15)  , respectively. They are no exception to the analysis proposed here. Although their initial site is licensed, thus triggering the movement of the causative prefix, the gemination of their prefix results from the use of the empty onset of the first syllable in the stem. This is shown in the representation in (16) The gemination of the causative prefix is naturally expected in this context. The opposite would have been surprising since an onset position would have remained empty inside the syllabification domain.

Reciprocal forms
The reciprocal forms behave similarly to the causatives just discussed. In both varieties, the reciprocal prefix geminates when attached to a stem beginning with a CC cluster ((17b), (18b)); it remains ungeminated when followed by a CV sequence ((17a), (18a)). The Tashlhiyt data in (17a) show cases where the vowels of the stem are replaced by an invariant vocalic melody /i…i/, while Kabyle has many instances of schwa, which according to our analysis surface in ungoverned vocalic positions (see section 3.1 for details). The distribution of these schwas is of paramount importance in determining the length of the reciprocal prefix in Kabyle, as opposed to Tashlhiyt where no such vowels exist.

The Kabyle data further show examples like [nnəqbəl] and [nnəfk]
, where the reciprocal prefix undergoes labial dissimilation. This phenomenon, to which we will turn in section 6, is found in many other Berber varieties, including Tashlhiyt. It occurs whenever the stem contains a labial consonant. The length variation of the reciprocal /m-/ can be analysed in the same manner as the causative /s-/. It is underlyingly simplex. Its quantity variation follows from the presence of the initial CV: (i) /m-/ moves into the initial site when it is licensed by the following vowel.
(ii) It geminates by spreading into the initial site when it is unlicensed, i.e. followed by an empty V position.
The forms represented in (20) and (21)  In summary, the status of the initial CV with respect to licensing accounts for the length variation of both reciprocal and causative prefixes. These prefixes remain ungeminated when the initial site is licensed, otherwise they geminate. Complex combinations of these prefixes further support the present analysis. As we will show in section 5, only the inner prefix varies in length. The outer one remains constantly ungeminated. Before doing this, let us briefly discuss the causative and reciprocal forms containing syllabic consonants.

Syllabic consonants
Tashlhiyt Berber has only three phonemic vowels /i, a, u/. The so-called "transitional vocoids", found in certain consonant clusters (e.g. [nkər] 'wake up, stand up', [lkəm] 'arrive'), have no syllabic status according to Dell & Elmedlaoui (2002), while Coleman (1996; argues that they are epenthetic, filling syllabic nuclei that would otherwise remain empty. However, in forms entirely made of voiceless obstruents, like [kʃf] 'be faded' and [fsx] 'cancel', no such vocoids are heard (see Ridouane 2008 for phonetic and phonological arguments). Following Dell & Elmedlaoui, I assume that in Tashlhiyt any segment, including a voiceless obstruent, can be syllabic, but contrary to their analysis (Dell & Elmedlaoui 2002: 76), I claim that no sonority hierarchy is needed for segments in their competition for the status of syllable nucleus. Rather I propose within the CVCV approach that any vocalic position which lacks proper government systematically hosts one syllabic consonant. The reader is referred to section 3.2 for details and representations about this aspect of the analysis.
The vocalic positions which host syllabic consonants are allowed to license empty V positions, including that of the initial CV. In Tashlhiyt  The length of the causative prefix in these examples varies under the same conditions as in the ones discussed in section 4.1, except that the vocalic positions which license the initial CV host syllabic consonants. More particularly, when the initial site is licensed by a vocalic position to which a syllabic consonant is associated, the prefix remains ungeminated (22a). It geminates when the initial site is unlicensed (22b). 6 The forms represented in (23)  In (23a), V 4 remains empty by virtue of being domain-final. V 3 , properly ungoverned is filled by the neighbouring consonant and therefore governs V 2 . V 1 , just as V 3 , allows the syllabic consonant to branch and then licenses the initial CV, leading to a situation where the prefix s-can move into its C position. This contrasts with the situation in (23b) and (23c) where V 2, properly ungoverned, hosts a syllabic consonant and then governs V 1 . As a consequence, the prefix geminates by spreading into the unlicensed initial site. The reader will have noticed that the medial consonants in (23b) and (23c) remain non-syllabic, regardless of their sonority status. If sonority hierarchy were decisive in the choice of the syllabic consonant, the form in (23b) would have selected the simplex variant of the prefix. That is, the medial consonant /r/ would have been syllabic, therefore licensing the initial CV.

Complex combinations of causatives and reciprocals
As stated so far, my analysis predicts that only the inner prefix will vary in length when both causative and reciprocal prefixes cooccur in the same form. This is exactly the situation in causativized reciprocals, as shown in Tashlhiyt  only the accounting for complex phenomena of selective harmony and dissimilation which involve the above morphemes. This is discussed in the following section.

Sibilant harmony and labial dissimilation
In (25a), the reciprocal m-remains ungeminated, as it moves into the initial CV licensed by the following vowel. This contrasts with (25b) where the initial site, unlicensed, allows the prefix to geminate by means of spreading. However, the causative s-, which lies outside the domain of the verb, remains simplex. This is all the more interesting that in the Aït Seghroushen variety studied by Guerssel (1992) and briefly discussed in section 2.2, the inner prefix remains constantly simplex (e.g. The domain of the verb, of which the initial site is part, also proves interesting in accounting for complex phenomena of selective harmony and dissimilation which involve the above morphemes. This is discussed in the following section.

Sibilant harmony and labial dissimilation
In Tashlhiyt, like in many other Berber varieties, the causative prefix undergoes voicing and anteriority harmony when the verb stem contains a sibilant consonant. This is shown in the examples below.  The reciprocal prefix undergoes dissimilation without regard of its length, similarly to what has been noticed in the case of sibilant harmony in causatives. However, only consonants trigger labial dissimilation (27a); labial vowels and glides are by no means involved in the process (27b). Focusing on sibilant harmony, Elmedlaoui (1992: 16) addresses the problem in terms of feature spreading. His analysis accounts for all cases where the causative prefix undergoes harmony. However, it fails to explain why this prefix remains unchanged when it is followed by another prefix. The same holds for the reciprocal prefix, which resists dissimilation when another prefix intervenes before the stem.
Strict adjacency to the stem seems necessary in order for both processes to occur. Consider the verb [ħʃʃm] 'be ashamed'. In the causative form [ʃħʃʃm] 'cause shame', the prefix undergoes anterior harmony (/s-/ > [ʃ]). Interestingly, when adding the reciprocal prefix immediately before the stem, the causative /s-/ remains unchanged while the reciprocal /m-/ dissimilates into [n], leading to the form [snħiʃʃim] 'cause (them) shame each other'.
In light of these facts, one can suggest that sibilant harmony and labial dissimilation operate within a specific domain, of which the causative and the reciprocal prefixes are not part when they occur in the outer position. This domain, put in brackets throughout this paper, is defined as follows: The domain of the verb consists of the stem template, preceded by an initial empty CV.
The initial CV allows for the causative and the reciprocal prefixes to attach inside the domain of the verb. In complex forms where both prefixes cooccur, only the inner prefix has access to the initial site, and can therefore interact with the segmental content of the stem. As shown above, the initial CV is licensed by the following vowel, thereby allowing the adjacent prefix to move into the domain of the verb. As a result, this prefix dissimilates into [n] while the outer prefix resists anterior harmony.
takes precedence over imperfective gemination and tt-prefixation. We need to explain how the causative prefix has access to the initial empty CV before the imperfective is formed. One way to address this issue lies in the syntactic structure of the verbal phrase. Causativization as an operation on argument structure is generally argued to occur lower in the structure than the aspect marker. 8 Using three distinct projections and a head movement approach (cf. Travis 1984 on the Head Movement Constraint; Baker 1988 on Incorporation), we can posit that the initial CV is generated under VP, the projection immediately above which the causative is formed, while the imperfective is generated higher in the structure under AspP. 9 The form in (33) illustrates the derivation.
The empty initial CV in (33) moves from the head position of the VP to adjoin to the upper causative morpheme /s-/. When raised to Asp, the initial site no longer contains a C position available for the imperfective marker, thus ruling out the form *[ttsgawar]. Recall from the above discussion that [tt-] connects to two C slots, one of which belongs to the initial site. However, the vowel /a/ between the last two consonants, which comes with the imperfective prefix, remains in situ in the causative form.
However, one problem might arise with the linearization of the syntactic structure in (33). Under the standard view that moved heads left-adjoin the upper heads, we expect the linear order of affixes to be mirror-image of the order of syntactic constituents. That is, we expect the imperfective marker to appear before the causative. Actually, this is true 8 The vP projection has been originally proposed for causative formation (cf. Chomsky 1995;Kratzer 1996). Arbaoui (2010: 77-78) argues that in Classical Arabic the causative (also called Form II) is generated under vP, lower in the structure than the imperfective. 9 The same reasoning underlies the analysis proposed by Achab (2006: 187), although the author dispenses with little v. Instead, he uses an additional VP projection as the locus of causativization (see also Guerssel 1986).
to the extent that it concerns languages with concatenative morphology, and in which templates play no specific role. In Berber, templates play a central role in arranging the order of morphemes. The structures of the type in (33) are meant to reflect the alleged templatic character of this language. Under the assumption that templates determine the linearization of grammatical information, the empty initial CV, which undergoes movement, is intended to explain the competition between the causative and the imperfective markers. The outcome of this competition does not require any specific statement about the order of affixes, since only the causative marker is realized. If the realization of the causative and the imperfective markers were not under templatic control, we would have had a form like *[ss-t-gawar], where the causative marker occurs in the outer position. 10 Similarly to the case explained in (33) above, the incompatibility of the causative maker with the imperfective gemination follows from the fact that these markers compete for the same CV site. Since the causative is generated lower in the structure than the imperfective, the medial root consonant /k/ no longer has a C position for gemination. The form in (34) (1985). One interesting example comes from verbal inflection: Under the view that Gender is projected lower in the structure than Number and Person (cf. Fassi Fehri 1984;Shlonsky 1989 on Semitic), we expect the verb stem to be followed immediately by the gender marker, then number and person (i.e. verb + gender + number + person). The order attested in Berber is actually the reverse. In the preterit, the verb stem is followed by number, then gender, while the person marker is prefixed (e.g. [t-nkr-m-t] 'you stood up (fm)'). Similar facts are found in Arabic (Lowenstamm 2011: 170) and Ethio-Semitic (Rose 1996).
The empty initial CV in (34) moves from the head position of the VP to adjoin to the upper causative morpheme /s-/. When raised to Asp, the initial site no longer contains a C position available for medial consonant gemination, thus ruling out the form *[sslkkm].
In the absence of the causative morpheme, the initial site provides the empty C position needed to achieve gemination, yielding the form [lkkm] 'arrive-imperfective'. Complex combinations of the causative and the reciprocal morphemes are obtained in the same way. The causative reciprocal forms of the type discussed in section 5 show that only the reciprocal /m-/ (inner prefix) varies in length for it has access to the initial CV, as opposed to the causative /s-/ (outer prefix) which remains constantly simplex. Unlike the causative, the reciprocal prototypically selects transitive verbs, be they lexical or derived (i.e. causativized), possibly turning them into intransitives in the sense of Buerning (2004: 03). That is, the reciprocal derives a verb whose argument is both the agent and the internal argument.
Assuming a Reciprocal Projection (or a Reciprocal VoiceP as proposed by Buerning 2004), which takes a VP as its complement, and immediately above which little v introduces causativity, the derivation of the form [smmrwi] proceeds as shown in (35) The empty initial CV moves from the head position of the VP to adjoin to the upper reciprocal morpheme m-. Because unlicensed (see section 3.2), the initial site is filled by means of spreading, resulting in the gemination of the reciprocal morpheme. The causative, projected immediately above RecipP, is left with only one consonantal position and therefore remains ungeminated.
Templatic competition at the interface between syntax and phonology allows unifying the analysis of the cooccurrence restrictions of the causative and the imperfective, and the length variation of the reciprocal and the causative. However, it only partially accounts for the facts about the imperfective of reciprocals. Unlike causatives which resist both gemination and tt-prefixation in the imperfective, reciprocals only repel gemination. That is, all reciprocal forms resort to tt-prefixation, regardless of the mechanism their bases select. Consider the following forms in Tashlhiyt. from "medio-passive", whereas gemination is viewed as the prototypical marker of the imperfective. It is precisely this marker that behaves uniformly throughout the derivation, combining neither with the causative nor with the reciprocal.

Conclusion
The typology put forth in Lowenstamm (1999) distinguishes two types of languages: (i) type I languages such as French where the initial CV is always licensed and used as a cliticisation site, and (ii) type II languages such as Hebrew where the initial site is not always licensed but where a uniformity convention allows using it as if it is always unlicensed. In this paper, an attempt has been made to pinpoint a third type of languages, namely Berber, in which the initial site is not always licensed, and an item-by-item computation is necessary to determine the way it is activated. I have argued that the length variation of the causative and the reciprocal prefixes depends on the licensing status of the initial site. They remain simplex when the initial CV is licensed, and geminate when it is unlicensed. Complex combinations of those prefixes have shown that only the prefix which has access to the initial site varies in length. The same reasoning holds for apparently unrelated phenomena of selective harmony and dissimilation. I have argued that the causative prefix harmonizes with the sibilant in a stem when it is associated inside the domain of the verb, of which the initial site is part. Likewise, the reciprocal morpheme undergoes labial dissimilation when it has access to the initial site. Within a syntactic approach, I have also discussed the cooccurrence restrictions the causative and the imperfective markers undergo: I have argued that the causative marker is generated lower in the structure than the imperfective, resulting in forms where only the former is realized. The same reasoning underlied the incompatibility of the imperfective gemination with the reciprocal marker.