The apophonic chain and the form of weak and strong verbs in Palestinian Arabic

4.1 Some more on the vocalization classes of Palestinian Arabic

For expository reasons, the table in (11) recalls the strong bases of inflection in all measures, already shown in (1). As noted in the introduction, M1 verbs belong to one of two major groups: [a]-theme or [i]-theme (these vowels are in bold in 11), while verbs in other measures belong to the [a]-theme exclusively.

The division of M1 verbs into vocalization classes has taken the data at face value, as they are presented in the dictionary in Elihay (2007). However, from a scrutiny of all of the 384 verbal entries in the dictionary, it turns out that not all combinations of consonants and vocalization classes are possible or equally likely. Indeed, the identity of the second or third consonant of the stem has an effect on the distribution of verbs in vocalization classes:

In the [a]-theme, the vocalization class [a-a] is exclusive to verbs with a post-velar 2nd or 3rd consonant (post-velars are uvulars [x, ʁ] or gutturals [ħ, ʕ, h, ʔ]). All verbs with both emphatics and post-velars in these positions are in this group. Still, a total of 18 verbs with 2nd or 3rd post-velars (and no emphatics) appear in the two other vocalization classes of the [a]-theme. Verbs with 2nd or 3rd emphatics (unless accompanied by a post-velar) appear in the [a]-theme only in the [a-o] vocalization class, where there is also a large proportion of verbs with 2nd and 3rd plain consonants, and 11 verbs with post-velars. The [a-ɛ] class harbors mostly verbs with 2nd and 3rd plain consonants, with seven cases involving post-velars. There are also two exceptions to the generalization regarding emphatics and the [a-o] class, [waṣal] “connect” and [waṣaf] “describe”, which I will assume can be accounted for by dissimilation from the [a-o] class owing to the initial [w] (no verb in [a-o] has an initial [w]). As for the [i]-theme, all combinations are common. Interestingly, of the 89 verbs of the [i]-theme, 82 are in the [ɛ-a] class. For the remaining seven the dictionary reports that the vowel of the imperfective varies between [a] and [ɛ], depending on the region. ^[7],[8]

There is only one exceptionless generalization in (12): if a verb in the [a]-theme has a 2nd or 3rd emphatic, then it is certain to belong to the [a-o] group. The two other distributional generalizations, regarding verbs in the [a]-theme with 2/3 post-velars and the vocalization classes of the [i]-theme, are only strong tendencies. Any analysis must, besides covering the regular cases, leave room for these exceptions. It is also noteworthy that as (1) above showed, the [a-ɛ] vocalization is the only option for verbs in measures other than M1, regardless of their consonantal make-up (with M6 being in fact [a-a]). As pointed out for CA by Guerssel (2003), for this reason, it cannot be maintained that the [a] vocalization of the [a-a] group is a phonological consequence of the proximity of the post-velar, even if one ignores the exceptions in M1; we will return to this proposal below.

In the introduction, it was claimed that Guerssel and Lowenstamm’s (1993, 1996) analysis can explain the attested classifications: given the perfective unsuffixed stem, the imperfective can be predicted. In the next subsection, we will examine the CA account and the differences between the CA and PA data sets and see how PA can be accounted for.

4.2 The logic behind the subclasses in PA: strong verbs

In this subsection, it will be shown that the same logic that was proposed for Classical Arabic can be used to motivate the regularities in PA. There is no implicit claim that the latter is derived from the former, but rather that the same logic holds in both.

In Classical Arabic, strong verbs in M1 are not divided into an [a]-theme and an [i]-theme. Instead, the first vowel of the stem is always [a]:

All of the verbs in class (13a) have 2nd or 3rd post-velars. Guerssel (2003) considers this “guttural” group (13a) to be the phonologically-conditioned allomorph of the [a-i] group (and thus not purely a phonological issue). No exceptions to this rule are reported. In this state-of-affairs, the problem that Guerssel and Lowenstamm faced when they approached the data in (13) was mainly to distinguish between (13b) and (13c), both of which have a vowel [a] in the perfective, but [i] and [u] respectively in the perfective. Their answer was simple: the [a] of (13b) is a copy of the other /a/ of the base, the one in V₁, whereas that of (13c) is a real lexical /a/. The lexical vowel of [ḍarab] is in fact /ø/ (i. e. it is null). This, with the aid of the apophonic chain in (10b) above, puts the system in perfect order: perfective /ø/=> imperfective [i], and perfective /a/=> imperfective [u].

Let us now recast the PA data in (14) as we did for CA:

It is not hard to relate the CA set in (13) to the PA set in (14). The [a]-theme of PA (14a-c) corresponds to (13a-c); in both systems, the [a-a] set is restricted to verbs with second or third guttural consonants. The [i]-theme of PA (14d) corresponds to (13d). There are four noteworthy differences between the two systems. First, the PA system seems to have five vowels, not three, with [ɛ] and [o] appearing in PA where [i] and [u] did in CA. Second, in PA not every verb in the perfective has [a] in V₁; instead, the correct generalization is that in the perfective, the two vowels of the stem have the same value for the feature [low]. Third, there is a small but important class [ɛ-ɛ] that does not correspond to any of the classes of CA. Finally, the [u-u] class of CA does not exist in PA.

Let us begin with the mid vowels and with those vocalization classes that exhibit apophony. All of the mid vowels that we have seen until now were in word-final CVC syllable: [yíktɛb] “that he write”, [yúdros] “that he study”, [ríkɛb] “he rode”. A closer examination reveals that short [ɛ] and [o] in word-final CVC syllables alternate with [i] and [u] respectively, when they are not word-final, for instance when an object suffix is added: [yíktibo] “that he write it”, [yúdruso] “that he study it”, [rikíbhom] “he rode them”. ^[9] If so, [ɛ,o] are in fact phonetic realizations of /i,u/ in this position. The analogy with CA becomes almost perfect: we have three classes [a-i], [a-u], and [i-a]. As the lexical representations in (15) show, all three instances of the apophonized imperfective are accounted for by the apophonic chain.

For (15a,b), one can adopt the CA analysis into PA. In (15a), the second [a] of [katab] is copied from the theme /a/, while that of [daras] (15b) is a real lexical /a/. Accordingly, the imperfective forms are respectively /ktib/ and /drus/, which are realized as [ktɛb] and [dros]. The representations in (16) are identical to those proposed by Guerssel and Lowenstamm for CA. In (16a), there is one default /a/ which spreads to the V₂ position; in (16b) there is also a lexical /a/ alongside this default [a]. Note that because this is not a derived form, the derivational CV is not accessed.

However, the analysis is not easily extended to the perfective /rikib/ in (15c). In (16), it must be assumed that the PA perfective template has a default /a/ in V₁, but /rikib/ does not have a vowel /a/ at all, but rather two /i/ vowels. In addition, in the CA analysis, the existence of a default vowel /a/ for all verbs was motivated by its ubiquity not only in all perfective forms, but also in the imperfective M1 prefix e. g. [kataba]-[aktub]. In Guerssel and Lowenstamm’s analysis, the two vowels are the same “templatic /a/”. This motivation, too, is absent from PA, where the prefix always has the quality of an epenthetic vowel and is realized only when stressed or before preceding a consonant cluster: [tíktɛb] “that you write” and [tiktíbhom] “that you write them” but [tṣi:r] “that you become” or [tħibb] “that you love”. How is the system of PA different from that of CA?

Let us reconsider the data set from PA. Examining all the perfective bases of PA, and regarding [ɛ] as /i/, the emerging regularity is that the stem must have identical vowels in V₁ and V₂. This fact must be built into the template, as in (17):

The vowel of the stem can be either [i] or [a]; this is what I have been calling [a]-theme and [i]-theme. In the [i]-theme, there is no reason for a learner of PA to ever assume a lexical /a/. I conclude that /a/ is not present in all M1 verbs in PA. Instead, I propose to formalize the theme distinction that we have been using. With the aid of the template in (17), we may derive the two representations in (18) (the derivational CV is absent from the representations; <Q,T,L> stand for the consonants of the stem):

(18)

Two verbal themes in PA: first attempt

a.	[a]-theme	b.	[i]-theme

The position of the theme vowel in (18) is V₁. But why should this be? Given the template in (17), the surface form would be identical if the vowel were in V₂. In other words, since the two vowels of the perfective are always traceable to a single lexical vowel, nothing in the phonetic output indicates that the theme should be represented in V₁; another possible representation would have the vowel originating in V₂. There are thus four possibilities, as in (19).

(19)

Four possible representations for perfective verbs

a.	[a]-theme, in V1	b.	[a]-theme in V2
c.	[i]-theme, in V1	b.	[a]-theme in V2

The learner cannot tell, on the basis of the perfective, which is the right representation for a [QaTaL] or [QiTɛL] surface form. ^[11] This is where exposure to the imperfective is important. Recall that V₂ of the imperfective is derived by apophony from V₂ of the perfective. If the theme occupies V₁ in the perfective, then V₂ is empty. Thus, apophony will derive /i/ in V₂ of the imperfective (step ø=>i), regardless of the perfective theme. On the other hand, if the theme occupies V₂, then the quality of the theme matters. If it is /a/ we predict /u/ in the imperfective, and if it is /i/ we predict /a/. ^[12] In other words, the prediction is that both themes will have imperfectives with /i/, but only the [a]-theme will have imperfectives with /u/, and only the [i] theme will have imperfectives with /a/. The prediction is correct: this analysis derives all and only the attested patterns, and most importantly incorporates the seven [ɛ-ɛ] verbs in (14e) above:

By replacing the default status of /a/ in M1 with the template in (17), the lexical vowel /i/ of CA corresponds to a “theme”, on a par with the lexical or default CA /a/. Given (17) and the availability of two slots in the perfective, four representations were possible for the attested array of perfective bases; on the basis of the imperfective, it was shown that all representations are possible. Just as in CA, the learner needn’t memorize the imperfective vocalization, but rather only the correct representation of the perfective.

The analysis above involves a symmetry between the [a] and [i] themes, which derives the exceptional [ɛ-ɛ] vocalization class (20d). One possible objection is that there are only seven verbs in this class. Indeed, while the distribution the two vocalization classes in the [a]-theme is even, their distribution in the [i]-theme is extremely uneven. Still, the reason is clear, at least under the probable assumption that the ancestor of PA was more like CA in the relevant respect. The two vocalization classes of the [a]-theme are present in CA, whereas the [ɛ-ɛ] vocalization class is an emergent development of PA. Most verbs in the [i]-theme correspond to QaTiL verbs in CA, and accordingly have imperfectives with [a], so it is likely that most perfectives will be lexicalized as /QøTiL/. It is rather a strength of the analysis and not a weakness that it depicts an architecture from which this new vocalization class can emerge naturally.

The final difference between CA and PA was the absence of the [u-u] vocalization. Why can’t the lexical vowel /u/ of PA be a theme vowel? Why is [QuToL] not a legitimate perfective form? The answer is probably diachronic, but the fact can nevertheless be stated synchronically with the help of the apophonic chain. Assuming that the perfective is always the base of derivation, the synchronic statement in (21) suffices:

This statement rules out not only an [u-u] class, where /u/ would have been situated in V₂ and the input to apophony, but also an [u-ɛ] class where /u/ would have been in V₁.

For completeness, a word is due regarding the large class of verbs with 2nd or 3rd post-velar consonants in the [a]-theme. This class resists apophony, with both the perfective and the imperfective exhibiting [a]. The corresponding class in CA was analyzed by Guerssel (2003) in a way that is by and large compatible with the present analysis. Verbs of the [a-a] class, Guerssel claimed, are the guttural version of the CA [a-i] class, where there is no lexical vowel. Because of this absence, the low vowel of the perfective is represented as originating in the guttural. This proposal is adapted to the PA analysis in (22a). For such verbs, I propose, the theme and its position are irrelevant, because the vocalization coming from the guttural will cancel the theme. Just like in CA, because the vowel is linked to an adjacent consonant, it will remain stable in the imperfective too.

(22)

Representations of the perfective and imperfective with guttural 2nd or 3rd consonants

a.	/føtøħ/ => [fataħ]	b.	/iftøħ/ => [iftaħ]

Guerssel argues convincingly that this is not a phonological process, because it does not apply whenever its conditions are met. Indeed, there is no trace of it in measures other than M1 in CA. This is also true in PA. In Guerssel’s analysis, the association of the post-velar to V₂ is blocked when there is a lexical vowel, and this can be shown for PA too, by the handful of verbs which belong to the [a-u] and [i-a] vocalization classes (23a,b). PA also adds the 7 exceptions in [a-i], one of which appears in (23c). This class was analyzed as having an [a]-theme in V₁. The existence of such verbs further supports Guerssel’s view of the process (22) as non-phonological, in that there are exceptions even within words with no vowel in V₂. ^[13]

(23)

The three other possibilities for M1 verbs with 2nd or 3rd post-velars

a.	[daxal]-[udxol] ‘enter’	b.	[rijɛʕ]-[irjaʕ] ‘return’
c.	[raʕad]-[irʕɛd] ‘be thunderous’

With these representations, we have covered the perfective and imperfective stems of all strong verbs in PA. For each of the two possible perfective vocalizations, the speaker has to interpret the position of the vowel in the perfective, and this position will determine whether the vowel will be a source for apophony or not. Having made that decision, the apophonic chain will yield all four patterns that are open for all consonants. The fifth, exclusively post-velar [a-a] pattern was accounted for by attributing the perfective vocalization to the post-velar. We will now see that the same principles predict correctly the behavior of weak verbal stems.

4.3 Confirming the analysis beyond strong verbs

In this subsection, we explore the attested patterns in weak verbs. The table in (24) repeats the three types of weak verbs: biradical, hollow and weak-final verbs:

Each of the weak types enters one of the groups already established for strong verbs. In the biradical and hollow verbs, the base never belongs to the [i]-theme. ^[14] The base vowel can either be the input to apophony or not, and this decision will determine the vocalization of the imperfective according to the apophonic chain. ^[15] In weak-final verbs, there are once again two themes. Unlike in strong verbs, there is no ambiguity as to the status of the vowel: all weak-final verbs of the [a]-theme behave as if this vowel were default and not the input to apophony, and in the [i]-theme the vowel of all weak-final verbs of is always treated as lexical. ^[16] Weak verbs, if so, exhibit a greater extent of predictability: only one type of weak verb has a theme distinction, and within that theme there is only one way to interpret the vowel. It is further noteworthy that like strong verbs, weak verbs exhibit all and only the three stages of the apophonic derivation ø=>i=>a=>u. In that, they support the analysis of the system with this theoretical tool.

Let us now go over the representations of all these verbs, however briefly, in order to be explicit about the analysis proposed. First, I assume that the association of segments to template proceeds from the edges inwards, with consonants spreading leftwards (otherwise, there are problems with M2; see Yip 1988, Buckley 1990). Second, as will be elaborated upon in the next section, I assume that verbs with two surface consonants are based on roots with a /ø/ element. This /ø/ must be represented as occupying a position in the template, because otherwise, given the edge-in mapping, all weak verbs would behave identically.

The representations of the two types of weak-final verbs are provided in (25a,b). The ø element occupies the final position. All verbs of the [a] theme will be represented as in (25a), and all those of the [i] theme as in (25b). The representations of hollow verbs are provided in (25c,d). The ø element occupies the second consonantal position of the root. Again, whether the theme /a/ is positioned in V₁ or V₂ will determine the vocalization of the imperfective.

(25)

The representation of bases with weak elements

a.	[rama] ([-irmi]) ‘throw’	b.	[nisi] ([-insa]) ‘forget’
c.	[ṣa:r] ([-ṣi:r]) ‘become’	d.	[ka:n] ([-ku:n]) ‘be’

The substance of ø becomes crucial in distinguishing biradicals from hollow verbs. If mapping is edge-in, the initial representation of biradicals must be as in (26), with the second consonant mapped to the last consonantal position. The element ø must be absent from this representation in order to explain the subsequent spreading of the second consonant to the second consonantal position in biradicals but not in hollow verbs above. Independently of this issue, biradicals present a slight complication, because V₂ is never realized in the perfective. This might give the impression that our analysis wrongly predicts only one vocalization class, or even no vocalization class. But this impression is a false one: apophony takes as its input the vowel present in the lexical representation of V₂. In strong verbs, apophony takes as its input an empty V₂ before the V₁ vocalization occupies it ([katab]-[iktɛb], ø=>i); similarly, it may take as its input an empty/full vocalization before the spreading of the final radical disallows its realization in V₂. Thus, the input to apophony in (26a) is /ø/, because /a/ is in V₁, but it is /a/ in (26b).

(26)

Biradicals: perfective bases

a.	[ħall]-([-ħill]) ‘solve’	b.	[ħaṭṭ]-([-ħuṭṭ]) ‘put’

Biradicals attest to the complete absence of a V₁ theme vowel in the imperfective. In biradical imperfectives, V₁ – the position immediately following the first consonant - is the only position realized. Nevertheless, it is the apophonized V₂ of the perfective which is found in this position in the imperfective stem, not the /a/ of the perfective.

As mentioned in the beginning of this subsection, there is no [i]-theme in hollow and biradical verbs. As it stands, the account cannot motivate this distribution, as it cannot explain the unambiguous treatment of the [a]-theme of weak-final verbs as V₁ themes. However, the representations in (25) and (26) do allow us to state the gaps in distribution as in (27):

To summarize, the analysis of the bases of inflection in PA easily extends from strong verbs to weak ones. All of the patterns in weak verbs are also attested in weak bases. A few distributional gaps among weak verbs, though still awaiting motivation, can be easily defined by independently-necessary analytic tools.

Verbs in measures other than M1, as mentioned several times above, also follow the apophonic chain while showing more predictability than strong M1 verbs. There is only an [a]-theme, and it is always treated as appearing in V₁, so that the imperfective always has a vowel /i/ (realized [ɛ] word finally), the exception being M6, which is consistently derived by concatenation rather than by apophony. When the imperfective form necessitates more than one vowel, the /a/ of the theme reappears if the position to be realized immediately following the first radical, as in M2. If it is the prefixal position to be realized, it is the default [i] which is used, as in M1.

The entire system of bases can be summarized as in (29), where <Q,T,L> stand for the consonants of the stem and G for guttural and post-velar cononants. Every branching node represents arbitrary distinctions that the learner must record. Given a root, the speaker has to know in which measure it appears. In measures other than M1, the derivation of both perfective and imperfective will be predictable: the imperfective stem will have a vocalization <a,a>, and the apophony that derives the imperfective will take as its input the lack of lexical vowel. ^[17] If the verb is in M1, then the number and character of its radicals is relevant. If it is a biradical or a hollow verb, then there is no doubt about the surface form of the perfective stem, because of the statements in (27) above. The speaker will nevertheless have to decide on the representation of the theme of this stem as originating in V₁ or V₂, depending on the imperfective form. For the weak-final verbs, the theme is not predictable: the speaker will need to remember what root goes with what theme. But once the theme is chosen, there is no ambiguity: the weak-final [a]-theme must originate in V₁, because of (27b), and the [i]-theme will originate in V₂. Finally, triradical roots involve the most choices. First, the theme of the root must be memorized. Then, it must be decided, on the basis of the imperfective, whether the theme originates in V₁ or V₂ (recall that if the theme is /a/ and the 2nd or 3rd consonants an emphatic, then there is no choice ̶ the theme is treated as originating in V₂). Finally, a large majority of the [a]-theme verbs with 2nd or 3rd post-velars do not alter their V₂ vocalization in the imperfective. Note that in all of the cases in which there is a choice for the original position of the theme, the speaker must be exposed to the imperfective form. This exposure helps establish the representation of the perfective base as one of the five options in (19) and (22) above. Once that representation is established, the form of the imperfective needn’t be stored, because it is predictable by the apophonic chain.

With the exception of the triradical post-velar verbs, all of the perfective-imperfective pairings in PA involve a single step in the apophonic chain of Guerssel and Lowenstamm (1993, 1996). The analysis thus provides synchronic support for existence of this derivational schema.

An additional note is due with regards to storage. In the last 20 years or so, a current in the study of Semitic Morphology has emerged where roots are not viewed as morphemes (e. g. Bat El 2003). Rather, it is stems that are stored, and the form of the stem establishes its classification in a given measure. In this analysis, the only psychologically real level of morphological activity is that represented by the “stem” level in (29). The analysis in this paper is compatible with this view too, as long as the stems that are stored can make similar distinctions between a V₁ and V₂ position of the theme.

To summarize the discussion of the bases of inflection in this section of the paper, it was shown that all of these bases behave analogously and according to the apophonic chain. In the next section, an account will be provided for the inflectional paradigm of these stems.

5.1 Strong verbs

The table in (30) presents the inflection of the representative strong verbs [katab] “write”, [daras] “study” and [rikɛb] “ride”, with stress also marked. As is well known (e. g. Brame 1974), stress in PA is completely predictable from \syllable structure. It falls on the last heavy syllable, namely non-final closed or long syllables. If there is no heavy syllable, stress falls on the antepenultimate syllable. ^[18]

In terms of the form of the stem, the imperfective is always [VCCVC], while the distribution of stems in the perfective depends on their theme and on the environment of the stem. One must distinguish four environments: pre-consonantal, unsuffixed, before 3pl [-u], and before 2fmsg [-at]. In the preconsonantal environment, verbs of the [a]-theme remain segmentally identical to the disyllabic unsuffixed stem, whereas verbs of the [i]-stem are reduced to a monosyllabic stem [CCVC]. Before vowel-initial suffixes, verbs of the [i]-stem have their second vowel syncopated. The same thing happens in the [a]-theme, but only before the [-at] suffix; before [-u], the stem remains identical. These alternations are summarized in (31):

The phenomena to explain are syncope and its sensitivity to vowel quality and syllable structure. For the vowel [i], Kager (1999) formalizes a constraint banning this vowel in an open syllable (based on Brame 1974). In the present framework, one can also understand why /i/ has to syncopate in an open syllable: its position is governed. This is shown in (32) for both cases of /i/-syncope. For the unsuffixed case /misik/, one may assume, as Kager does, that because the vowel is stressed it cannot be syncopated.

(32)

Syncope of /i/ in unstressed, governed syllable ^[19]

a.	/misik-t/ => [msikt]	b.	/misik-u/ => [misku]

The same cannot be said of the syncope of the vowel /a/ in [katbat] “she wrote” vs. [kátabu] “they wrote”, if only because unlike [i]-syncope, [a]-syncope is not a general phonological process of the language. Rather, it seems that it is the difference in prosody between the suffixes [-at] and [-u] that must be responsible for the syncope in this case. A possible explanation is the following. There is a requirement for base+agreement-suffix to be prosodified under a single foot. As has been claimed many times for various dialects of Arabic (see Watson 2011 for an overview), final vowels are extrametrical, so the sequence /katab-u/ is prosodified [{kata}b-u], and no violation of the prosodic requirement is incurred. In contrast, the vowel of the suffix /-at/ is not final and thus cannot be extrametrical. Thus, in order to prosodify the entire sequence /katab-at/ into one foot, one of the vowels must be syncopated. Syncopated positions must be governed, so either the first or the second /a/ can syncopate. Since initial clusters are more marked than medial, heterosyllabic ones, it will be the second /a/ that will syncopate to yield [{katbat}]. For this reasoning to be generalizable, one must assume that /i/-syncope holds regardless of metrical considerations; otherwise one would expect *[mísiku].

5.2 Weak-final verbs

The table in (33) presents the inflection of the two types of weak final verbs, contrasting them with their strong counterparts:

As can be seen in (33), weak-final verbs are well-behaved in terms of stress. They are also well-behaved in terms of [i]-syncope: the first underlying /i/ of /nisi+t/ is not stressed in the surface [nsi:t], and therefore syncopates under government. Like strong verbs, weak-final verbs exhibit a single stem in the imperfective. There is one difference: the final vowel of both types of weak-final verbs does not surface when a suffix is attached: /tinsa+i/ => [tinsi] “that you (fmsg) forget”; /tirmi+u/ => [tirmu] “that you(pl) throw”. In the perfective, we observe the same phenomenon, but only in the [a]-theme: /rama+u/ => [ramu] “they threw”, but /nisi+u/ => [nisyu]. Another important fact is that the preconsonantal weak-final verb has a long vowel, [e:] in the [a]-theme and [i:] for the [i]-theme. There are thus three phenomena to explain:

1. The disappearance of the second (or only) stem vowel in certain cases, but not in others;

2. The lengthening of the preconsonantal vowel; and

3. The quality of the preconsonantal vowel.

Now recall our question regarding weak verbs. What does the speaker know about the missing element in weak verbs? Chekayri and Scheer (1996) is devoted to answering this question for CA. Their main claim is that such verbs in fact have a null element in the weak position. The position is filled in a manner that is predictable from its environment. Roughly the same claim can be made for PA, as we will now see.

In (34), the preconsonantal weak-final forms are presented in the same template as the preconsonantal triradical in (32a) above. As we saw in the previous section, in weak-final forms there is no doubt whether the vocalization of V₂ is a copy of V₁ or present lexically in V₂: if it is [a] – it originates in V₁, and if it is [i] – it originates in V₂. For this reason, the representation of the [a]-theme weak-final verb [rame:t] “I/he threw” in (34a) has a floating /a/ above V₁, while V₂ is empty. The parallel representation in (34b) has /i/ above the V₂ position. What makes weak-final verbs special is that the last consonantal position C₃ is also empty. According to Chekayri and Scheer, this position may be filled by apophony or by spreading from adjacent vowels. In (34a), there is never an adjacent vowel – an independently-established fact as the input to the imperfective apophony is /ø/ – and the result is that the ø in C₃ itself apophonizes to become /i/. The V₁ /a/ will eventually occupy also V₂, and the combination of the copied /a/ and that /i/ gives the long [e:]. ^[20] In (34b), there is always an adjacent lexical vowel /i/; the C₃ position is simply engulfed by the spreading of that vowel and may therefore remain unassociated. ^[21]

(34)

Weak-final verbs: pre-consonantal forms

a.	[rame:-t] ‘I/you(msg) threw’	b.	[nisi:-t] ‘I/you(msg) forgot’

With the representations above, both the length and the quality of the preconsonantal vowel have been explained. No underlying phonemic identity has to be attributed to a third element in the weak-final root.

The unsuffixed forms, whose representations we have already encountered in the previous section, do not necessitate postulating a specific phonemic identity for the third element. An equally probable assumption is that the edges of templates simply drop when not identified, as in (35). ^[22] The imperfective vocalization is obtained as claimed, by apophony from the perfective base.

(35)

Weak-final verbs: pre-consonantal forms

a.	[rama] ‘he threw’	b.	[nisi] ‘he forgot’
c.	[yirmi] ‘that he throw’ (ø=>i)	d.	[yinsa] ‘that he forget’ (i=>a)

We may now turn to the elision of certain vowels in contact with vowel-initial suffixes. As we saw, [a] is always elided, whereas [i] is only elided in the imperfective, where it is the result of apophony. In the perfective, where [i] is lexical, it does not disappear before vowel-initial suffixes. As we will now see, these facts confirm the analysis of the weak-final [a]-theme as positioned in V₁ and the [i]-theme as positioned in V₂.

In the unsuffixed stem [rama] in (35a) above, the second vowel is a copy of the first; in consequence, when there is no need for copy, or when copy would create an illicit hiatus, it simply does not occur. Such a case is presented by suffixed [ram-u] in (36a): the suffix vowel (which again is to be analyzed as long, McCarthy 2005) is added to the rightmost edge of the template and occupies the two rightmost V positions. As a result, there is no position requiring to be filled. The rightmost stretch of the template, the suffixal CV, is dropped. Contrast this to the situation in (36b), where the vowel is not copied, but lexically associated to V₂. In this case, the suffix cannot replace the base vowel. The /i/ must be realized. To resolve the resulting hiatus, /i/ branches onto C₃. This permits the syncope of the governed branch and its dissociation from V₂. The forms of both imperfectives can be achieved by stating that apophony, whether its output is [a] or [i], simply does not occur when the position can be filled by a suffixed vowel (as it would create a hiatus).

(36)

Weak-final verbs: pre-consonantal forms

a.	[ram-u] ‘they threw’
b.	[nisy-u] ‘I/he forgot’
c.	[yirm-u] ‘that they throw’
d.	[yins-u] ‘that they forget’

Thus, the distinction between a lexically-positioned vowel and one derived by apophony receives empirical support. Even though the vowel is [i] in both [nisi] “he forgot” and [yirmi] “that he throw”, the [i]’s behave differently before the same plural suffix [-u].

In none of the representations above were there phonemic properties assigned to a third element in the root. Rather, the independently-established lexical, default or apophonized status of the perfective or imperfective vocalizations, in combination with the suffixal environment, determined how the verb would be realized.

For completeness, it is worth to examine the situation of weak-final verbs in measures other than M1. The representative measures used in this paper are repeated in (37), contrasting strong and weak verbs:

As can be seen again in (37b), strong verbs in M2-5 behave like the [a-i] class of M1, while M6 always has [a] in V₂, regardless of aspect (this vowel is in bold). Weak verbs conform to exactly the same generalizations: all weak-final verbs in M2-5 behave like the [a-i] class of M1, while those in M6 have a stable [a] in V₂. If somehow weak-final verbs had anything other than /ø/ as a third radical, we could expect a different behavior in these different environments. But in reality, the situation is perfectly compatible with the main claim of the subsection, namely that weak radicals are simple ø, as claimed in Chekayri and Scheer (1996, 2003). Whereas these authors made that claim for a standardized, unspoken language, in the present paper it is shown to hold in synchrony, too, in the PA system. ^[23]

So far, the emergent status of weak radicals has been proven for weak-final verbs only. In the next subsection, hollow verbs are analyzed along similar lines.

5.3. Hollow verbs

As we saw in the section on the inflectional bases, hollow verbs appear only with the [a]-theme. Discussing only the inflectional bases disregards the interesting behavior that these verbs display in their preconsonantal form of the perfective. As shown in (38), whereas in strong verbs the same stem is used throughout the perfective (with the exception of the syncope before [-at]), in hollow verbs there is a special stem in the preconsonantal position, showing the same apophonized vowel as in the perfective, only short. Interestingly, Chekayri and Scheer (2003) leave the similar fact from CA for further investigation, which is yet to be initiated.

We have seen that weak-final verbs also have a specific form for the preconsonantal stem; but it was the missing consonant that underwent apophony, not the vocalization… Why would there be apophony of the vocalization in the precosonantal position? As we will see presently, in hollow verbs, too, it is the consonantal position that is apophonized - but the syllabic conditions obscure this fact.

First, consider the unsuffixed and prevocalic forms of the imperfective paradigm, presented in (39). In the unsuffixed forms (39a,b), the theme provides the vocalization. It is a fact about PA that it allows final CVVC syllables, so there is nothing wrong with the representation. In the prevocalic stems in (39c,d), the same long vowel stands in a non-final open syllable. In such a configuration, as shown in the representations, the right branch of the long vowel is governed by the following nucleus. We know that PA allows such syllables too. Note that in all these representations the vowel is long, and C₂ is engulfed and need not be realized.

(39)

Prevocalic and unsuffixed hollow verbs in the perfective

a.	[ṣa:r] ‘he became’	b.	[ka:n] ‘he was’
c.	[ṣa:rat] ‘she became’	d.	[ka:nat] ‘she was’

We know at the same time that PA, like many other languages, does not allow long vowels in closed syllables. Within CVCV, this was formalized in Lowenstamm (1996) by stating that the nucleus under the right branch of non-final long vowels (V₂ in 39c,d) has to be governed by a following nucleus. If it isn’t, it cannot be realized. This of course formalizes the problem of preconsonantal stems, shown in (40). If the second nucleus of the long vowel cannot be realized, C₂ is no longer engulfed and cannot remain silent.

(40)

Preconsonantal hollow verbs: impossible for a vowel to branch onto V₃

a.	[ṣirt] ‘I/you(msg) became’	b.	[kunt] ‘I/you(msg) were’

We have seen a similar situation in weak-final verbs: the consonantal position was then realized by apophony. When the theme was in V₁, the input to apophony was ø. This gives the prediction in (41c) for verbs like [ṣa:r]: C₂ will be realized as /i/. We would thus expect the form to be realized as /ṣayrt/. Because diphthongs cannot survive in closed syllable, the vowels will coalesce to /e/. But as we also saw, mid vowels are never found in CVCC syllabifications. Elsewhere, one finds [i] in their place, as indeed one finds here. In other words, apophonizing C₂, when combined with the general rules of the language, it yields the correct [ṣirt]. Contrast this to the situation depicted in (41b), in which the theme is independently known to be situated in V₂ (by the fact that it is the input to imperfective apophony). Since V₂ vowels can be apophonized, in order to fill the C₂ position, the lexical vowel is apophonized and gives /u/. We expect /kawnt/, which by the same generalizations as its unround counterpart is correctly predicted to be realized as [kunt].

(41)

Preconsonantal hollow verbs: apophony

a.	/ṣert/=> [ṣirt] ‘I/you(msg) became’	b.	/kont/=> [kunt] ‘‘I/you(msg) were’

Note that the same vowel that appears in the preconsonantal perfective also resurfaces in the imperfective [ṣi:r], [ku:n]. This fact follows from our understanding of the mechanism at work in realizing the ø consonant, namely the apophonic chain. More precisely, in the present proposal such positions needn’t be realized if a long vowel engulfs them. But if that is not possible, then the position will be filled by apophony from the theme in V₂, if there is one. If there is no theme in V₂, the basis of the apophonic derivation will be ø. The logic is the same for both types of weak verbs.

This section explored the inflectional paradigms of PA verbs. We have seen that apophony is a central mechanism in this domain, too. Most notably, this argument was made for the inflectional paradigm of weak verbs, which has only been partially treated in the past. Chekayri and Scheer (1996, 2003) have proposed for CA that all weak verbs in fact have a null element in the weak position, which is in some cases realized according to the apophonic chain. That proposal was shown to be correct for PA too, and was successfully extended to the inflection of hollow verbs, to which it has never been applied.

We can now turn to the problem introduced in the beginning of the paper, regarding the inflection of biradicals.