This means that in the binyan system there is a degree of syncretism, with some pat-‐
terns displaying more syncretism than others (cf. Ch. 4).
In a nutshell, there is no direct mapping from verbal semantic properties to binyan forms. On the one hand, the same binyan may host verbs of different semantic types (with the exception of binyan IX, which is associated almost exclusively with in-‐
choatives). On the other, verbs of the same semantic type (e.g., causative) may appear in different binyanim (e.g., I, II, III). While rejecting the correlation, so neatly presented in grammars, between certain binyanim and specific interpretations, one cannot fail to acknowledge that there are pockets of regularity in the binyan system, as has already been observed in Borg (1981, 1988). In Ch. 4, I show that argument alternations are to a large extent morphologically regular and quantify the regular and irregular share in the system. Before, however, I introduce the second assumption underpinning this work, which helps predict and explain the regularity and irregularity found in Maltese tem-‐
platic verbs.
3.3 Root derivation and word derivation
One central argument put forward in this chapter is that the binyan system in Maltese is neither perfectly transparent nor completely opaque. When looking at a large body of data, it turns out that, while the semantic relations between certain binyanim created from the same root can be predicted, there is a great deal of variation and idiosyncrasy in the system. Interleaving a root with different binyanim may result either in regular changes in the semantic and syntactic configuration of the verbs they form, or in verbs that are semantically unpredictable and too distant from each other to allow for a deri-‐
vational account, even if they share some phonological material, e.g., the consonantal root.
To account for these regular and irregular aspects within the binyan system, I posit, following Marantz (1997, 2001, 2007) and Arad (2003a, 2003b, 2005), a distinction in the syntactic composition of words, which in several respects echoes the distinction found in traditional lexicalist theories of morphosyntax between lexical and syntactic word formation (cf. Siegel 1974; Wasow 1977; Kiparsky 1982; Spencer 1991; Giegerich 1999; inter alia). Words formed in the lexicon are more prone to phonological and se-‐
mantic irregularities; they often receive an idiosyncratic interpretation and lexical pho-‐
are merged. Once this interpretation is assigned, it is carried along throughout the deri-‐
vation”. In other words, while root-‐derived verbs may pick numerous interpretations in different environments, word-‐derived verbs are tied to the meaning of the verb, noun or adjective from which they are derived. They have no access to the root, but only to the words they are derived from (cf. Marantz 1997; Embick 2008; Alexiadou 2009; inter alia).
In order to motivate the generalization that root-‐derived words are more likely to display morphonological irregularities and semantic idiosyncrasies than word-‐derived words, Arad (2005: Ch. 7) draws a connection with the Chomskyan condition on phase impenetrability (Chomsky 2001, 2004, 2005), even though she does not fully discuss the implications of linking the generalization to phases. The main idea is that the com-‐
bination of a root with a category-‐assigning head defines a phase, and phases are gen-‐
erally considered to be impenetrable once they have been spelled out. In short, the form and meaning of a root-‐derived word are set immediately after it is formed by the spell-‐
ing-‐out of the phase. Any further (word) derivations are compositional and do not have access to the root, but only to the morphonological and semantic interpretation of the base form they are derived from (cf. Marantz 2007). This connection to phases has be-‐
come something of an Achilles’ heel for the Marantz-‐Arad approach because, as dis-‐
cussed in Sect. 3.3.2, it makes predictions that turn out to be somewhat restrictive.
In addition, word formation from roots is quite often non-‐productive, as the combi-‐
nation of a root with a particular category-‐defining head but not with others is to a large extent arbitrary. Recall the examples from English in Sect. 2.2.2: the root √RELIG
arbitrarily combines with the adjectivalizing head -‐ious (relig-ious), but not with -y (*relig-y), and √STICK arbitrarily combines with the head -y (stick-y), but not with -‐ious (*stick-ious). It is therefore expected that word derivation, in contrast to root deriva-‐
tion, displays no or, at any rate, very few gaps.
The main differences between root-‐derived and word-‐derived words are listed in Table 3.3 (cf. Marantz 2001, 2007). From this follows the second general assumption in this study. The prediction is that root-‐derived words are more prone to phonological and semantic irregularities (special phonological processes, idiosyncratic and idiomatic meanings, etc.), while word-‐derived words are morphonologically regular and have semantically predictable meanings.
Table 3.3 Key differences between root derivation and word derivation
Root-‐derived words Word-‐derived words
merger with a category-‐assigning head merger above a category-‐bearing head
idiosyncratic, idiomatic meanings predictable meanings
apparent semi-‐productivity apparent complete productivity
independent of argument structure possible operations on argument structure
The Marantz-‐Arad model makes the correct predictions with respect to word for-‐
mation processes in a number of languages, such as zero derivation in English and Dutch, denominal verbs (as well as other verbs) in Hebrew, and plural formation in Amharic. I now discuss each one of these cases (Sect. 3.3.1), and then turn to an exami-‐
nation of the phenomenon in the binyan system in Maltese, motivating the use of the framework in this study (Sect. 3.3.2). Finally, in Sect. 3.3.3, I tackle one weak point of Arad’s analysis, concerning the impenetrability of the root and categorizing head.
3.3.1 Cross-‐linguistic evidence
Arad (2005) argues that the locality constraint introduced above is a universal prop-‐
erty. In other words, (i) the ability to be assigned multiple interpretations is strictly re-‐
served to roots, and (ii) the assumption that once the root merges with a category head and forms a word, its interpretation is fixed and is carried along throughout the deriva-‐
tion, constitute a constraint that holds across all languages. On the other hand, the abil-‐
ity of roots to be assigned a variety of interpretations in different morphological envi-‐
ronments is language specific. For instance, while the roots in a language like Hebrew typically appear in various verbal (and nominal) environments, the roots in a language like English generally appear in only one verbal environment.
In view of the claim that the locality constraint on the interpretation of root-‐derived and word-‐derived words is a universal property of languages, in this section, I review some studies which put to the test the syntactic distinction between root derivation and word derivation in four languages from two typologically distinct families. I first con-‐
sider studies on zero derivation in two Germanic languages, English and Dutch. Follow-‐
ing that, I turn to studies on Semitic languages, namely plurals in Amharic and denomi-‐
nal verbs in Hebrew.
Zero derivation
Zero derivation or conversion, that is items which can function as different grammatical categories, is a good testing ground in which to start a discussion on category-‐less roots that receive their grammatical category when inserted in syntax. A typical example is verb-‐noun pairs like searchV and searchN or paintV and paintN. In such cases, it is as-‐
sumed that the roots √SEARCH and √PAINT are combined with a verbal or nominal functional node, depending on the syntactic environments in which a verb or a noun is required. The relation between such pairs thus consists in their common underlying root, with their separate categories taking shape in syntax.
However, in a number of studies it is argued that not all verb-‐noun conversion pairs constitute root level derivations. Rather, one form is more basic than the other; that is, they are either nouns derived from verbs or verbs derived from nouns. Building on ob-‐
servations by Kiparsky (1982), Myers (1984), and others, Arad (2003a, 2005) distin-‐
guishes between root-‐derived and noun-‐derived verbs on the basis of stress assignment and semantic relations in English zero-‐related pairs. The generalization is that when the verb and the noun share phonological properties, such as the same stress pattern, as in cómmentN,V, dísciplineN,V and cóntactN,V, there tends to be a tight semantic relation be-‐
tween the two members of the pair. By contrast, in pairs like óbjectN and objéctV, récordN and recórdV, cóntestN and contéstV, not only is stress assignment different in the verbs and the nouns, but they also stand in a looser semantic relation to each other. In light of the assumptions made above, the conclusion is that the latter cases are root-‐derived while the former are word-‐derived. In short, being derived from a common root, pairs like óbjectN and objéctV have the typical nominal and verbal stress in English and distant semantics. Pairs like cómmentN,V, on the other hand, are noun-‐derived because they have the same stress properties (the typical nominal stress in English) and the semantic relation between the two members is clear.
A similar directional approach to verb-‐noun conversion pairs is taken by Don (2004, 2005) and Kraaikamp (2008) for Dutch (cf. also Acquaviva 2009). Two proto-‐
typical properties of nouns and verbs in Dutch are gender and conjugational class.
Nouns are either neuter, as in het huis ‘the.NEUT house’, or non-‐neuter, as in de weg
‘the.NON-‐NEUT road’. Verbs are either regular, if they are stem-‐invariant, or irregular, if they display stem vowel alternation. Compare, for example, the regular present–past–
participle triplet of ‘to work’, werk – werk-te – ge-werk-t, with the irregular triplet of ‘to walk’, loop – liep – gelopen.
Now, it has been shown that, in the case of zero derivation, the verb is regular if it is derived from a noun, and the gender of the noun is non-‐neuter when it is derived from a verb. Considering this, the assumption, then, is that irregular verbs and neuter nouns, being root-‐derived, have more or less deviant semantic interpretations as opposed to word derivations. This prediction is in fact borne out.
To take an example, consider the noun-‐verb pair prijs ‘price’ and prijzen, which has two meanings: a denominal interpretation ‘put a price on’, and the meaning ‘to praise’, which cannot be considered to be transparently derived from the noun. As expected, when used in the denominal meaning, the verb is regular, but when used in the ‘to praise’ meaning, it is an irregular verb (cf. Kraaikamp 2008: 12-‐13). This is taken as evidence that prijzen ‘to praise’ is a root derivation, whereas prijzen ‘put a price on’ is derived above the root level, i.e. it is derived from the noun prijs ‘price’, as it inherits both the semantics and phonology of the base noun. Thus, as in the case of English zero-‐
related pairs, morphonological regularity seems to go hand-‐in-‐hand with a semantics that suggests a derivational relation with the word, rather than with the root.12
Plural nouns and denominal verbs
Let us now turn to data from two Semitic languages, starting with Amharic plurals. Am-‐
haric has a regular and an irregular way to form the plural of nouns. The regular forma-‐
tion involves attaching the suffix -očč (or one of its allomorphs, generally conditioned by nouns ending in a vowel) to a noun like bet ‘house’, giving bet-očč ‘house-‐PL’. Many nouns, however, take the so-‐called irregular plural markers, which may be external, namely the suffixes -an, -at or -t, or internal, i.e. ablaut and a change in the prosodic template (cf. Leslau 1995). Once again, irregular morphology runs parallel to unpre-‐
dictable semantics, as the examples in (9) demonstrate.
(9) wär ‘month’ wär-at ‘season’
näfs ‘soul’ näfs-at ‘small insects’
12 However, this does not mean that any morphonological irregularity corresponds to semantic idiosyn-‐
crasy. Take, for instance, verb inflection in English: the interpretation of an irregular past tense form like broke is just as predictable as that of a regular verb like braked. Such examples seem to rest uncomfortably with the assumption in Marantz (2007) that the domain for morphonological irregularity is the same as the domain for semantic idiosyncrasy. Distributed Morphology generally accounts for this by invoking ‘Read-‐
justment rules’ to change break to broke, etc. Embick (2010) deals with this mismatch between morpho-‐
nological irregularity and predictable interpretations by delaying the phonological spell-‐out of morpho-‐
syntactic features until the next cycle head is merged. See also the discussion in Svenonius (2011).
There seems to be a lot of variation in the semantic relations between the singular and the (root-‐derived) irregular plural forms. The relation may be reasonably straightfor-‐
ward, as in a set of months making up a season, or more tenuous, as in souls and small insects. In addition, according to Kramer (2009: Ch. 4), the (word-‐derived) regular forms are very productive; that is, there are no paradigmatic gaps in regular plural for-‐
mation. “All nominals in Amharic may be regularly pluralized, even if they also have an irregular plural” (Kramer 2009: 145). By contrast, irregular plurals are non-‐productive and have paradigmatic gaps, which suggests that they are formed by the combination of a root with n[+PL].
Finally, let us consider Hebrew denominal verbs, which take us back to binyan morphology. Arad (2003a: 745-‐747, 2003b: 89-‐90, 2005: 245-‐247) shows how a root like √sgr interleaves with various verbal and nominal patterns, some of which are listed in (10).
(10) √sgr C1aC2aC3 (v) sagar ‘to close’
hiC1C2iC3 (v) hisgir ‘to extradite’
C1oC2C3ayim (n) sograyim ‘parentheses’
miC1C2eC3et (n) misgeret ‘frame’
CiC1C2eC3 (v) misger ‘to frame’
Particularly interesting for our discussion is the verb at the end, misger, which bears not only morphonological similarity to the noun misgeret (i.e. the typically nominal pre-‐
fix m-, apart from the same root consonants), but also shares an interpretation with the noun from which it is derived. Arad (2003b: 90) notes that “[i]t may seem natural or even trivial that the verb made from the noun frame means to frame. But this, I argue, is a crucial property of noun-‐derived verbs.” While the root-‐derived verbs sagar and his-
gir are assigned different, hardly related interpretations, misger, which is word-‐derived, as suggested by the presence of the prefix m- that is carried over from the nominal pat-‐
tern miC1C2eC3et into the verbal form, is semantically tied to the noun it is derived from.
On the basis of this data, Arad (2003b: 90) concludes that:
Although the verb misger contains the consonants of the root √sgr, it cannot have access to the underspecified core meaning of the root, or to all the interpretations assigned to that root in dif-‐
ferent environments: something seems to interfere between the verb misger and the root √sgr.
This interfering element, I argue, is the noun misgeret.
With this cross-‐linguistic evidence in hand, we can now examine whether the Aradian predictions also hold for the Maltese verbal system. This is the topic of the next section.
3.3.2 Word formation in Maltese
In this section, I demonstrate that the transparent (regular meaning associations, phonological similarity, and full productivity) and opaque aspects (idiosyncratic, un-‐
predictable meanings, phonological irregularity, and gaps) of the binyan system in Mal-‐
tese can be explained by drawing a distinction between word formation from roots and non-‐roots, along the lines of Marantz (2001, 2007) and Arad (2003a, 2003b, 2005).
More specifically, on the basis of morphonological and semantic evidence, I argue that argument alternations (namely the active-‐passive and transitive-‐reflexive alternations) are word-‐derived, whereas all other instances found in the binyan system, including the phenomenon of multiple interpretations (cf. Sect. 3.2.1), roots which derive singleton verbs, and roots which derive verbs with identical meanings (cf. Ch. 4), are root deriva-‐
tions. The status of the causative-‐inchoative/noncausative alternation with respect to root derivation vs. word derivation is discussed in Ch. 4 and Ch. 5.
Consider, to begin with, the data in (3) above, reproduced here as (11), which pro-‐
vide an example of a root creating different verbs and nouns, some of which are semantically close, such as xorob ‘drink’, inxtorob ‘be drunk’, xarba ‘a drink’, and others which are not, such as xorob ‘drink’ and xarrab ‘wet’. (I will, for the time being, pass over the secondary meaning of inxtorob ‘shrink’. The problems posed by this and similar verbs are discussed in some detail below).
(11) √xrb C1vC2vC3 (v) xorob ‘to drink’
C1vC2C2vC3 (v) xarrab ‘to wet’
nC1(t)vC2vC3 (v) inxtorob ‘to be drunk/shrink’
C1vC2C3a (n) xarba ‘drink’
tvC1C2iiC3 (n) tixrib ‘wetting’
Interestingly, xorob ‘drink’ and xarrab ‘wet’ are not only semantically distant, but also have a different vowel pattern (cf. Ch. 4 for discussion on vowels). By contrast, the vow-‐
els of xorob ‘drink’ and inxtorob ‘be drunk’ are the same and the semantic relation be-‐
tween the two is straightforward: they mark a regular change in the structure of the arguments, namely the active-‐passive alternation, as the sentences in (12) illustrate.
(12a) It=tifel xorob l=ilma.
DEF=boy drink.PFV.3SG.M DEF=water
‘The boy drank the water.’
(12b) Matul ir=Ramadan ix=xarba ta-l=ħarrub
terpart of ħareġ ‘take out’. Likewise, tħarreġ is derived above the root level and receives a phonological form (note the gemination of the second radical) and a semantic inter-‐
pretation based upon the binyan II verb, i.e. it is the reflexive of ħarreġ ‘train’.
Particularly significant in this respect is the fact that root-‐derived verbs like ħareġ are typically more polysemous than verb-‐derived verbs like inħareġ. While ħareġ can mean ‘to go out, take out, publish/issue, turn out, defray, break out, etc.’, its passive counterpart can mean ‘to be taken out, be published/issued, be defrayed’. Note that an active reading of each of these three meanings is (already) included in the base it is de-‐
rived from, supporting the assumption that word-‐derived words inherit interpretations that have been assigned above the root level.
To take another example, the binyan II verb niġġeż, which is directly derived from the root √ngż, means ‘to prick, hurt/upset, administer an injection, hook a fish which snaps off, etc.’ By contrast, binyan V tniggeż, being verb-‐derived, is tied to the interpre-‐
tations of niggeż, and receives either a passive ‘to be pricked’ or a reflexive ‘to prick oneself’ interpretation. These and many other cases where root derivations, unlike word derivations, are highly polysemous suggest that verb-‐derived verbs take as their input items whose meaning has already been fixed at the first merging of a root with a category-‐assigning head.14
As a side note, we can put the syntactic distinction in word formation to test also on some concatenative verbs and nouns in Maltese. Consider the triplets in (13).
(13a) iwweldja ‘to weld’
iwweldjar ‘the act of welding’
welding ‘soldering alloy, the act of welding’
(13b) iċċita ‘to quote, cite’
iċċitar ‘the act of quoting, citing’
ċitazzjoni ‘citation, summons, fine’
(13c) aġixxa ‘to act, behave’
aġir ‘way of acting, behaving’
azzjoni ‘action, behavior, share, (religious) action’
14 In a discussion on affixal polysemy (i.e. that an affix like -ize can sometimes mean ‘cause to become x’, e.g., privatize, sometimes ‘cause to go into x’, e.g., containerize, sometimes ‘perform x’, e.g., anthropologize, etc.), Lieber (2004: 43) relates polysemy to semantic underspecification, claiming: “One reason that affixes tend to be highly polysemous is that their actual semantic content is vastly abstract and underdetermined.”
This fits neatly with our analysis that consonantal roots are underspecified and words directly derived from roots tend to be more polysemous than words derived from (semantically and phonologically speci-‐
fied) words.
It appears that the difference between the -‐ar/-ir suffixed nouns and the -ing/-zzjoni nouns (e.g., the heading Applikazzjoni retroattiva ta’ politika ġdida dwar iċ-ċitar ‘Retro-‐
It appears that the difference between the -‐ar/-ir suffixed nouns and the -ing/-zzjoni nouns (e.g., the heading Applikazzjoni retroattiva ta’ politika ġdida dwar iċ-ċitar ‘Retro-‐