ISLAND MORPHOLOGY: MORPHOLOGY’S INTERACTIONS IN THE STUDY OF STEM PATTERNS

ISLAND MORPHOLOGY: MORPHOLOGY’S INTERACTIONS IN THE STUDY OF STEM PATTERNS The paper discusses the notion of morphological complexity, with a focus on stem patterns. Stem patterns, creating stem-based inflectional classes, are morphological constructs which come about as a result of observing the patterns rendered by the stem-form alternations (or stem splits (Baerman/Corbett forthcoming)), which one extracts after the formation of word-forms within paradigms. Stem-based inflectional class formation constitutes one aspect in the analysis of non-canonical paradigms, which also include affix-based inflectional classes, syncretism, defectiveness, and overabundance Corbett 2005, 2007, 2009; Baerman, Brown and Corbett 2005; Thornton 2010). While these non-canonical instances are in themselves interesting to observe, it is even more intriguing to be able to see what interactions can arise, which at times do not seem to be the result of something exterior to morphology proper. Through data taken from Maltese verbal paradigms the phenomenon of stem-based inflectional classes will be explored, which will exhibit how internal to the paradigm there exists a complex system in itself, which is based on the distinct organisation of different conflated morphosyntactic features which come about via syncretism. These patterns should illustrate a paradigm-internal morphological phenomenon that is irrelevant to the syntax, where while morphology borders with it, there need not be any interaction at this interface. At the same time, it will also be shown how at times, the border with phonology is blurred, where while the phonology may often try to build bridges that interface with the morphological island, the island’s internal forces that drive its autonomy may deem to be more superior than the phonology’s strive to impose its interacting requirements, which render some interesting morphophonological mismatches as a result. Lingvistika-2011-01-93 1/5/12 1:32 PM Page 84


INTRODUCTION
Discussing island morphology, or morphology in itself, for that matter, entails the presupposition that morphology exists as a component in the grammar that has its own complexity and idiosyncrasies, and need not always interface and interact with others.The use of the metaphorical term island morphology should be equated to an autonomous level of morphology where complexity can be observed. 1he morphological complexity at issue here is the redundant type; that which makes no contribution to the form-meaning relation, and can be treated as a system of 'pure morphological functions' that have an independent 'life of [their] own' (Brown 2008: 6).The study of such functions, which has to do with the observation of how forms change, and what patterns result from this, which essentially have no value/relevance to the syntax and/or meaning, is referred to as morphological features (Corbett/Baerman 2006: 232).Asserting that such features have no interaction with syntax, the study will also probe the extent to which such features are allowed to border with phonology.As will be shown, the significance and autonomy of such features is best displayed in instances where the phonology seems to predict a pattern with which the morphology does not abide, thus resulting in a mismatch between the phonological and morphological forces.
What follows in section (2) is a discussion on morphological features, and their significance in a canonical typology account within the context of canonical paradigms.Section (3) briefly discusses the traditional distinction across verbs in the Semitic part of Maltese (henceforth SM), which will be the source of the paradigmatic data. 2 This is then followed by the verbal paradigmatic data which will show how syncretic stem indexations render patterns that may in turn vary across lexemes, thus creating stem-based inflectional classes; the subject matter at hand.Section (4) summarises the findings and concludes the paper.

Form Classes
Morphological features are best illustrated through the study of form classes.The analysis of form-classes looks at output realisations; focusing solely on their surface structure (refer to Anderson's (2011) reply to Corbett (2011)), and observes the patterns formed across distinct segments, such as affixation, tone variation, and/or stem-forms.The study related with the autonomous morphological system displayed by the affixal material, or the relation of this with tone patterns, is referred to as the analysis of inflectional classes, while that involving stem-forms, or the relation of stem-forms with affixes and/or tone patterns, which create some sort of pattern in the paradigm's cells grouping, is referred to as syncretic indexing (Corbett 2004(Corbett , 2007(Corbett , 2009)), (Baerman 2011), (Corbett/Baerman 2006, 2010), (Baerman/Corbett forthcoming).These features can be understood as distinct parameters of complexity.The former can be thought of as involving a horizontal parameter that looks across lexeme sets, and observes any comparisons/contrasts in the patterns displayed by the affixal material, while the latter, a vertical parameter that looks at the complexity which arises from within the paradigm.In both instances the patterns come about as a result of the different organisation of the array of inflected and/or lexical material in a given paradigm or across paradigms.Such organisatory patterns come about through different groupings of the surface forms as a result of their morphological sameness -syncretism, when these should in theory be distinct, since they are essentially the realization of distinct morphosyntactic features (Corbett 2005) (Baerman et al. 2005).The reason for assuming that such form classes are morphological features and hence irrelevant to the syntax (Alexiadou 2004: 21), is that, for example, the syntax does not care whether the particular affix which is being used is solely related with a set of lexemes and not another, as long as the correct word-form and realisational affix is selected to satisfy the morphosyntactic requirement.
The discussion here will pivot both the vertical and horizontal parameters mentioned above.On the vertical parameter we will observe the grouping of paradigmatic cells on the basis of the pattern rendered by the stem-form alternations in the different cells based on the syncretic stem index features.To this, a horizontal dimension will be added, which will illustrate the different paradigmatic stemalternation patterns that cut across different lexical classes, which, as will be shown in section (3.2), may or may not be phonologically-conditioned, and whose distribution across the classes is also 'syntactically-unmotivated' (Corbett 2009: 5).The interaction of both the vertical and horizontal parameters will result in the analysis of stem-based inflectional classes (hereafter SBICs), which are taken to be analogically parallel to the inflectional classes mentioned above, but which are derived from observing stem patterns instead of affixal material patterns3 .In this study, stem-form patterns from SM will be looked at in relation to their interaction with affixes, as well as traditional verbal classifications, described in section (3.1), which rely heavily on an account based on a reference to the underlying representation (UR), consisting of a reference to an underlying triconsonantal root (C-root) morpheme.By virtue of this SBIC analysis of verbal paradigms in SM, verbs will be categorised systematically, not anymore on the basis of an abstracted UR, which may at times, not even be representative of the differences or similarities across verbal sets, but rather on the basis of their surface structure stem patterns and their distinct ways of interaction with other elements in the word-form, such as the inflectional affixes.

Stem patterns as a consequence of non-canonical paradigmatic effects
What is interesting about form-classes is that essentially these come about as a result of a non-canonical behaviour, or the interaction of non-canonical phenomena.The canonical approach to morphology can be understood as that manifestation of a morphological phenomenon/element, that unarguably represents that which best fits the set parameters or criteria (Corbett 2007: 9), which however need not be that which is prototypical, as it may even be non-existent (Corbett 2010: 3).Looking at that which is noncanonical means that one moves away from the canonical, and the further away one is from the canon, the more of a non-canonical instance it becomes (Corbett 2011: 2).
The non-canonical paradigmatic aspect which this study will be concerned with is internal non-canonicity, whereby while the lexical material in the paradigm i.e. the stem is expected to be the same; non-alternating (Corbett 2007: 10), this is not the case in non-canonical paradigms, giving rise to the presence of distinct stemforms which may pattern differently across lexemes, thus creating SBICs by virtue of an independent organisatory system.The independent organisatory system is determining factor, since different SBICs may have the same number of stem-slots, but differ solely on the organisatory pattern.The data will also provide instances where there is an interaction of alternating stem-forms with other non-canonical phenomena such as syncretism.As will be shown, this will be a rather important interaction as it will account for the mismatches that may exist between the number of designated stem-slots, which comes about as a result of the syncretic stem indexation, vs. the actual number of stem-forms in the paradigm, and will be responsible for the internal SBIC stem pattern variation.The discussion in section (3.2.4) will also include a case illustrating the interaction of internal non-canonicity with suppletion, where the stem-form is not only altered, but involves a radical change.
Apart from being non-canonical because the stem alternates and creates distinct patterns, there is an additional non-canonical behaviour in SM verbal paradigms.
The stem may not simply bear lexical meaning alone, but is at times also endowed with grammatical features.This happens when in a SBIC, such as that discussed in section (3.2.4), the stem alternates on the basis of a coherent 1^2 vs. 3 PERSon feature distinction in the PERFect sub-paradigm, where the stem-forms used for the lexeme żar 'visit', for example, are żor and żār respectively.These features in turn overlap with those realised by the inflectional affixes (Baerman/Corbett 2011).This is regarded as additionally non-canonical because in 'the canonical situation, lexical meaning (and only that) is conveyed by lexical material, the stem; grammatical meaning, and only that is conveyed by the inflection' (Corbett 2011: 112).

Degrees of non-canonicity
Having established that the presence and formation of stem alternation patterns is non-canonical, it will be shown that there is a gradience of non-canonicity, such that for example, the more stem-form overlap across the PERFect-IMPERFect subparadigms, i.e. the more the syncretic interactions, which will result in an increased feature conflation on the stem, the more non-canonical the pattern is.The same follows for the number of members and/or stem patterns a SBIC takes.While a SBIC may be the most populous, thus potentially more canonical, among the other classes, it may still, on the other hand, be fragmented in distinct stem patterns, thus perhaps making it less canonical, in that it does not display uniformity across its members.Another parameter on which to grade the classes' canonicity may potentially be related with the actual (in)stability of the SBIC's members (refer to sections (3.2.2) and (3.2.4)),where it seems to be more canonical to have members participating in one SBIC only, rather than in two.All these parametric ideas can thus be further imposed on this non-canonical phenomenon to be able to capture the variation and the behaviour pertaining to these distinct SBICs.

THE DATA 3.1 Verbs in Semitic Maltese
The traditional classification of SM and Arabic verbs is that, on the basis of their UR, there is a: Strong, deaf, and weak verbal distinction; e.g.kiber 'grow', mess 'touch', and mar 'go' respectively.The weak class of verbs is further divided into three, depending on the position of the weak consonant (i.e. a w or j) in the UR; initial (assimilative) wiret 'inherit' (w-r-t), medial (hollow) mar 'go' (m-w-r), or final (lacking/defective) qara 'read' (q-r-j).Ryding (2005: 55) treats these distinctions as inflectional classes, however, as the reader must realise, this term has nothing to do with that used in section (2.1) above.inflectional classes, as a term used in the wider literature, refers to the variation found in the realisation of affixal material across different classes of lexemes.This is not what we have here, however, as the inflectional material does not vary across these different types of verbs.What we do see however, in relation to the above verbal classifications is that at times, this distinction does in fact correlate with a distinct SBIC.Thus, for want of clarity, this term, referring to the traditional classification will be abandoned, and the term used in Aquilina (1973: 144); verbal bases, will be used instead.This term is in Maltese traditionally used to refer collectively to the distinct types of verbs that exist in SM with reference to their URs, such as C-C-C (strong), w-C-C (weak-initial), C-C i C i (geminate) etc.The use of the term here, whilst retaining the same verbal classification understanding, will do this not on the basis of an UR, but rather on the stem-shape, such as CVCVC (which would include both strong and weak-initial verbs), CV:C (for hollows), and so on.In this paper it will be shown that the taxonomy of SM verbal bases on the basis of their UR does not always crosscut/correlate with the classification that will be provided here, based on the observed surface stem-form patterns of the 1 st binyan verbs.On the other hand, in some instances, membership in a SBIC does correlate exclusively with a particular verbalbase.We will also see that while traditional Maltese grammars always distinguish a handful of verbs as irregular, such as ħa 'take' on the basis of an obsolete UR that includes the Arabic alīf as the C-root's first radical, it will be shown that in some cases, there is nothing that indicates, at least from the surface paradigmatic pattern that these should be distinguished from other regular verbs.A real instance of irregularity is, under the following classification, equated with a case of non-canonicity which follows from the discussion in section (2.2.1),where an irregular verb is understood as that which is a member of its own distinct SBIC or pattern.Thus, the classification of verbs, and the regular/irregular dichotomy will be motivated solely from that which can be observed from the surface paradigmatic patterns, based on a non-autonomous morphological system that seems to coordinate across surface forms through the creation of stem patterns, within and across paradigms.

The inventory of stem-based inflectional classes in SM
Before describing the SBICs, a brief overview of the data that will be used to illustrate the phenomenon will be given.The syncretic stem patterns which shall be looked at are derived from observations from the PERF-IMPERF sub-paradigm systems in SM.The PERF sub-paradigm in Maltese is characterised by suffixes which realise the conflated morphosyntactic features for PERS, NUMber, and ASPect, as well as GENDer where appropriate.In the IMPERF sub-paradigm prefixes realise PERS, ASP, and GEND, while suffixes realise NUM feature values.While the paradigm of a Maltese verb also includes an IMPERATIVE sub-paradigm, yet, this will not feature in the discussion here.Stem-alternations in SM verbal paradigms render their own stem-classes; i.e. different means with which stem-form alternations come about, which include syllable-structure type alternations, ablaut-changes, stemextensions, and suppletion.While it is only the end result that we will be concerned with here, i.e. the stem-alternation in itself, whenever required there will be a reference to the actual mechanism which is responsible for this change.
Recall from section (2.2) that the criterion responsible for distinguishing among SBICs is the varied syncretic stem indexing pattern they employ.Thus, while the number of stem-slots can be the same across two classes, they have to differ by virtue of the stem indexing pattern involved, as is the case for SBIC I and III in sections (3.2.1) and (3.2.3) respectively, which both have five stem-slots but whose syncretic pattern is grouped differently.We will also see how members of a given class need not be homogeneous, and may have varied stem patterns resulting from non-distinct stem-forms filling in separate stem-slots.Sections (3.2.1) till (3.2.6) below will illustrate and further discuss the individual six distinct SBICs in SM, and the variation involved.From the data which will be given below one would want to say that some classes come about as a result of a stem-alternation that is determined by morphotactic constraints that depend on the suffix's phonology, particularly in the PERF sub-paradigm.However, having said this, while the suffixes are generally invariant, we still get SBIC and stem pattern variation.Furthermore, if morphotactics determined patterns, then we would expect the stem-form used with V-initial suffixes in the IMPERF sub-paradigm to be always syncretic with the PERF stem-forms that are followed by a V-initial suffix themselves.This is not the case however.When mismatches of the sort arise, then we have a case of real morphological patterns which cannot be explained through any phonological considerations, thus highlighting the independent island morphology.

SBiC i
Stem-based inflectional class I is distinguished by having five stem-slots which require an alternating IMPERF sub-paradigm where the stem-pattern splits and is indexed for the NUM feature values SG vs PL; lqot-and lqt-, respectively in the laqat 'hit' paradigm, for example, and a conflated, non-motivated, non-coherent, morphomic split in the PERF, involving three distinct stem-slots that are indexed for 3SGM, 3SGF^3PL, and 1^2 (laqat, laqt, and lqat). 4This class happens to be the most populous, including almost all the distinct types of verbal bases in SM apart from the deaf CVC i C i base (as displayed in sections (3.2.1.1-3.2.1.3)).The biggest set of paradigms however, belong to the traditional strong class with verbal-base C [-w/j] VCVC. 5 This SBIC will illustrate the stem pattern variation that can exist within the same class, where while being unified by the same number of stem-slots and the syncretic stem-pattern in the members' paradigms, varied patterns are brought about as a result of overlapping forms across the PERF-IMPERF paradigms.What is interesting to observe is that while the strong triliteral class involves six ablaut classes (section 3.2.1.1),the pattern of variation does not necessarily cross-classify with the distinct ablaut-classes, and there may be distinct sub-pattern variations within the same ablaut-class.6

Strong verbal base ablaut classes a-a
From the above paradigm we can see that the stem-form in the IMPERF sub-paradigm alternates between the SG vs. PL NUM values.Thus, to economise on the space available, it is only the 3SGM and 3PL values for the IMPERF paradigm that will be displayed.In the case of the PERF sub-paradigm, the 1SG cell will be taken to represent the underspecified 1^2 PERS.On the other hand, all three 3 rd PERS slots will be displayed, even if in the above paradigm the 3SGF^3PL stem-forms are syncretic.The need to render all three 3 rd PERS cells is important as it is precisely on the basis of the distinct syncretic stem pattern oragnisation within these cells, that the PERF sub-paradigm of the distinct SBICs, varies.The paradigm for laqat 'hit' above, along with three others, to illustrate the variation in this ablaut-class, are given below.
An interesting variation can be seen across the different members of the a-a ablaut-class.Three distinct paradigmatic patterns are found, and are taken to represent sub-set of members within this ablaut class, which in turn transcends to having 7 ABCC represents the 1^2, 3SGM, and 3SGF^3PL stem-indexing in the PERF sub-paradigm and SG ≠ PL implies that the IMPERF sub-paradigm is alternating.The SG and PL stem-slots in the IMPERF are not given a letter due to the fact that while this pattern of alternation is retained across the members of this SBIC, the stem-forms that are actually present in these stem-slots may not be new, but may involve a partial or total overlap of forms used in the PERF sub-paradigm, as shown to be the case in ħataf 'grasp' and ħarab 'escape', respectively.
distinct sub-patterns within SBIC I. 8 The essential distinction between the laqat-talab and ħataf-ħarab sets is that the former's pattern arises out of having five distinct stem-forms filling the five designated stem-slots in this class, implying that there are no overlapping forms across the sub-paradigms.In the latter set, on the other hand, while ħataf has four stem-forms, which result from the syncretic stem-form across the PERF 1^2 and IMPERF SG realised by ħtaf-, ħarab only has three, since its IMPERF sub-paradigm displays total overlapped forms.Just like the ħataf class its PERF 1^2 and IMPERF SG stem-forms are syncretic (ħrab-), however there is additional syncretism across the PERF 3SGF^3PL and IMPERF PL stem-slots (ħarb-), which in turn, with both instances of syncretism, explain the three-fold stem-form alternation across the whole paradigm.
The significance of talab 'pray/ask' and ħarab 'escape', in relation to laqat 'hit' and ħataf 'grasp' respectively, (where laqat and talab display the same pattern) is that, the stem's second consonant (C 2 ) in talab/ħarab is a resonant, belong to the set of {m, n, l, r, għ} phonemes in Maltese.From this determining phonological similarity one would assume that these two verbs should behave in the same way in their respective paradigms, since this is the behaviour of l and r across the grammar, however they don't.Rather, as a result of something other than phonology, while the stem-vowel of the IMPERF SG stem-form in talab involves an ablaut change to o (tlob-) from the stem-vowel a of the PERF 1^2 stem-form (tlab-), ħarab does not involve this change, thus resulting in the stem pattern variation.Since ħataf-ħarab do not involve ablaut change across the IMPERF SG and PERF 1^2 slots, retaining the stem-vowel a throughout, then, to account for the distinct number of stem-forms, there has to be stem-form overlapping across the lexemes' sub-paradigms, which further conflates features on the stem-form, resulting in stem-form losses.The loss of two stem-forms in the case of the ħarab set (vs. the one lost in the ħataf set) comes about as a result of having a resonant C 2 coupled by a non-ablaut triggered stem-form change.
a-e e-e These two ablaut-classes pattern exactly with the laqat-talab stem-pattern in the a-a ablaut class.

e-a
In the case of the e-a ablaut class, we have similar distinctions that match exactly with two of the three patterns we had in ablaut class a-a.The beżaq-feraq class patterns exactly like the laqat-talab class, with five distinct stem-forms filling in the five indexed stem-slots, whereas, interestingly, while the ħarab class differs from that of ħataf in ablaut class a-a, mesaħ-feraħ form one class, and pattern with the ħataf subclass having four stem-forms in their paradigms, with the forms msaħ-and fraħrespectively being used across the PERF-IMPERF sub-paradigms in the PERF 3SGF^3PL and IMPERF PL stem-slots.Considering that the stem's syllabic structure in the IMPERF PL cell of a resonant C 2 verb with a strong C [-w/j] VCVC verbal base such as feraħ 'be happy' should be the same as that of the PERF 3SGF^3PL, i.e.CVCC, the fact that it still patterns in the same way as its non-resonant C 2 counterpart mesaħ 'wipe' implies that since feraħ will not allow a *CCC syllable-structured stem-form in the IMPERF PL cell, unlike mesaħ, than ablaut change is involved, which changes the stem-form from ferħ to firħ.If this were not the case, we would have expected the same stem-form overlap observed in the ħarab sub-class, rendering a syncretic stem-form in the PERF 3SGF^3PL and IMPERF PL slots.
i-e i-e In the i-e ablaut class, just like the ħataf-ħarab pattern in ablaut class a-a, there is a varied stem pattern which cross-classifies across resonant and non-resonant C 2 stems.While the latter class of verbs patterns with what seems to be the most common pattern in this SBIC, i.e. having five distinct stem-forms, the resonant C 2 class renders an interesting variation from the other patterns observed so far.What differs is that, unlike all other verbs with this characteristic resonant C 2 property in their stems, the IMPERF SG stem-form tlef-does not overlap with that in the PERF 1^2 cell tlif-, but rather includes a new distinct form.This class of verbs thus introduces a distinct sub-pattern that comes about as a result of having a distinct IMPERF SG stem-form and a syncretic one in the IMPERF PL slot.9 o-o This class of verbs follows exactly with all those verbs that pattern with ħarab in ablaut class a-a.
From the distinct patterns observed across one verbal base within SBIC I, it would be interesting to be able to come up with a parameter that could calibrate which of the stem patterns within a given SBIC appear to be the most canonical.To do so, in the meantime one would need to determine whether ablaut change, as opposed to syncretism, is preferred across the patterns.It seems intuitive to assume that if a SBIC stipulates a particular number of stem-slots, then the most canonical stem-pattern would be that which fills in these stem-slots with distinct forms.Hence, since from the paradigms above we can see that ablaut change renders distinct stemforms, distinct stem-forms in every slot would be preferred over syncretic ones, as is the case in analogy with the different word-forms which fill a lexeme's paradigmatic slots.It may seem that syncretism vs. ablaut change is correlated with ablaut class membership, where overlapping stem-forms have only been found in the a-a and o-o ablaut-classes.However, from the a-a ablaut class we see that this employs both ablaut changing and syncretic strategies, and thus, the correlation of ablaut class with a preference for a kind of stem-pattern over the other cannot quite be taken as an indicator of what one is likely to find.Before concluding this section, one should mention that while it was the verbal base C [-w/j] VCVC discussed here, weak-initial verbal bases, whose initial C is a w (assimilative), including wasal 'arrive', wiret 'inherit', and wiżen 'weigh', model exactly, and fit within the largest sub-pattern in this SBIC, having five distinct stem-forms.The purpose for this note here is to illustrate that it is surface form morphological patterns that account for SBIC memberships, and not phonological or UR accounts, as will also be displayed in the following two sections.

Non-CVCVC SBiC members
The paradigms discussed here should illustrate how SBIC membership is not determined by a type of morphophonological stipulation having to do with the verb's morphophonological base.It is true however that one may say that, interestingly enough, the biggest majority of verbs in this class happens to have a CVCVC verbal base.Mixed memberships however, are what further make it even more of a morphologically-motivated system based solely on surface forms, without showing any relevance to the UR and with no reasonably explicable border with phonology, since SBIC membership need not necessarily cross-classify morphophonologically distinct verbal bases, as mentioned in section (3.1).While mar 'go' is the sole hollow verb that participates in this SBIC (see section 3.2.4 for the discussion on the class of traditional hollow verbs), this non-CVCVC verbal base membership is shared with the a-a ablaut class of verbs belonging to the lacking verbal base (CVCV), illustrated below by qara 'read'.While mar 'go' patterns with the majority stem pattern, interestingly, the lacking verb sub-set, which only includes C 2 resonant stems, patterns exactly like the tilef class of verbs in ablaut class i-e, with a syncretic PERF 3SGF^3PL-IMPERF PL stem-form, and a distinct IMPERF SG form.

irregulars
The sub-heading here may be confusing and will be subject to falsification.While the paradigm for ħa 'take' below, in its surface pattern it is nothing but regular, and is a member of the largest sub-pattern in this SBIC, traditional Maltese pedagogical grammars treat this verb along with the other handful of irregulars mentioned in section (3.1).Its irregularity-based account stems from resorting to an UR, which however poses a stark contrast with what is observed from the surface paradigmatic pattern.It may only be suggestive of the fact that the verbal base inventory based on the 3SGM's stem-form should be expanded in SM, where we have to add a verbal base which includes only one C instead of the typical two or three.This is all that should be done in the overall grammar, instead of trying to impose that which is underlying onto the surface, precisely when affixation and stem-alternation patterns suggest an otherwise non-irregular verb.irregularity, on this account, should only be taken to be something which comes out from the surface, as would be the case in an instance where there is a SBIC or stem pattern whose membership is highly restricted.

Summary
Our overview of SBIC I should have illustrated the subject-matter in further detail, focussing on the interesting array of variation within the same class, rendering four different sub-patterns.It was shown how a morphological pattern will not necessarily cross-classify the verbal bases of the morphophonological properties of the stem-form, or will not necessarily rely on morphotactic constraints, but is rather an autonomous morphological drive that classifies members, based on the stemform organisation realising the syncretic stem index pattern.

SBiC ii
This SBIC shows a distinct stem pattern which involves a non-alternating IMPERF sub-paradigm, and a pattern reorganisation in the PERF sub-paradigm, where the 3SGM stem-slot fuses with that of the 3SGF, while the 3PL takes its distinct stem-slot, as displayed in the ABBC vs. the ABCC pattern in SBIC I, illustrating how the PERF 3PL need not always be conflated with the 3SGF, justifying the reasoning behind the three-slotted 3 rd PERS representation in the beginning of section (3.2.1.1).SBIC II is populated with the sub-class of strong verbs traditionally referred to as deaf/geminate, whose UR involves two identical Cs at the right-edge of the stem (or of the C-root), such as ħass 'feel'.Without needing to refer to any underlying morphological representations when referring to the members of this class, it suffices to say that the stem's C 2 and C 3 are the same.Since this class is exclusive to such verbs, it seems that we may have a case of a morphophonologically-conditioned SBIC membership, although see below.As displayed from the above paradigm, the members of this class are unstable.Unstability, in our context here, should be understood through what can be illustrated from the above alternating pattern, which represents the whole class (except for verbs that patten with mess 'touch' and għadd 'enumerate', where these verbs participate in the ABBC SG=PL SBIC only, implying that their paradigms do not allow for an alternation in the PERF 3PL cell, and hence no shift in SBIC membership).This variation comes about as a result of the interaction with a further non-canonical morphologically-complex phenomenon; stem-form overabundance in the PERF 3PL cell, (illustrated in bold).This stem-overabundance, i.e. having more than one stemform in a given cell, patterning the word-form overabundance which one gets when more than one word-form fills in a paradigmatic cell (Thornton 2010), creates a case of allomorphy for the inflectional material.While allomorphic alternation seems to be phonologically-conditioned, yet, interestingly, this has a further effect on the morphological features discussed here.Such allomorphy results in the possibility of having an additional complexity rendered by the availability of SBIC shifts.Thus, through the non-canonical occurrence of stem-overabundance we get what seems to be a non-canonical SBIC which allows its members to participate in two distinct SBICs.When the SBIC alternation takes place, the 3 rd PERS stem-slots in the PERF sub-paradigm are levelled, resulting in a neat stem-split (Baerman/Corbett forthcoming) which renders a stem-alternation based on a discourse-participant vs. other PERS split (3 rd vs. 1^2), represented by the ABBB PERF sub-paradigm pattern.This process results in a reduction of a stem-slot and a stem-form, i.e. from four to three.
What should be said is that while a stem-form overabundance has here resulted in an SBIC-shift, this is not always necessarily the case, as we will see in section (3.2.4).At the same time, it is not only a non-canonical phenomenon such as overabundance that can cause this non-canonical SBIC shift/alternation in SM paradigms.Rather, this can also result from interactions external to the stem, such as when paradigms involve additional morphological elements, including the negation realising exponent, and attached pronouns.The other interesting thing which can be observed has to do with the relevance of the morphological island property which seems to motivate class formation.This is because when the deaf/geminate verbal base shifts its class membership, it starts patterning exactly in the same way as hol- low verbs with a CV:C verbal base, as will be shown in section (3.2.4), which is taken to imply that SBIC membership is not triggered by any morphophonological stipulation from the verbal base.Although it appears that in many instances there is a crossclassification of SBIC membership and verbal base type, when this is not the case, as also illustrated in sections (3.2.1.2-3)one is able to appreciate and observe the independent status of island morphology at play.

SBiC iii
SBIC III may be thought of as a heteroclite class that takes the PERF stem pattern of SBIC II, with the indexed 1^2, 3SG, and 3PL stem-slots, and the more general alternating IMPERF SG≠PL pattern also found in SBIC I.This class thus displays the same number of stem-slots as SBIC I, however, since they are not the same class, then this means that their syncretic stem indexing pattern organises the slots differently, reflecting the distinct ways how features can be conflated together on the stem in the PERF sub-paradigm.This class illustrates what is here taken to be a real case of irregularity.While there is nothing irregular in the pattern as such, as the island morphological system is able to create whatever pattern the mechanism allows, what makes this pattern irregular in our understanding is that only one lexeme has this pattern in its paradigm.As a result of this single membership the class is regarded as non-canonical, by the criterion mentioned in section (2.2.1).What is further the case here with ġie 'come' is that traditional Maltese grammars treat it along with ħa 'take', mentioned in section (3.2.1).Yet, these two verbs do not illustrate similar paradigmatic patterns.While the surface pattern of ħa is regular, belonging to the biggest SBIC, and can only be regarded as irregular if one is concerned with the verb's UR, ġie, which is also irregular in its UR, is here taken to be irregular only because it participates alone in its own separate class.

SBiC iV
The pattern in this class has been mentioned in section (3.2.2) where members of SBIC II alternate with SBIC IV.This class involves a neat PERF-IMPERF distinction where unlike SBIC I there is no case of stem-form overlap across the two subparadigms, as shown in the representations from żar 'visit' and far 'below', with only one exception; qal 'say', to which we will return below.Furthermore, there is a neat 3 vs.1^2 PERS split in the PERF sub-paradigm.Stem-Class IV is characterised by the traditional class of hollow verbs with a CV:C morphophonological verbal base, however, as shown from section (3.2.2) this base is not exclusive, and CVC i C i bases can also participate in this class.What is interesting to observe that while the class of verbs belonging to this SBIC is traditionally divided according to whether the verb has an underlying -w-or -j-medial consonant, from the surface form stem pattern representations this does not hold as a relevant distinction.
The żar-far class of verbs, representing the whole of the SBIC members.Qal happens to represent its own separate way with which it realises the class' pattern, which is not observed anywhere else across SM verbal paradigms.This pattern is essentially derived from the interaction of the non-canonical phenomena of suppletion and syncretism (rendering higher order exceptionality, Corbett 2011).Therefore, qal is not only irregular because of its particular stem pattern, but also because of the suppletion involved in the PERF sub-paradigm, which, coupled by the subsequent overlap of the suppleted form across the PERF-IMPERF paradigm, results in the only stem pattern that involves a PERF 1^2 stem-form used as the non-alternating stem of the IMPERF sub-paradigm, resulting in a stem-form loss.Unlike ġie 'come', which has been treated as irregular, in the previous section, through evidence from its unique paradigmatic organisation of syncretic stem-slots, which happened to overlap with an irregular UR, qal is treated as regular from a UR perspective, at a par with other -w-medial verbs like far 'overflow' in traditional grammars.In our study here we see that qal displays an irregular means with which it realises its stem index pattern.For this reason, classifying it with the rest of the hollow class in which it belongs, misses out on the interesting variation and the contribution this paradigm gives to the analysis of the realisation of SBICs across 1 st binyan SM verbal paradigms.The same follows for mar 'go', which does not belong in this SBIC altogether, as mentioned in section (3. at the alternating paradigms below represented here by sam 'fast', we observe that we get a case of overabundance in the PERF 3PL cell, and not in the PERF 1^2 cell, as is the case with far 'overflow', għam 'swim', mentioned in (ftn.9), and other members of this class.The alternation here parallels that which we had in SBIC II, illustrated by ħass 'feel', even though the PERF 1^2 alternation is also possible here, which as also mentioned in (ftn.10) above, will not lead to any distinct results.While there it was a case of a SBIC II to SBIC IV alternation, here we have the reversed direction i.e. an alternation from SBIC IV to SBIC II, which results in an additional stem-form.What one wants to say here is that the overabundant alternation comes out from a redundant morphological selection of a distinct allomorph realising the PL features in this cell.As a result of this redundant change in form, the phonology comes to ensure that the stem-form's form fits the phonological requirements.
3.2.5 SBiC V SBIC V includes two members, each with its own separate stem pattern, and just like SBIC III appears to be heteroclite, involving a PERF sub-paradigm that patterns the stem-pattern organisation in SBIC IV having a 3 vs.1^2 stem-alternation, and an alternating IMPERF sub-paradigm that patterns with SBIC I and III.What is interesting to observe is that the irregular heteroclite SBICs III and V overlap with the majority stem-slot organisatory pattern in SBIC I, in their IMPERF pattern.Just like ġie 'come' in stem-class III, kiel 'eat' and ra 'see' are also treated as morphologically irregular in traditional grammar, on the basis of the odd and obsolete C-root in their UR.Although this SBIC is made up of two members, and may be regarded as less irregular than SBIC III, which only has one member, yet, the two members have their own separate stem patterns, which individuates them once again.While kiel 'eat' matches its number of stem-slots with the same number of stem-forms, ra 'see' involves an overlapping stemform, thus having three stem-forms filling in the four stem-slots.

SBiC Vi
SBIC VI is characterised by having the highest number of stem-slots in its members' paradigms, whose syncretic stem organisation pattern renders six distinct stemslots in the overall paradigm.While the IMPERF's pattern is not innovative and alternates on the basis of an SG vs. PL distinction, in the PERF sub-paradigm we have four stem-slots: 1^2, 3SGM, 3SGF, 3PL, which pattern is thus the main motivator behind having the four distinct PERF stem-slots represented in our data.As we will see, this class is divided in two stem patterns; one that has six distinct stem-forms, each one mapping to a designated stem-slot, and the other having five forms, involving an overlapping stem-form across the sub-paradigms.
The members of this SBIC are all members of one of the two ablaut classes found in the traditional class of lacking verbs; e-a.The other class of verbs belonging to the a-a ablaut class participate in SBIC I, as mentioned in section (3.2.1.2).Interestingly then, the class of CVCV verbal bases is divided into separate SBICs on the basis of the ablaut class they belong to, which is not what happens across the different ablaut classes of the traditional strong class, which, as we have seen in section (3.2.1.1),all participate in the same class.It is important to mention that traditional Maltese grammars only distinguish different members of the lacking class on the basis of whether the final weak consonant (which never surfaces in verbal paradigms) is -w or -j.The a-a/e-a ablaut class distinction is not given any independent attention as something that characterises the lacking class, in contrast with the great deal of attention given to the fact that there exist six ablaut classes in the strong class.The importance given to the lacking class' ablaut class division only surfaces to illustrate the significance these distinct ablaut classes had within the diachronic process involved in the integration of the Romance part of the language within the set Semitic verbal paradigmatic model, and are thus only attended to in relation with diachrony (Mifsud, 1995, Hoberman/Aronoff, 2003).What is being proposed here is that, on the contrary, these two ablaut classes are significant within the synchronic context.Furthermore I go a step ahead to say that, synchronically, due to what we observe on the basis of paradigmatic data, the distinction across the a-a vs. e-a ablaut classes is more relevant than that found in the strong class, as the distinction in the lacking class has been grammaticalised and rendered as a split membership in two SBICs.
The above paradigms illustrate how ablaut class e-a is further divided into two distinct sub-classes that essentially depend on the nature of the stem-vowel present in the IMPERF sub-paradigm.While in the ħeba-kera class there is an ablaut change from PERF 1^2 e (ħbej) to IMPERF SG i (ħbi), in the beda-mela class the PERF 1^2 e changes to IMPERF SG a (bdej to bda), which is the pattern that involves a syncretic form across the PERF 3PL and IMPERF PL.It is important to emphasise on the fact that this ablaut change does by no means cross-classify any UR, rather, it happens that all the verbs in the above four paradigms have a -j final radical, and where the ħeba-kera sub-class includes both -j and -w final verbs.Precisely because there is no logically possible explanation as to why the beda-mela class does not function in the same way as the ħeba-kera class or vice-versa, all this is taken to be significant as it implies that SBICs and stem-patterns do really involve internal drives, and are not affected by the external syntax or phonology.

CONCLUSION
From the above 1 st binyan verbal data from SM, we can now define the stem-based inflcetional class as a morphological phenomenon that arises as a result of the nonadherence of internal-paradigmatic canonicity, which follows from the treatment of the stem as pure lexical material that requires to be inert and non-alternating.SBICs are thus formed as a result of different indexing patterns of syncretic stem-slots, whose pattern of stem-form realisation within the same organisatory pattern may vary, and has to do with the number of stem-forms filling the available slots.As a result of alternating stems in paradigms, the canonical distinction/split between affixes as inflectional material vs. stems as lexical material, is blurred, which in SM verbal paradigms results in the conflation of morphological features on the stem, and/or feature overlap The paper discusses the notion of morphological complexity, with a focus on stem patterns.Stem patterns, creating stem-based inflectional classes, are morphological constructs which come about as a result of observing the patterns rendered by the stem-form alternations (or stem splits (Baerman/Corbett forthcoming)), which one extracts after the formation of word-forms within paradigms.Stem-based inflectional class formation constitutes one aspect in the analysis of non-canonical paradigms, which also include affix-based inflectional classes, syncretism, defectiveness, and overabundance Corbett 2005Corbett , 2007Corbett , 2009;;Baerman, Brown and Corbett 2005;Thornton 2010).While these non-canonical instances are in themselves interesting to observe, it is even more intriguing to be able to see what interactions can arise, which at times do not seem to be the result of something exterior to morphology proper.Through data taken from Maltese verbal paradigms the phenomenon of stem-based inflectional classes will be explored, which will exhibit how internal to the paradigm there exists a complex system in itself, which is based on the distinct organisation of different conflated morphosyntactic features which come about via syncretism.These patterns should illustrate a paradigm-internal morphological phenomenon that is irrelevant to the syntax, where while morphology borders with it, there need not be any interaction at this interface.At the same time, it will also be shown how at times, the border with phonology is blurred, where while the phonology may often try to build bridges that interface with the morphological island, the island's internal forces that drive its autonomy may deem to be more superior than the phonology's strive to impose its interacting requirements, which render some interesting morphophonological mismatches as a result.
the affixal material, both resulting in multiple exponence(Baerman/Corbett  2010).Having established what such classes look like in SM, on the basis of the data in section (3.2), we have seen that there are six SBICs that classify the 1 st binyan verbs.While SM appears to allow up to six distinct stem-slots across the PERF-IMPERF paradigm, it is only SBIC VI that fulfills this maximal distinction, and is only restricted to a small sub-set of lexemes, which seems to have resulted from an actual grammaticalisation of something that was merely a phonological ablaut class difference.It was further shown how with the classification of SM verbs in such classes, surface form morphology may be in a mismatched relation with the traditional UR of these same verbs.It is not just the UR that does not necessarily predict SBIC membership and organisation, but also not even surface structure morphophonological similarities.This was shown to be the case especially when the same verbal base participates in more than one SBIC or when distinct verbal bases participate in the same SBIC.With these occurrences, we can clearly bring out the internal autonomous drive upheld by the morphological component, especially when its forces override the potential borderbridging and interaction with the phonological interface, hence reinstating further morphology's island-status.comments by Stephen Anderson, "An I-language view of morphological 'exceptionality': comments onCorbett's paper",[127][128][129][130][131][132][133][134] andreply "Exceptions and what they tell us: reflections on Anderson's comments", 135-137].CORBETT, Greville G./BAERMAN, Matthew (2006) "Prolegomena to a typology of morphological features" Morphology 16, 231-246.CORBETT, Greville G./BAERMAN, Matthew (2010) "Deep Irregularity": When the expected lexical splits follow the wrong pattern.Paper presented at the 14th International Morphology Meeting, Budapest, Hungary.HOBERMAN, Robert D./ARONOFF, Mark (2003) "The verbal morphology of Maltese: From Semitic to Romance."In: J. Shimron (ed) Language Processing and Acquisition in Languages of Semitic, Root-based, Morphology.Amsterdam: Benjamins, 61-78.MIFSUD, Manwel (1995) Loan verbs in Maltese.Brill: Leiden.RYDING, Karen C. (2005) A Reference Grammar of Modern Standard Arabic.Cambridge: CUP.THORNTON, Anna M. (2010) Overabundance, shape conditions and paradigms.University of Surrey. with