This chapter attempts to demonstrate that the toolkit developed in earlier chapters can account for a “welter of descriptive complexity” (Chomsky 2000: 122) by simply combining elements that are independently justified. If correct, such results attest to a certain elegance of the model proposed. Section 1 of this chapter builds on the discussion in Chapter 2, where I analyzed different kinds of affixation in Precedence-Based Phonology. In PBP, affixation in general is the result of morpheme concatenation and linearization. I now will apply these ideas to the
abbreviated reduplication of Guarijio, analyzing it as a combination of two precedence
relations that descriptively would be called reduplication and subtractive morphology. In the second section of this chapter, I examine the interaction between aL-affixation and vowel nasalization in Sundanese, cumulatively building on the discussion in chapters two and three.
Like the Guarijio abbreviated reduplication, the affixation portion once again demonstrates how general enough PBP is to accommodate to different kinds of affixation. The nasalization portion aims to demonstrate how recipient-initiated Search works together with Unify for local and long-distance phonological processes, as seen in the previous chapter. That is to say, these two processes, as well as their interaction, in Sundanese are to be realized as the joint result of PBP, Search, and Unify. If my analyses are convincing, this suggests that PBP, Search, and Unify are good candidates for the universal morphophonological toolkit.
4.1 Guarijio
The Uto-Aztecan language Guarijio is recorded as having a construction called abbreviated reduplication. Targeting verb roots which are “iterative, construed as instantaneous” (Caballero 2006: 279), the abbreviated reduplication adds an inceptive aspect to these roots. Interestingly, in addition to reduplication, both copies of the root appear to undergo subtraction—there is no full copy of the root in the output form. So, for the three-syllable root muhíba ‘to throw’ in (1g), its form for the reduplication in question is simply two exact copies of its first CV: mu-mú ‘to start throwing’. The same observation can be seen for the rest of the items in (1).1
(1) abbreviated reduplication in Guarijio (Caballero 2006: 278, 279, Haugen 2014: 9)2 a. toní ‘to boil’ to-tó ‘to start boiling’
h. pe’ná4 ‘to gather/collect’ pe’-pé ‘to start gathering/collecting’
i. či’í ‘to suck’ či’-čí ‘to start sucking’
j. to’ná ‘to knock’ to’-tó ‘to start knocking’
k. ko’á ‘to eat’ ko’-kó ‘to start eating’
l. yo’á5 ‘to throw up/vomit’ yo’-yó ‘to start throwing up/vomiting’
1 Two exceptions to the abbreviated reduplication are: čeha ‘to sting’, čeh-če ‘to start stinging’; teha ‘to kick’, teh-te ‘to start kicking’ (Haugen 2014: 9).
2 Haugen’s (2014) c and ʔ in the examples are replaced with Caballero’s (2006) č and ’, respectively. The glottal stop (’) in (1h-p) is considered to be an underlying property and is realized between the end of first syllable and the beginning of the second syllable (Caballero 2006: 279, Haugen 2014: 8). This point will not be dwelled upon further. When forms with underlying glottal stop are cited, their URs are followed by {+constricted glottis}. The acute accent (´) is for stress. The stress is morpheme- and construction-specific and can fall on the first three syllables (Caballero 2006: 278 note 8, Haugen 2014: 9 note 3).
3 su-su in Haugen (2014: 9).
4
m. čo’ná ‘to grind’ čo’-čó ‘to start grinding’
n. ki’čú ‘to bite’ ki’-kí ‘to start biting’
o. wo’ná ‘to bark’ wo’-wó ‘to start barking’
p. a’či [sic.] ‘to laugh’ a’-á ‘to start laughing’
Descriptively speaking, the abbreviated reduplication is decomposed of reduplication and subtraction. For the reduplication part, the added precedence relation specifies that “the first V precedes the first X”. Two-syllable long roots may suggest the other possibility that the “last”
V precedes the first X, but items like (1g) whose number of syllables is three (and more) make no room for this ambiguity. Example (1p) has the identified two points converged on the X-slot which is associated with /a/, so a self-loop is added on this timing X-slot (as later shown in (4)). For the subtraction part, Gagnon (2008) and Raimy (2000a: 27ff) provide the clue. The deletion of b in (2) can be phonological or morphological.
(2) two types of deletion
a. phonological deletion b. morphological deletion
#→X→X→X→%
a b c
#→X→X→X→%
a b c where the output of both is a string ac
The familiar phonological deletion (2a) is carried out by a phonological rule. The morphological deletion (2b) is attributed to the added precedence relation, “a (directly) precedes c”. As linearization takes the newly added precedence relationship first, b is not realized after linearization. I consider that the subtraction of the abbreviated reduplication is
due to the “jump link” in (2b). Specifically, I analyze that the added precedence relation reads,
“the first vowel precedes %”. In sum, the abbreviated reduplicative morpheme adds to a verb root two precedence relations in morphology, as formalized in (3).
(3) Guarijio abbreviated reduplication in PBP
Σ (string in the active workspace): punctual and iterative verb root ς (initiator of search): ϛi → ϛj … ςk → %
γ (target of search): γi = first V; γj = first X; γk = first V δ (direction of search): δi = R; δj = R; δk = R
β (beginning point of search): βi = #; βj = #; βk = #
Since both precedence relations are newly added concurrently, when linearization travels to the first vowel, its effect to cover as many precedence relations as possible makes the route which goes back to the first timing slot be taken before the other route which leads to %. In (4), derivations for (1f,p) are given.
(4) suhku ‘to scratch body’, su-sú ‘to start scratching the body’; a’či ‘to laugh’, a’-á ‘to start laughing’
a. underlying representation
#→s→u→h→k→u→% #→a→č→i→% {+constricted glottis}
b. abbreviated reduplicative morpheme concatenation
% %
#→s→u→h→k→u→% #→a→č→i→% {+constricted glottis}
c. linearization
#→s→u→s→u→% #→a→a→% {+constricted glottis}
The particular contents which a morpheme brings along with are idiosyncratic, but, other than that, all the PBP mechanisms are the same in chapter two, in this section for Guarijio, and
is only apparent on the surface. Once analyzed in PBP, it is simply another instance of affixation.
Since the core of PBP is to state precedence explicitly, the above analysis of (3) is yet another piece of evidence to consolidate this core element of morphophonological representations—
precedence.
4.2 Sundanese
We have seen Sundanese infixation in section 2.1. Building on that discussion, I now look further into the details of this plural affix al/ar and discuss how this affixation interacts with vowel nasalization. The formalization of plural affixation, the formalization of vowel nasalization, and the interaction of the two reinforce the importance of the fundamental mechanisms studied in this thesis—computations involving precedence, Search, and Unify.
The plural morpheme has two allomorphs al and ar, and I assume that this morpheme contains a segment L underspecified for {lateral}. When the aL is concatenated with a base, the surface {lateral} value of L is computed in the phonology by three rules. The relevant data is shown as in (1).
(1) aL affixation in Sundanese (Cohn 1992: 201, 206-207, Robins 1959: 343-344) a. lɤmpaŋ l-al-ɤmpaŋ ‘to turn’ e. rɨwat r-ar-ɨwat ‘startled’
b. gɨlis g-ar-ɨlis ‘beautiful’ f. hormat h-al-ormat ‘to honor’
c. ŋoplok ŋ-ar-oplok ‘flop down’ g. sare s-ar-are ‘to sleep’
d. moal m-ar-oal ‘not to want’ h. combrek c-al-ombrek ‘cold’
i. bɤŋhar b-al-ɤŋhar ‘to be rich’
j. ŋumbara ŋ-al-umbara ‘to go abroad’
k. anjɨn ar-anjɨn ‘you’
l. ayɨm ar-ayɨm ‘patient’
When a base begins with a consonant (1a-j), aL is infixed between this consonant and the first vowel counting from the beginning of the word. When a base begins with a vowel (1k-l), aL comes before this vowel and is descriptively a prefix. In other words, aL always precedes the
after that timing slot. To account for this variation between infixation and prefixation, I propose that the plural morpheme has the form in (2).
(2) formalization of aL affixation
a. Σ (string in the active workspace): verb/noun base b. ς (initiator of Search): # → ςj → a → L → ςi c. γ (target of Search): γi = first V; γj = first X d. δ (direction of Search): δi = R; δj = L
e. β (beginning point of Search): βi = #; βj = γi
The morpheme in question as shown in (2b) initiates two rounds of Search but brings with it the beginning symbol (#). In order to utilize the alternative γn-1 for the beginning point of Search, I make what comes later in terms of precedence, closer to the end of the word, (the i Search) take place before what comes earlier in terms of precedence, closer to the beginning of the word, (the j Search). The i Search says to look rightward from the beginning of a string for the first vowel. From where the i Search found its target, the j Search looks leftward for the first timing slot. Although both Searches occur, Search can return with the target or not. In the present case, the i Search always returns with the target, but the j Search may return empty-handed when the first vowel happens not to be preceded by another X-slot, that is, examples like (1k-l). Therefore, instances like (1a-j) whose first vowel is preceded by another X-slot would end up with the configuration (3a) after concatenation and lead to infixation after linearization. As for instance like (1k-l) whose first vowel is not preceded by another X-slot, the configuration of (3b) is resulted in after concatenation and gives prefixation after
linearization.
(3) the first vowel being preceded by another timing slot (a) or not (b) a. # → C → V …
linearization>>> # → C → a → L → V … a → L
b. # → V …
linearization>>> # → a → L → V … a → L
In (3a), because the j Search returns with the target, three precedence relations are added. In (3b), because the j Search returns with nothing, two precedence relations are added. Importantly, no matter the first vowel is preceded by another timing slot or not, the j Search proceeds anyway.
We have encountered a similar situation when the Siraya plural reduplication was discussed in 2.2.1. There the second main Search was accompanied a subsidiary Search. The subsidiary Search looked for the first X preceding the penultimate V. Finding the target or not, this Search applied always. The same mechanical application of a Search also happens here. The j Search just goes on. The surface difference between infixation and prefixation is because the Search in question finds the target or not, never due to the occurrence of the Search. In general, Search performs what its subparts instruct mechanically with absolutely no clue about what the result will be. Search knows what it looks for, but it cannot control the consequences—is the target found or not? Where is the target found? Search per se proceeds without any anticipation.
As for how aL is realized as either ar or al, Robins (1959: 343) has made that clear:
-al- is used with forms whose initial consonant is l, and with those containing a following r, except as initial consonant of the second syllable. Words of any other structure regularly infix -ar-.1
Examples (1a-d) contain l anywhere in the base before aL affixation. Examples (1e-j) contain
r anywhere in the base before aL affixation. Examples (1k-l) do not contain r or l anywhere in
the base before aL affixation. Example (1a) has l which begins the onset of the syllable to the immediately left of the plural morpheme (l-a.l-ɤm.paŋ.), so aL is realized as al.2 Example (1g) has r which begins the onset of the syllable to the immediately right of the plural morpheme (s-a.r-a.re.), so aL is realized as ar.3 Interestingly, when the base r is syllabified as the coda of the syllable which contains L of aL (1f: h-a.l-or.mat.) or as the second member of the onset of the syllable to the immediately right of the plural morpheme (1h: c-a.l-om.brek.), aL is realized as al. Examples (1i-j) where the base r is further to the right of the plural morpheme, irrespective of syllable constituency, also realize aL as al. Lastly, (1k-l), where there either is no base r, or (1e), where the base r is to the left of the plural morpheme, aL is realized as ar.
So, I propose three rules for the laterality alternation: one for (1a), another for (1g), and the other for the rest.
1 Robins (1959: 344) records the following exception: gəde, g-al-əde ‘to be big’.
2 Currently no data is available for bases whose initial syllable has l as non-initial member of the onset.
3 To make sure what matters is the immediately right onset or the immediately right syllable, we would like to have a base with the following criteria for its first three syllables: the first syllable contains no r or l; the second syllable is onsetless; the third syllable has only r as its onset.
(4) laterality alternation rules i.
a. Search
ς (initiator of Search): [-vocalic, +consonantal, +sonorant, +continuant, -nasal, +coronal, +anterior, -distributed, -labial, -round, -high, -back, -low, +voice, -strident, -ATR, -spread glottis, -constricted glottis] (i.e. l, r, L)
γ (target of Search): first onset
ς (initiator of Search): [-vocalic, +consonantal, +sonorant, +continuant, -nasal, +coronal, +anterior, -distributed, -labial, -round, -high, -back, -low, +voice, -strident, -ATR, -spread glottis, -constricted glottis] (i.e. l, r, L)
γ (target of Search): first onset
ς (initiator of Search): [-vocalic, +consonantal, +sonorant, +continuant, -nasal, +coronal, +anterior, -distributed, -labial, -round, -high, -back, -low, +voice, -strident, -ATR, -spread glottis, -constricted glottis] (i.e. l, r, L)
γ (target of Search): [-vocalic, +consonantal, +sonorant, +continuant, -nasal, +coronal, +anterior, -distributed, -labial, -round, -high, -back, -low, +voice, -strident, -ATR, -spread glottis, -constricted glottis, -lateral] (i.e. r)
δ (direction of Search): R
4 The symbol l is used for ease of parsing. With more sophisticated formalization, perhaps some action has to be taken to check if the target of Search begins with /l/.
Because (4i-ii) take the value of {lateral} from the target but (4iii) does not, the former two rules are ordered before the latter one. If the ordering is reversed, due to the definition of unification, rules (4i-ii) are always bled. As for the ordering between (4i) and (4ii), for the time being there lack bases which begin with lVrV, so the ordering is undetermined. In sum, the ordering among the three rules in (4) is: (4i), (4ii) > (4iii).
I turn now to vowel nasalization. The nasality of a vowel in Sundanese is determined by the first non-glottal consonant to the left of the vowel.6 The relevant data is given below.
(5) vowel nasalization in Sundanese (Anderson 1972: 254, Robins 1957: 91, 94)
a. maro mãro ‘to halve’ i. ɲahokɤn ɲãhõkɤn ‘to inform’
b. maneh mãnẽh ‘you’ j. bɤŋhar bɤŋhãr ‘to be rich’
c. mandi mãndi ‘to bath’ k. marios mãrios ‘to examine’
d. ɲiar ɲĩãr ‘to seek’ l. ŋuliat ŋũliat ‘to stretch (intr.), to warp’
e. ɲaian ɲãĩãn ‘to wet’ m. ɲiruan ɲĩruan ‘bee’
f. niis nĩʔĩs ‘to take a holiday’7
n. molohok mõlohok ‘to stare’
g. miasih mĩʔãsih ‘to love’ o. ŋawidaŋ ŋãwidaŋ ‘to dry skins’
h. kumaha kumãhã ‘how?’ p. ŋayaktikɤn ŋãyaktikɤn ‘to make certain’
Examples like (5a,d,f-h,j-l,n) where there is only one nasal tell us that the direction of Search is leftward. That the target of Search is any non-glottal consonant is most obvious from examples where a glottal fricative follows a nasal and precedes a vowel (…N…h…V), as in (5h-j). Since h is oral, if h were the target, (5h-j) should be *kumãha, *ɲãhokɤn, and *bɤŋhar,
6 The insertion of glottal stop will be left for future study. For the original description of glottal stop’s appearance, please refer to Robins (1953: 134-141).
7 Robins (1957: 95) gives another translation ‘to cool oneself’.
respectively. Although the last vowel in (5n) fits the above configuration (…N…h…V), it is still not nasalized because the last two vowels in (5n) both take their value of {nasal} from /l/.
The vowel which immediately follows the glide in (5o-p) is oral, which means that the target of Search should correctly be called non-vocalic. Following Chomsky and Halle (1991: 307), I exclude the glottal fricative from the target of Search by use of the feature {low}. So, the target of Search is represented by [-vocalic, -low]. With the assumption that all the vowels in Sundanese are {nasal} underspecified, the vowel nasalization is then proposed as in (6).8
(6) formalization of vowel nasalization a. Search
ς (initiator of Search): [+vocalic]
γ (target of Search): [-vocalic, -low]
δ (direction of Search): L β (beginning point of Search): ς b. Unify
ς ∐ {αnasal}γ
Descriptively speaking, the glottal is transparent because it is specified {+low} and thus not in the class of, and the oral non-vocalic segments are opaque because they are in the target class.
As /r/ and /l/ in (5a,k-n) behave as the rule (6) specifies, the peculiarity which occurs when the plural morpheme joins bases containing nasal needs some elaboration.
Since the consonant of the plural morpheme is a legitimate target of Search, we expect that all the vowels which follow this consonant in (7) should be oral. This prediction, however, is
not completely borne out. Although the vowel which immediately follows r/l shows normal application of the nasalization, the vowels which are farther apart display over-application.
(7) interaction between aL infixation and nasalization in Sundanese (Robins 1957: 93, 101) a. miak mĩãk m-ar-iak m-ãr-iãk ‘to stand aside’
b. niir nĩʔĩr n-al-iir n-ãl-iʔĩr ‘to pierce’
c. ɲaur ɲãũr ɲ-al-aur ɲ-ãl-aũr ‘to say’
d. moekɤn mõẽkɤn m-ar-oekɤn m-ãr-oẽkɤn ‘to dry’
e. ɲaian ɲãĩãn ɲ-ar-aian ɲ-ãr-aĩãn ‘to wet’
f. mahal mãhãl m-ar-ahal m-ãr-ahãl ‘to be expensive’
g. ɲaho ɲãhõ ɲ-ar-aho ɲ-ãr-ahõ ‘to know’
h. niis nĩʔĩs n-ar-iis n-ãr-iʔĩs ‘to take a holiday’
There are at least two alternatives to deal with the over-application of vowel nasalization. The first option is to have the vowel nasalization rule apply cyclically (e.g. Cohn 1992: 202), as shown in (8) for ‘to dry’.
(8) cyclic vowel nasalization
#→m→O→E→k→Γ→n→% vowel nasalization>>> #→m→õ→ẽ→k→ɤ→n→% AL affixation, linearization>>> #→m→A→r→õ→ẽ→k→ɤ→n→% vowel nasalization>>>
#→m→ã→r→õ→ẽ→k→ɤ→n→%
In (8), Greek capital letters stand for vowels whose {nasal} is underspecified. When the vowel nasalization first applies, the vowels O and E are nasalized due to m. The vowel Γ, however, becomes ɤ because k is oral. When the same rule applies again later, the vowel A of the plural morpheme gets nasalized because of m. Since Unify only produces consistent feature specification, the vowels õ and ẽ are not denasalized even though r is oral. The vowel ɤ vacuously undergoes the nasalization rule. The second option is to apply the vowel nasalization
rule after AL affixation and before linearization, as shown in (9) for ‘to dry’.
(9) vowel nasalization before linearization
#→m→O→E→k→Γ→n→%
A→L
where A finds m; O finds m and L; E finds m and L; Γ finds k
In the above configuration, the vowels O and E each find the two targets (one nasal and the other oral). (Recall we have mentioned the situation of one initiator to more than one target in the end of section 3.3.) Then, the unification of {αnasal} in (6b) is split into {+nasal} and {-nasal}, as shown in (6'b).
(6') formalization of vowel nasalization a. Search
ς (initiator of Search): [+vocalic]
γ (target of Search): [-vocalic, -low]
δ (direction of Search): L β (beginning point of Search): ς b. Unify
i. ς ∐ {+nasal}γ
ii. ς ∐ {-nasal}γ
The unification of {+nasal} (6'bi) is ordered before the unification of {-nasal} in (6'bii), yet both steps are to be applied. When an initiator finds only one target and that target is oral (10a), the nasal unification portion does not apply because the target is unable to provide {+nasal}, and the oral unification portion applies and makes the initiator oral. The situation is reversed when an initiator finds only one target and that target is nasal (10b). When an initiator finds
more than one target and these targets are mixed in terms of nasality (10c), after the unification of {+nasal}, the unification of {-nasal} never applies because the {+nasal} of the initiator of Search is inconsistent with the {-nasal} of the oral target of Search. Conditions (10a) and (10b) have different results, but they share the reason why one of the two unifications does not apply—the target simply lacks the relevant valued feature. On the contrary, conditions (10b) and (10c) have the same result, but they differ in the reasons why the oral unification is not relevant—the former is because the only target is nasal; the latter is because of the definition of Unify.
(10) nasal unification before oral unification
a. b. c.
Vς to Oγ Vς to Nγ Vς to Oγ and Nγ
ς ∐ {+nasal}γ non-applicable Ṽς Ṽς
ς ∐ {-nasal}γ Vς non-applicable unification failure
With these two options, the oral vowel which immediately follows r/l in (7) is a puzzle as it should be nasal. Following Anderson (1972: 267), I assume that there is a vowel denasalization rule in Sundanese, as shown in (11).
(11) vowel denasalization (adapted from Anderson 1972: 267) a. Search
ς (initiator of Search): [+vocalic, +nasal]
γ (target of Search): [ ] δ (direction of Search): L β (beginning point of Search): ς b. Delete
if γ is [+consonantal, -nasal], ς – {+nasal}
c. Unify ς ∐ {-nasal}
The symbol [ ] stands for any segment Æ. Recall from section 3.1.2 that [αF] means {x: x {αF}}, i.e. any set which is a superset of {αF}. Once we have the set theory interpretation of segments, there is no a priori reason to rule out the possibility to find out any set which is a superset of the empty set, that is, {x: x { }}. By definition, the empty set is a subset of every set. In other words, every set is a superset of the empty set. In turn, the value of {x: x { }} is every set. As each segment, from a bare timing slot to a fully specified segment, is a set of features, the notation {x: x { }} (=[ ]) thus stands for every segment Æ. So, the rule (11) reads as: from a nasal vowel itself search leftward for any segment Æ (11a); if that segment is oral consonant, delete {+nasal} from the initiator of Search (11b); Unify {-nasal} with the initiator of Search (11c). Since Search terminates immediately when the target is found, to have [ ] as the target of Search means that Search would always terminates at the X-slot which either immediately precedes or immediately follows the initiator of Search, depending on the direction of Search. Because all the vowels in Sundanese are assumed to be underlyingly not
nasalization rule (6). If the ordering is the other way around, (11) never could have an influence.
The above analysis of a fraction of Sundanese morphology and phonology repeatedly demonstrates that precedence, Search, and Unify are necessary. The phenomena may seem to be diverse, but no new mechanism is needed when they are analyzed.
4.3 Summary
Two more case studies from Guarijio and Sundanese were presented in the previous pages.
Two more case studies from Guarijio and Sundanese were presented in the previous pages.