Introduction
This paper offers an optimality-theoretic account of the occurrence and non-occurrence of final /n/ deletion in Ghayeni Persian whose application in different domains and environments is affected by several opaque counterbleeding and counterfeeding interactions in addition to bleeding. The theoretical machinery utilized in this research builds directly on Optimality Theory’s (OT) existing approaches involving a novel combination of theoretical tools. It aimed to discuss how Parallel Optimality Theory (POT) and Harmonic Serialism (HS) in particular could treat these opaque interactions in Ghayeni dialect. That is why OT’s other existing approaches such as Stratal OT were not applied in this research. This research is intended to make a contribution by addressing certain conditions in which POT and HS might be able to treat opacity. It discusses although according to McCarthyFootnote 1 HS is not very successful in treating counterbleeding opacity, it might make it in derived words.
Phonological opacity was first defined by KiparskyFootnote 2 “as a measure of how far the context or the consequences of a phonological process may be determined only by examining the surface structure”.Footnote 3 According to KiparskyFootnote 4 “the concept of opacity comes from the SPE Footnote 5 tradition: a rule is opaque if the fact that it applied or the context that determined whether it applied is not visible in the surface form”.Footnote 6 “Although ChomskyFootnote 7 do not frame it in these terms, opacity is the crux of his argument against structuralist phonemics”.Footnote 8
“The phenomenon of phonological opacity has been the subject of much debate in the Generative Phonology especially following the appearance of OT when phonologists were quick to realize that, in its original version, OT was unable to describe a large set of phonological phenomena previously modelled by means of opaque rules”.Footnote 9 Phonological opacity is a challenge for parallel OT mainly because it does not allow for intermediate levels of representation. GreenFootnote 10 states that “opponents of OTFootnote 11 that opacity proves OT to be false and that phonology must be derivational.” According to McCarthyFootnote 12 “there have been many proposals for accommodating opacity in OT, they rely on a third level of representation, neither underlying nor surface, as a crucial part of the analysis of opaque alternations. The defining property of a derivation is the presence of this third (or fourth or fifth) level of representation.” HeFootnote 13 adds that “previous attempts to meld OT with serial derivations or their analogues have not been fully successful. The challenge, then, is to make use of the derivational insight without losing hold of OT's essential properties and basic results. “ UrekFootnote 14 states that “several modifications have been proposed over the years to incorporate opacity in OT. However, any of those approaches is associated with a number of problems, which sometimes prove fatal under closer examination.” She adds that “in ‘classic’ (parallel) OT the Input- Output map has no internal structure and all possible variants are produced by Gen in one step and evaluated in parallel. Precisely this property makes opacity a challenging issue for OT.”According to AnttilaFootnote 15 “most approaches that have been proposed to accommodate opacity assume that the problem lies in an inadequate theory of FAITHFULNESS and posit new kinds of faithfulness relations, e.g. Output-Output correspondence,Footnote 16 Sympathy,Footnote 17 and Turbidity.”Footnote 18 He adds that “other approaches assume that the problem lies in an inadequate theory of MARKEDNESS and enhance the power of markedness constraints, e.g. Targeted ConstraintsFootnote 19 and Comparative Markedness.Footnote 20 Yet Stratal OTFootnote 21 assumes that phonological constraints indeed always interact transparently, contrary to appearances, and that non-interactions arise from outside phonology proper, in particular morphology.” But GreenFootnote 22 notes that “the solutions proposed within OT, such as sympathy theory and stratal OT, have proved to be unsatisfying to many OT proponents, who have found these proposals to be inconsistent with the parallelist approach to phonological processes otherwise characteristic of OT.” Also KagerFootnote 23 believes that “each attempted solution has certain advantages and disadvantages and none of them seems to truly solve the opacity problem.” However, as McCarthyFootnote 24 states that “opacity is deeply connected with the phonology of a language, and any adequate theory of opacity must recognize this.”
As stated above several modifications of the theory have been proposed over the years to incorporate opacity. To this end, this research adopts HS as the derivational implementation of OTFootnote 25 to accommodate counterbleeding opacity. As Torres-TamaritFootnote 26 describes HS, “it is a derivational version of OT in which EVAL imposes the same constraint hierarchy in every step of the derivation. In other words, HS is just a version of OT that combines optimization (i.e., constraint interaction) with derivations.” McCarthyFootnote 27 points out that “after each evaluation, the optimal candidate selected by EVAL is fed back into GEN as a new input, from which a new candidate is constructed. This GEN- EVAL loop continues until there is convergence, when the optimum chosen by EVAL is identical to the most recent input to GEN. At that point, the derivation terminates, and the convergent form is the final output of the grammar.”
Regarding counterfeeding opacity, we adopt POT using local constraint conjunction (LCC) originally proposed by Smolensky.Footnote 28 It is worth noting that discussing opacity in this research requires showing intermediate representations, although it is less common in OT works. A case of bleeding in which final /n/ deletion is bled by a prior phonological process is discussed as well using local constraint conjunction.
This paper is organized as follows; in the rest of this section after providing general features of Ghayeni phonology, we introduce and analyze final-n deletion in this dialect. In §1, we discuss final-n deletion in the plural marker once it is added to nouns ending in a consonant, nouns ending in vowels /ɑ/ or /e/, and nouns ending in [ou]. In §2, we analyze non-occurrence of final /n/ deletion in Ghayeni Persian in coda clusters including the /nd/ cluster and the glottal + /n/ cluster. Finally, in §3, we discuss both application and non-application of final /n/ deletion in the Ezafe construction including its application in the singular pronoun “in” and the plural pronoun “inɑn”, and its non-application in the possessed and modified nouns as well as in nouns with the plural marker [au]. This research is concluded in §4 with a brief summary of some of the article’s key points.
Ghayen is a city located in the northwest of South Khorasan province in Iran. A dialect of Persian (an Indo-European language) called “Ghayeni” is spoken in this city. General features of Ghayeni phonology based on ZomorrodianFootnote 29 are given below:
− Six vowel phonemes: /i/, /e/, /a/, /u/, /o/, /ɑ/ (Vowel length is not contrastive).
− Twenty-three consonant phonemes: /p/, /b/, /m/, /n/, /f/, /v/, /t/, /d/, /s/, /z/, /l/, /r/, /ʃ/, /ʒ/, /tʃ/, /dʒ/, /k/, /ɡ/, /j/, /χ/, /G/, /ʔ/, /h/.
− Three syllable structures: CV, CVC, CVCC (glottal stop is inserted at the beginning of vowel initial words, and hiatus is resolved by deletion of a vowel or insertion of an intervocalic consonant).
− Stress pattern: The main stress is word-final in nouns, adjectives and adverbs. For the verbs, it depends on morphological and syntactic properties. (Since the main stress is always word-final in nouns, adjectives and adverbs and since it has nothing to do with final /n/ deletion, it is not an issue in the examples presented in this paper).
In Ghayeni dialect final /n/ deletion is a productive phonological process. We have to state explicitly that all of the data in this paper come from Zomorrodian,Footnote 30 approved and double- checked by one of the coauthors as a native speaker of Ghayeni Persian. Besides, in this research the underlying representation of Ghayani words is the Ghayani dialect rather than standard Persian although both dialects share the same underlying representation (henceforth UR) in many words especially those including underlying final /n/ which originate in Middle Persian spoken during the Sassanid Empire (224–651 AD) as the ancestor of both Ghayeni and Modern Persian.Footnote 31 Application and non- application of final-n deletion in this dialect is shown in the data in (1):
(1)
As evidenced in (1) final /n/ in the words on the left deletes, while non-final /n/ in their suffixed versions on the right does not. In addition, despite the productivity of this phonological process in Ghayeni dialect, it is being bled by some processes. In fact one reason we discuss cases of non-application of final-n deletion is to show that final /n/ does synchronically exist in the UR.
Now we introduce the constraints responsible for the occurrence of final /n/ deletion in Ghayeni Persian. Then, in the sections that follow we explore more complex applications and non-applications of this process in different environments in this dialect.
Based on the fact that the underlying /n/ deletes at word final position in Ghayeni Persian, we formalize the markedness constraint ‘*n]’ which bans final [n].It is worth mentioning that just like Ghayeni Persian, Catalan and standard Dutch have a sort of final /n/ deletion. But unlike CatalanFootnote 32 and Standard DutchFootnote 33 in which stress and schwa precede the deleted /n/ respectively, final /n/ deletion in Ghayeni Persian is not restricted to stress conditions or the presence of a certain vowel.
According to the ranking in (2), the markedness constraint ‘*n]’ that triggers final /n/ deletion dominates the faithfulness constraint ‘MAX’ which mandates that every segment in the input has a correspondent in the output.
(2) *n] >> MAX
It is worthy of note that *n], just like other universal phonological constraints, is applicable in other languages. However, the ranking is different in standard Persian as well as in many world languages where final /n/ deletion does not apply i.e., it is MAX that dominates *n] (MAX] >> *n).
The ranking in (2) is capable of accounting for both application and non-application of final /n/ deletion in the words in (1):
(3) Final /n/ deletion in Ghayeni Persian
As shown in tableau (3a), candidate (a.i) is optimal as it has not violated the dominant markedness constraint ‘*n]’. However, since in tableau (3b), candidates have no final [n] they have not violated ‘*n]’. Then, as in the second round candidate (b.ii) has unlawfully undergone non-final /n/ deletion, candidate (b.i) with no deleted phoneme is chosen as optimal.
1. The Plural Marker
In standard Persian, words are commonly pluralized with the suffixes “-hɑ“ and “-ɑn”. While “-hɑ“ is used to pluralize all nouns, “-ɑn” is used to form the plural of human nouns and some other living beings as animals and plants:
(4) Standard Persian
Two interesting features of Ghayeni Persian are that it lacks the suffix “-hɑ“, and the plural suffix “-ɑn” does not surface in plural forms. Instead, “-u” and [au] which are regarded as plural markers appear after consonants and vowels respectively.
Nouns Ending in a Consonant
The plural suffix in the following plural forms seems at first sight to be “-u” in Ghayeni Persian:
(5)
However, we argue that the plural suffix is in fact “-ɑn” which has undergone pre-nasal raisingFootnote 34 and final /n/ deletionFootnote 35. The former is a common phonological process in the majority of Persian accents and dialects including the standard accent and Ghayeni through which the low vowel /ɑ/ converts to the high vowel [u] in the environment before nasals, as shown in the following examples:
(6)
The interaction between pre-nasal raising and final /n/ deletion is shown in the following derivation:
(7)
Upon the deletion of /n/, the environment that caused the occurrence of pre-nasal raising has disappeared in the phonetic representation (henceforth PR). This is an instance of counterbleeding opacity as defined in AntilaFootnote 36 because despite the fact that the conditioning environment for pre-nasal raising is not met on the surface, it applies (overapplication). Moreover, if the rules were applied in the opposite order, final /n/ deletion would “bleed” pre-nasal raising by depriving it of the opportunity to apply. This hypothetical derivation is shown in (8):
(8)
In counterbleeding opacity, “an unfaithful mapping occurs for reasons that cannot be explained with classic OT markedness constraints because the conditions that encourage the unfaithful mapping are no longer apparent in surface structure”.Footnote 37 Discussing HS’s difficulty in handling counterbleeding opacity, McCarthyFootnote 38 asserts that “despite having derivations with intermediate stages, HS is not very successful in treating opacity as it runs into problems because of the durability of the constraint hierarchy and the markedness/faithfulness split, basic characteristics that it shares with the parallel implementation of OT.” However, heFootnote 39 accepts that in a limited and rather arbitrary set of cases, HS can handle this type of opacity. He concludes that HS is worth studying, and may very well reward further examination under assumptions different from those he had entertained in his paper. According to McCarthy,Footnote 40 Pruitt,Footnote 41 Elfner,Footnote 42 and KimperFootnote 43 among others, from an empirical point of view, it has been demonstrated that HS is able to accommodate particular cases of counterbleeding opacity better than POT. Therefore, we have adopted HS to accommodate particular cases of counterbleeding opacity in our research.
Jam & TeymouriFootnote 44 propose that the markedness constraint that triggers pre-nasal raising is “*[ɑ]N” which prohibits the sequence of [ɑ] followed by a nasal. However, *[ɑ]N is only accounting for part of the pre-nasal raising phenomenon. It explains why the vowel changes, but not why it changes all the way from /ɑ/ to [u] rather than to [o]. So in order to correct this shortcoming we formalize *[-high]N as defined in (9):
(9) *[-high]N The sequence of a non-high vowel followed by a nasal is prohibited.
The markedness constraint *[-high]N is acoustically motivated. The results of an acoustic salience analysis by Salehi KoopaeiFootnote 45 indicate that pre-nasal raising occurs because the high vowel + nasal sequence is more perceptible than the low vowel + nasal sequence. So, pre-nasal raising is a listener-oriented phonological process.
While pre-nasal raising is obligatory in most Persian accents and dialects, it exceptionally fails to occur in some words due to different linguistic and non-linguistic reasons. For instance, while /tɑbestɑn/ ‘summer’ changes to [tɑbestun], /dabestɑn/ ‘elementary school’ fails to change to *[dabestun]. Furthermore, in some other words its occurrence is optional as an instance of free variation. For example, “Tehran” is optionally pronounced [tehrɑn] and/or [te:run].Footnote 46
IDENT[height]Footnote 47 is the faithfulness constraint that militates against any change in the value for the feature [high] in an output segment compared with its corresponding segment in the input. The markedness constraint *[-high]N must dominate IDENT[height] to insure pre-nasal raising. Furthermore, in order to prevent the transparent candidate *[doχtarɑ] (in the hypothetical derivation in (8)) from winning (incorrectly)Footnote 48 we formalize the morpheme-specific markedness constraint *ɑ]PL which confirms that final [ɑ] is not the plural marker.Footnote 49 It is worthy of note that using morpheme-specific constraints like (Edgemost(um; L))Footnote 50 was common in OT from its emergence.Footnote 51 Moreover, with the incorporation of *ɑ]PL, our analysis survives a pitfall challenging HSFootnote 52 in handling counterbleeding opacity.
Explanation of this counterbleeding interaction between pre-nasal raising and final /n/ deletion requires the incorporation of the abovementioned three constraints in the initial ranking in (2) which yields the following ranking:
(10) *[-high]N >> *ɑ]PL >> *n] >> IDENT[height], MAX, DEP
In this research we have used multi-step tableaux.Footnote 53 The semi-circle arrows to the left side of the tableau indicate which form in a given step is selected as the input form for the next step.Footnote 54
In multi-step tableau (11) which includes steps 1, 2, and 3, the word /doχtar+ ɑn/ is analyzed under the ranking in (10). Since this counterbleeding interaction is between two processes, two steps are required before convergence: /doχtar+ɑn/ → [doχtarun] → [doχtaru]. In step 1, the input to GEN is the underlying form /doχtar+ɑn/, and the candidates include faithful [doχtarɑn] as well as three unfaithful [doχtarɑ], [doχtaron] and [doχtarun]. The four candidates are evaluated and the most harmonic one, [doχtarun], becomes the new input to GEN in step 2. The candidates in step 2 include faithful [doχtarun] and unfaithful [doχtaru]. The grammar chooses [doχtaru], which becomes the new input to GEN in step 3. The progression of harmonic improvement from the UR to the output of step 1 to the output of step 2 is obvious. In each step, the winner is more harmonic than the input. Finally, there is convergence in step 3. Underlying /doχtar+ɑn/ has realized all of its potential for harmonic improvement under this grammar, so the output of EVAL and the input to GEN are identical.
(11) /doχtar+ɑn/ → [doχtaru]
Nouns Ending in /ɑ/ or /e/
One of the features of Ghayeni dialect is that if a singular noun ends in vowel /ɑ/ or /e/, its plural form ends in diphthong [au] which is regarded as a plural marker as evidenced in the data set in (12): (the plural forms in parentheses are standard Persian)
(12)
A careful examination of the change of /ɑ/ or /e/ to [au] reveals that it occurs through the following derivation:
(13)
As shown in the derivation above, this is an instance of counterbleeding opacity because upon the deletion of /n/, the environment that caused the occurrence of pre-nasal raising has disappeared in the PR. Furthermore, pre-nasal raising has created diphthongs “eu” and “ɑu” in the intermediate level. However, since these two diphthongs are absent in Ghayeni Persian, their first vowels change to [a] to form [au] which is regarded as a plural marker in this dialect. Note that contrary to languages like English, Persian has no phonemic diphthongs. Moreover, contrary to the unmarked plural marker /ɑn/, [au] is the marked plural marker because as shown in the derivation in (13) it is the result of the application of three phonological processes (it does not exist in the UR as Persian has no phonemic diphthongs). According to KagerFootnote 55 in comparison with unmarked properties, marked properties occur in sound systems with less frequent frequency.
Based on the fact that [au] is regarded a plural marker, we formalize the markedness constraint “[au]PL”:
(14) [au]PL Plural marker diphthong is [au].
Note that *ɑ]PL and [au]PL are not positive and negative versions of each other. They were formalized for different reasons. The negative constraint *ɑ]PL was formalized to be violated by the transparent candidate. As mentioned earlier, the great weakness of the HSFootnote 56 in dealing with counterbleeding opacity is how to get rid of the transparent candidate. But the positive constraint [au]PL was formalized to prevent the candidate with a wrong diphthong from winning. Also as the underlying plural marker “-ɑn” includes no diphthong, [au]PL is not at odds with it. Even if [au]PL could affect “-ɑn”, since it is not on top of the ranking it cannot rule out the faithful candidate in step 1namely in tableaux (16), (20), and (50) where[au]PL technically becomes decisive in step 3 or 4 (where there is no “-ɑn”) to rule out candidates with [ɑu], [ou] or [eu] which are not plural marker diphthongs.
Explanation of this counterbleeding interaction requires the incorporation of [au]PL in the ranking in (10). Moreover, since in the vowel change the [-back] vowel [a] replaces the [+back] vowel /ɑ/, the faithfulness constraint “IDENT[back]” which militates against this vowel change must be added to the ranking which ultimately yields the following ranking:
(15) *[-high]N >> *ɑ]PL >> *n] >> [au]PL >> IDENT[height], IDENT[back], MAX, DEP
In multi-step tableau (16) which consists of 4 steps, the word /gedɑ+ɑn/ is analyzed under the ranking in (15). Since this counterbleeding interaction is between three phonological processes, three steps are required before convergence: /ɡedɑ+ɑn/ → [ɡedɑun] → [ɡedɑu] → [ɡedau]. In step 1, the input to GEN is the underlying form /ɡedɑ+ɑn/, and the candidates include faithful [ɡedɑɑn] and three unfaithful [ɡedɑɑ], [ɡedɑon] and [ɡedɑun]. The four candidates are evaluated, and the most harmonic one, [ɡedɑun], becomes the new input to GEN in step 2. The candidates in this step include faithful [ɡedɑun] and unfaithful [ɡedɑu]. Step (2) shows that the grammar chooses [ɡedɑu], which becomes the new input to GEN in step 3. The candidates in this step include faithful [ɡedɑu] and two unfaithful [ɡedau] and [ɡedɑun]. The candidates [ɡedɑun] and [ɡedɑu] have violated *n] and [au]PL respectively. Therefore, the grammar chooses [ɡedau] as the plural form, which becomes the new input to GEN in the convergence step in which the output of EVAL and the input to GEN are identical.
(16) /ɡedɑ+ɑn/ → [ɡedau]
Nouns Ending in /ou/
Another feature of Ghayeni dialect is that nouns ending in [ou] are singular while the same nouns ending in [au] are Plural:
(17)
A closer look reveals that [au] replaces /ou/ in the plural form through the following derivation:
(18) UR /ʃou+ɑn/
As shown above, the consecutive occurrences of pre-nasal raising, hiatus resolution and final /n/ deletion change the plural /ʃou+ɑn/ to “ʃou” which is the same as its singular form, as if nothing happened. However, since in Ghayeni Persian [-au] is regarded as a plural marker, it has to replace /ou/ to differentiate the plural from the singular. This is imposed by the markedness constraint [au]PL through changing /o/ in /ou/ to [a].
Persian in general does not allow onsetless syllables and hiatus. Resolving hiatus requires the incorporation of the markedness constraint “ONSET”Footnote 57 in the ranking in (19):
(19) *[-high]N >> *ɑ]PL >> ONSET >> *n] >> [au]PL >> IDENT[height], IDENT[back], MAX, DEP
In multi-step tableau (20) which consists of 4 steps, the word /ʃou+ɑn/ is analyzed under the ranking in (19). This counterbleeding interaction is between four phonological processes. But since hiatus resolution and final /n/ deletion occur unrelatedly at the same level of representation, three steps are required before convergence: /ʃou+ɑn/ → [ʃou.un] → [ʃou] → [ʃau]. In step 1, the input to GEN is the underlying form /ʃou+ɑn/, and the candidates include faithful [ʃou.ɑn] and four unfaithful [ʃou.ɑ], [ʃou.on], [ʃou.un] and [ʃoun]. The five candidates are evaluated and the most harmonic one, [ʃou.un], becomes the new input to GEN in step 2. The candidates set at step 2 includes [ʃoun] and [ʃou]. The grammar chooses [ʃou], which has not violated “*n]”. Then [ʃou] becomes the new input to GEN in step3. The candidates in step 3 include [ʃau], [ʃou] and the [ʃoun]. The grammar chooses [ʃau] as the plural form, which becomes the new input to GEN in step 4. Finally, there is convergence in step 4; underlying /ʃou+ɑn/ has realized all of its potential for harmonic improvement under this grammar, so the output of EVAL and the input to GEN are identical.
(20) /ʃou+ɑn/ → [ʃau]
2. CODA CLUSTERS
As we mentioned in the introduction, this paper aimed to discuss both the occurrence and non-occurrence of final /n/ deletion in Ghayeni Persian. To this end, this section explores two environments where this process fails to occur. Moreover, the data are regarded as additional evidence for the existence of final /n/ in the UR.
/nd/ Cluster
Ghayeni Persian has the postnasal /d/ deletion rule which deletes /d/ when it is preceded by /n/ within a coda cluster:
(21)
After the application of postnasal /d/ deletion, it seems that the environment is ready for final /n/ deletion. But final /n/ is not deleted if it becomes final through /d/ deletion. Although the conditioning environment for final /n/ deletion is met on the surface, it fails to apply (underapplication). This is an instance of counterfeeding opacity, as defined in Antila.Footnote 58 The PRs in (21) fail to undergo final /n/ deletion because this process is incapable of affecting the output created by the postnasal /d/ deletion rule. This implies that final /n/ deletion has to apply before postnasal /d/ deletion. In other words, final /n/ deletion has priority over postnasal /d/ deletion. This priority prevents the occurrence of final /n/ deletion, though its conditioning environment is met on the surface.
(22) Counterfeeding interaction between final /n/ deletion and postnasal /d/ deletion
The markedness constraint that triggers postnasal /d/ deletion is *nd] defined in (23):
(23) *nd]
The word final sequence of [n] followed by [d] is prohibited.
On the first attempt we incorporated *nd] in the initial ranking in (2) in the hope that it would explain this counterfeeding interaction. But as depicted in tableau (24), candidate (b) whose coda cluster is totally deleted is incorrectly selected as optimal over opaque (a). So the desired output [ban] is unattainable with just these three constraints.
(24) Incorrect analysis of counterfeeding between final /n/ deletion and final /d/ deletion
As shown in tableau (24) upon the application of both rules in candidate (b), the whole coda cluster has disappeared and subsequently the word has got damaged. In order to solve this problem we use a local constraint conjunction consisting of two locally conjoined (anti-deletion) MAX constraints ranked above *n] to ensure that the opaque candidate whose /n/ is not deleted is more harmonic than the transparent one which is damaged due to the deletion of two segments. This is a fact that /n/ never deletes when it is followed by another segment in the input, even if that segment is deleted later by a phonological process. The local constraint conjunction proposed in (25) includes MAX-[n] and MAX-[d] which militates against the whole coda cluster deletion. It is worthy of note that in each of the three instances of counterfeeding opacity as well as in one instance of bleeding interaction discussed in this paper we are dealing with two deletion processes within the same syllable domain which could be dealt with using locally conjoined MAX constraints.
(25) Local Constraint Conjunction
MAX[[n] & [d]]coda One violation is assigned for a candidate that violates both MAX-[n] and MAX-[d]. The domain for this constraint is the coda.
The ranking in (26) explains the counterfeeding opacity in the examples in (21):
(26) *nd] >> MAX[[n] & [d]]coda >> *n] >> MAX
Tableau (27) demonstrates that the opaque candidate (a) which has only undergone postnasal /d/ deletion is optimal.
(27) Counterfeeding opacity: no /n/ deletion
Glottal +/n/ Cluster
Ghayeni Persian has a glottal consonant deletion rule which deletes /h/ or /ʔ/ within a coda cluster. The application of glottal deletion creates a context in which the other consonant in the cluster including final /n/ has to remain to preserve the coda. Otherwise the whole coda cluster would disappear and subsequently the word would get harmed. This is an instance of bleeding order; although the conditioning environment for both glottal deletion and final /n/ deletion is met in the UR, the latter fails to apply because the application of glottal deletion creates a context in which final n- deletion can no longer apply. This distinct realization of word-final /n/ is represented by the examples in (28):
(28)
As we see in the examples above, upon the loss of the glottal consonant the vowel /a/ is lengthened. This is an instance of compensatory lengthening (CL) which occurs when the mora of a deleted segment survives and links another segment.Footnote 59 However, as compensatory lengthening has nothing to do with the bleeding of final n- deletion, it is ignored in our discussion.Footnote 60
The markedness constraint that triggers glottal loss in syllable coda is *Coda-glottal stated in (29):
(29) *Coda-glottal Glottal consonant is prohibited in syllable coda.
Here again local constraint conjunction MAX[[glottal] & [n]]coda militates against the whole coda cluster deletion. The bleeding of the final /n/ deletion rule by the glottal consonant deletion rule is due to the dominance of *Coda- glottal and MAX[[glottal] & [n]]coda over *n] represented in the ranking in (30).
(30) *Coda-glottal >> MAX[[glottal] & [n]]coda >> *n] >> MAX
(31) Glottal deletion bleeds final /n/ deletion
3. The Ezafe Construction
In Persian, the possessive form of nouns and also modification of nouns by adjectives are commonly introduced using Ezafe vowel /-e/, an enclitic which is attached to the noun being possessed or modified. In Ghayeni Persian, Ezafe vowel has a key role in creating different opaque interactions regarding final /n/ deletion discussed in the following four sections.
Possessed and Modified Nouns
In Ghayeni Persian, if a singular or a plural noun ending in /n/ is a possessee or is modified by an adjective, final /n/ deletion fails to occur:
(32)
As seen in the examples in (32), although the conditioning environment for final /n/ deletion is met on the surface, it fails to apply (underapplication). This is an instance of counterfeeding opacity which is more challenging than the one discussed in section “/nd/ Cluster” because due to our access to the UR in that section, we knew that the existence of /d/ at the end of those words prevented final /n/ deletion. But in the examples in (32), we have no access to the UR. However, we know there must be a reason why it failed to apply. That is, just like the words in (21) in which the existence of final /d/ prevented final /n/ deletion, we infer that the words in (32) must end in a hidden Ezafe enclitic. A closer look reveals that this morpheme is later deleted at an intermediate level. According to Ghomeshi,Footnote 61 Ezafe which literally means ‘addition’ is the unstressed vowel /-e/ that links together elements belonging to a single constituent. It appears between an adjective and its complement, between a preposition and its complement, and more generally between any two items that have some sort of connection. ZomorrodianFootnote 62 asserts that in Ghayeni Persian Ezafe does not appear after words ending in /n/ (p. 50). Accordingly, the UR of the words in (32) would be the following:
(33)
As seen in the following derivation, the Ezafe deletion rule in [ruGanzard] is incapable of feeding final /n/ deletion because it has no priority over final /n/ deletion:
(34) Counterfeeding interaction between final /n/ deletion and Ezafe deletion
The markedness constraint that triggers Ezafe vowel deletion is *neEz defined in (35):
(35) *neEz The sequence of /n/ followed by Ezafe [e] is prohibited.
The local constraint conjunction provides a straightforward solution to counterfeeding opacity in which the candidate whose Ezafe vowel is only deleted is intended to win. The LCC that we have used is MAX[[n] & [eEz]]σ.
(36) Local Constraint Conjunction
MAX[[n] & [eEz]]σ One violation is assigned for a candidate that violates both MAX-[n] and MAX -[eEz]. The domain for this constraint is the syllable.
The markedness constraint NUCLEUS/V (Tranel, 1996) triggers resyllabification:
(37) NUCLEUS/V Nuclei must be filled with vowels rather than consonants.
This constraint is never violated in Persian, and has consequently been awarded a
high rank. The ranking in (38) explains the counterfeeding opacity in the abovementioned examples: (38) ONSET, NUCLEUS/V>> *neEz >> MAX[[n] & [eEz]]σ >> *n] >> MAX
Tableau (39) demonstrates that the opaque candidate (a) whose Ezafe vowel is only deleted is optimal.
(39) counterfeeding between final /n/ deletion and Ezafe deletion
The Singular Pronoun “in”
In the following words, at first glance it seems that the final vowel [i] is a morpheme meaning “his/her’.
(40)
But the problem is that this morpheme does not have such a meaning in Persian. A closer look reveals that it is in fact the Persian pronoun “in” (meaning ‘this’) whose last /n/ is deleted together with the preceding Ezafe vowel. Also, in plural examples the plural suffix “-ɑn” has undergone pre-nasal raising. Accordingly, the UR of the words in (40) would be the following:
(41)
Given the derivation “/doʃman+ɑn+e #in/ → [doʃmanuni]” for instance, if we look at its PR we will find the following four facts which suggest that this derivation is not opaque:
a) The consonant [n] at the end of [doʃman] is not deleted. Since it is not the last segment in the PR, the conditioning environment for its deletion is not met.
b) The plural suffix is [un] rather than *[ɑn]. Pre-nasal raising has happened because its conditioning environment is met.
c) There is no Ezafe vowel [e] after the plural suffix [un] because due to the existence of [n] in this suffix, the conditioning environment for Ezafe vowel deletion is met.
d) As expected, the underlying final /n/ is not realized in the PR.
The following ranking explains these four facts:
(42) *[-high]N >> *neEz >> *n] >> IDENT[height], MAX, DEP
Tableau (43) demonstrates that the candidate (a) which has undergone pre-nasal raising, Ezafe vowel deletion and final /n/ deletion is optimal.
(43) Pre-nasal raising, Ezafe vowel deletion and final /n/ deletion
The Plural Pronoun “inɑn”
In the following words, the final morpheme [inu] means ‘their’.
(44)
Given “[maʃinuninu]” for instance, now we know that [maʃin] (an English loanword) means “car”, and [-un-] is in fact the plural suffix /ɑn/ whose vowel underwent pre-nasal raising. There remains [-inu] which has to mean “their” or “they/them”. In Persian the word “inɑn” means “they/them”. Regarding our previous analyses, it is not difficult to infer that [-inu] is the phonetic realization of /inɑn/ which underwent pre-nasal raising and final /n/ deletion consecutively. Accordingly, the UR of the words in (44) would be the following:
(45)
The processes involved in the conversion of /maʃin+ɑn+e #inɑn/ to [maʃinuninu] is shown in the following derivation:
(46)
As shown in the derivation in (46), there are two simultaneous applications of pre-nasal raising followed by final /n/ deletion and Ezafe vowel deletion. And just like the case of [doχtaru] in the derivation in (7), upon the deletion of /n/, the environment that triggered pre-nasal raising at the end of the word has disappeared in the PR. This is another instance of counterbleeding opacity between the second occurrence of pre-nasal raising and final /n/ deletion because despite the fact that the conditioning environment for pre-nasal raising is not met on the surface, it applies (overapplication). Moreover, if the rules were applied in the opposite order, final /n/ deletion would “bleed” pre-nasal raising by depriving it of the opportunity to apply.
As Torres-TamaritFootnote 63 notes, “GEN in HS is restrained by a gradualness conditionFootnote 64 on candidate generation by which candidates can introduce only one single “modification” with respect to the latest input.” The notion of gradualness requirement on GEN in HS is tied to faithfulness constraints, such that a single change in HS is equivalent to an unfaithful mapping.Footnote 65 In other words, the gradualness requirement only allows one “modification” or one change only, from an input form to each candidate in the candidate set. In HS as formulated by McCarthyFootnote 66 and subsequent works, as well as by Torres-Tamarit,Footnote 67 unrelated processes have to take place at different steps.
However, as depicted in the derivation in (46) since maʃinuneinun has undergone pre-nasal raising twice, it is two changes away from the input. If, in compliance with the gradualness condition that every candidate must only be one “modification” or one change away from the input, instead of maʃinuneinun we include maʃinuneinɑn and maʃinɑneinun as two candidates in step 1, both will violate *[-high]N and there will be no winner in this step. Therefore, HS would fail to analyze this counterbleeding opacity at the very first step. But if with regard to this fact that both “modifications” are identical we consider them one change, then we can include maʃinuneinun in step 1. This way our analysis survives the challenge. Since similar cases may be found in other languages, it would not be impossible that a candidate which undergoes the same process twice in the same step could also be included in the standard definition of gradualness condition. However, as HS is one of the latest versions which needs to be cross-linguistically attested much more work/data is necessary.Footnote 68
In multi-step tableau (47) under the ranking in (42), three steps are required before convergence: /maʃin+ɑn+e# inɑn/→ [maʃinuneinun] → [maʃinuneinu] → [maʃinuninu]. Therefore, in step 1, the input to GEN is the underlying form /maʃin+ɑn+e# inɑn/. As displayed, candidates (b), (c) and (d) (the faithful candidate) have fatally violated the undominated constraint *[ɑ]N because they failed to undergo pre-nasal raising. Thus, candidate (a) that has undergone two occurrences of pre-nasal raising, is selected as the input to GEN in step 2. In this step, the faithful [maʃinuneinun] violates *n], for it has final [n]. Hence, candidate (a) that has undergone final /n/ deletion is selected as the input to GEN in step 3. In step 3, the faithful [maʃinuneinu] violates *neEz for having the forbidden sequence of /n/ followed by Ezafe [e]. Therefore, candidate (a) that satisfies this constraint is selected as the input to GEN in step 4 (convergence). Finally, there is convergence in step 4. Underlying /maʃin+ɑn+e# inɑn/ has realized all of its potential for harmonic improvement under this grammar, so the output of EVAL and the input to GEN are identical.
(47) /maʃin+ɑn+e# inɑn/→ [maʃinuninu]
Nouns with Plural Marker [au]
In subsections 2.2 and 2.3, we argued how the Ghayeni plural marker [au] is made. In the examples in (48) and those shown later in (52), this plural marker changes to [a] when the word to which it is attached is a possessee:
(48)
Based on our previous analyses the UR of these possessees would be the following:
(49)
The processes involved in the conversion of /serɑ+ɑn+e# mɑ/ to [seran mɑ], according to our analysis, are shown in the following derivation:
(50)
As seen in the PR [seran mɑ], although the conditioning environment is ready for final /n/ deletion, it fails to apply. This is an instance of counterfeeding opacity. This phonetic representation fails to undergo final /n/ deletion because this process is incapable of affecting the output created by the Ezafe vowel deletion rule. As noted earlier regarding the data in section 4, final /n/ deletion has to apply before the Ezafe vowel deletion. This priority prevents the occurrence of final /n/ deletion, though its conditioning environment is met on the surface.
The application of pre-nasal raising plus Ezafe vowel deletion results in “serɑun mɑ“. But, as “ɑu” is not regarded a plural marker, its first vowel converts to [a]to yield “seraun mɑ“. Now the question raised is “why would “seraun mɑ“ change to [seran mɑ]?” Studying Ghayeni Persian reveals that this dialect never allows the sequence of a diphthong followed by [n]. This is the effect of the markedness constraint “*Diph[n]” whose satisfaction is obtained either from monophthongization or /n/ deletion. Since final /n/ deletion has already lost its chance to apply, monophthongization takes place. The ranking in (52) explains the counterfeeding opacity in the examples in (48):
(51) *[-high]N >> *neEz >> *Diph[n] >> MAX[[n] & [eEz]]σ >> *n] >> [au]PL >> IDENT[height], IDENT[back], MAX
Now we address the following examples:
(52)
Based on our previous analyses the UR of these possessees would be the following:
(53)
The processes involved in the conversion of /ʃou+ɑn+e# mɑ/ to [ʃan mɑ], according to our analysis, are shown in the following derivation:
(54)
As seen in the derivation in (54), there is no serial interaction between unrelated processes of hiatus resolution and Ezafe vowel deletion which occur simultaneously at two different points at the same level. The ranking in (55) which is the result of incorporating *ONSET in the ranking in (51) explains the counterfeeding opacity in the words in (52) as well as those in (49).
(55) *[-high]N >> *ONSET, NUCLEUS/V >> *neEz >> *Diph[n] >> MAX[[n] & [eEz]]σ >> *n] >> [au]PL >> IDENT[height], IDENT [back], MAX
Tableau (56) demonstrates that the opaque candidate (a) is optimal.
(56) Counterfeeding between final /n/ deletion and Ezafe deletion in a plural form
Conclusion
The theoretical machinery utilized in this research built directly on OT’s existing approaches involving a novel combination of theoretical tools. This paper explored the application and non-application of final /n/ deletion in different domains and environments in Ghayeni Persian. This productive phonological process is affected by several opaque counterbleeding and counterfeeding interactions as well as bleeding. The results of the study confirmed previous research that classic or parallel optimality theory is incapable of handling counterbleeding opacity. Therefore, HS was adopted to accommodate this type of opacity even though preventing the transparent candidate from winning (incorrectly) is a challenge for HS. However, thanks to the existence of the plural suffix in our data, with the incorporation of the morpheme-specific markedness constraint *ɑ]PL, our analysis survived the pitfall. So this analysis or its implications might extend to other cases/analyses where derived words undergo counterbleeding. More precisely, the transparent candidate could be ruled out if one of its morphemes which has unlawfully undergone a phonological process could be singled out by a morpheme-specific markedness constraint whose index refers to that morpheme. However, this claim needs to be explored through further research in other languages.
We discussed three cases of counterfeeding opacity in which /n/ never deletes when it is followed by another segment in the input, even if that segment is deleted by the application of a phonological process. In order to deal with this challenge we adopted POT using local constraint conjunction (LCC).
As another instance of non-occurrence of final /n/ deletion in Ghayeni Persian we addressed a case of bleeding interaction whereby word-finally /n/ surfaces if it is preceded by a glottal consonant within a coda cluster. In this environment, the glottal consonant deletes which consequently bleeds final /n/ deletion. Since in the bleeding order both rules theoretically have an equal chance to apply we used local constraint conjunction to dispose of the candidate which has undergone both final /n/ deletion and glottal deletion in defiance of the bleeding order.
Finally we argued that if in compliance with the gradualness condition that every candidate must only be one change away from the input, we include maʃinuneinɑn and maʃinɑneinun as two candidates in step 1, both will violate *[-high]N and there will be no winner in this step. Therefore, HS would fail to analyze this counterbleeding opacity. But if with regard to this fact that both “modifications” are identical we consider them one change, then we can include maʃinuneinun in step 1. This way our analysis survives the challenge. We concluded that as similar cases may be found in other languages, it would not be impossible that a candidate which undergoes the same process twice in the same step could also be allowed for in the standard definition of gradualness condition. However, as HS is one of the latest versions which is still not cross-linguistically well attested much more work/data is necessary.
