• Nie Znaleziono Wyników

Chapter 3: Theoretical approaches to epenthesis

3.1 Generative approaches

3.1.3 Optimality Theory

While Optimality Theory continued to employ some of the main postulates of SPE phonology, an important contribution of the theory was that OT systematically referred to syllable structure optimization in explanations of typological asymmetries.

The concept of markedness, on which the optimality-theoretic model relies, refers to the claim that linguistic structures are either ‘unmarked’, that is preferred cross-linguistically, or ‘marked’, that is avoided. Along these lines, CV syllables, which are widely preferred in many languages, are less marked than onsetless syllables. Hence, from the perspective of syllable structure, prevocalic epenthesis is driven by a requirement for syllables to have onsets (see, for instance, Prince and Smolensky, 1993; McCarthy and Prince, 1995). In terms of violable constraints, the requirement for a syllable onset is expressed with a markedness constraint ONSET.

(11)ONSET:syllables must have onsets (Prince and Smolensky, 1993: 17) However, segment insertion is prohibited by a faithfulness constraint DEP-IO

(12) DEP-IO: every segment of the output must have a correspondent in the input (McCarthy and Prince, 1995: 16)

80

Hence, for consonant epenthesis to take place, ONSET must be ranked higher than DEP -IO, which militates against insertion. The ranking is thus as follows.

(13) OT: ranking to generate epenthesis ONSET >> DEP-IO

Consonant epenthesis has been captured in OT, for instance, by McCarthy and Prince (1995: 26‒27). They analyze glide and glottal stop insertion in Madurese, an Austronesian language spoken in Indonesia. The illustrative data is given in (14) below (after McCarthy and Prince, 1995: 27). As shown in (14a), glides /j/43 and /w/share backness or rounding with the first vowel of the hiatus via spreading. A glottal stop is inserted when the hiatus starts with the low vowel /a/, which lacks a corresponding glide in Madurese, or the vowels forming hiatus are identical.

(14) Madurese: word-internal hiatus resolution UNDERLYING FORM SURFACE FORM GLOSS

a. /seaŋ/ [sejaŋ] ‘afternoon’

/rophia/ [rɔphijɤ] ‘wife’

/koa/ [khuwɤ] ‘cave’

b. /maen/ [mãɁẽn] ‘toy’

/leer/ [lɛɁɛr] ‘neck’

/soon/ [sɔɁɔn] ‘request’

An evaluation of candidates is provided in (15) below on the example of the input form /seaŋ/ ‘afternoon’.

(15) Madurese: tableau for /seaŋ/ (after McCarthy and Prince, 1995: 27) /seaŋ/ ONSET DEP-IO

 a. sejaŋ * b. se.aŋ *!

Candidate (15b) fatally violates ONSET, as it fails to provide the syllable with an onset.

Candidate (15a) violates the anti-insertion DEP-IO, as there is /j/ in the output, which

43 McCarthy and Prince (1995: 27) use the symbol /y/ for the high front glide. However, for the clarity of exposition, I replace it with the IPA symbol /j/.

81

is absent from the input, but it satisfies ONSET, which is ranked higher. Thus, candidate (15b) incurs a more serious violation, and candidate (15a) wins the evaluation.

Optimality Theory interprets the insertion of a prevocalic consonant as the result of the ranking of two constraints: a markedness constraint ONSET and a faithfulness constraint DEP. For a candidate which inserts a segment to provide an onset to win, ONSET must be ranked higher than DEP. However, the typological overview of prevocalic consonant epenthesis in Chapter 2 showed that prevocalic consonant epenthesis may occur in two distinct environments: in a vowel hiatus, and at the left edge of a vowel-initial constituent. Additionally, the quality of the epenthetic segment itself may differ. Apart from the position in which the process applies, the quality of the epenthetic consonant relies on other factors, such as segmental markedness.

Hence, OT needs to account for this cross-linguistic variation, and it has already proven useful in doing so (McCarthy and Prince, 1995; Keer, 1995; 1996; Rosenthall, 1994; 1997; Alber, 2001; Uffmann, 2007; Rubach, 2000; 2007; Czaplicki, 2007;

Staroverov, 2014). The following section provides an overview of how some of those analyses tackled the issue of prevocalic consonant epenthesis concerning the position of epenthesis and the quality of the epenthetic segment. The discussion is parallel to that in Chapter 2, with a focus on the two major aspects of consonant insertions: the site of epenthesis (3.1.3.1) and the quality of the epenthetic segment (3.1.3.2).

Additionally, section 3.1.3.3 provides an outline of an analysis couched in an extension of the standard OT, Derivational Optimality Theory (DOT), which, in contrast to standard OT, postulates the need for derivational levels. The levels provide an opportunity to explain, for instance, the choice of a different epenthetic segment word-initially and word-medially.

3.1.3.1 Site of epenthesis

In optimality-theoretic analyses, the main driver of the prevocalic consonant epenthesis is the constraint ONSET, which discriminates against onsetless syllables. In terms of a ranking of violable constraints, prevocalic consonant epenthesis is derived by ranking ONSET above the anti-insertion constraint DEP-IO. However, the systematization of the prevocalic consonant insertions in Chapter 2 indicated that the process applies in two specific environments, that is, in a vowel hiatus, and at the left edge of a constituent. While in some languages epenthesis has been postulated in both these sites, there are languages which restrict the application of the process to one environment only. The description of the above in OT is achievable by introducing

82

additional constraints which ban constituent-initial or constituent-medial insertions.

An appropriate ranking of those constraints produces the desired output.

As has been discussed in section 2.1 in Chapter 2, some languages, such as Modern Standard Ukrainian, resolve vowel hiatus inside words but allow onsetless syllables word-initially. Similarly, Axininca Campa (Payne, 1981), does not permit insertions word-initially, a fact which has become a textbook example of such analyses in OT (McCarthy and Prince, 1993; 1995; Kager, 2004). In Axininca Campa, word-internal vowel hiatus is resolved by epenthesis, as shown in the examples in (16) below (after Kager, 2004: 101).

(16) Axininca Campa: heteromorphemic hiatus resolution

UNDERLYING FORM SURFACE FORM GLOSS

/no-N-koma-i/ [noŋkomati] ‘I will paddle’

/no-N-koma-aa-i/ [noŋkomataati] ‘I will paddle again’

/no-N-koma-ako-i/ [noŋkomatakoti] ‘I will paddle for’

/no-N-koma-ako-aa-i-ro/ [noŋkomatakotaatiro] ‘I will paddle for it again’

Inside words, epenthesis is allowed by the constraint ranking ONSET >> DEP-IO.The evaluation of candidates for the input /no-N-koma-i/ under such a ranking is provided in (17) below.

(17)Axininca Campa: tableau for /no-N-koma-i/

/no-N-koma-i/ ONSET DEP-IO

 a. noŋkomati *

b. noŋkomai *!

Candidate (17b) incurs a fatal violation of ONSET by having an onsetless syllable word-initially. Candidate (17a) provides the syllable onset and violates the anti-insertion DEP-IO, which, however, is ranked lower than ONSET. Thus, candidate (17a) wins the evaluation. However, while the onset-driven consonant epenthesis is accepted inside words, the process is not found word-initially, where Axininca Campa allows onsetless syllables, as illustrated with the examples in (18) below.

83

(18) Axininca Campa: no epenthesis word-initially (after Kager, 2004: 110) UNDERLYING FORM SURFACE FORM UNATTESTED

SURFACE FORM

GLOSS

/osampi/ [osampi] *[tosampi] ‘ask’

/i-N-koma-i/ [iŋkomati] *[tiŋkomati] ‘he will paddle’

To account for a system which shows epenthesis inside words, but not word-initially, another constraint is required. The current ranking of DEP-IO higher than ONSET makes incorrect predictions for the words in (18), as in the evaluation the winning candidates would be *[tosampi] *[tiŋkomati] which insert /t/ word-initially.

Kager (2004: 111) notes that the lack of word-initial epenthesis refers to a broader phenomenon, namely the alignment of grammatical and prosodic structures. In terms of violable constraints, the need for such an alignment is expressed by ALIGN-L.

(19) ALIGN-L

‘The left edge of the Grammatical Word must coincide with the left edge of the Prosodic Word’ (McCarthy and Prince, 1993: 39)

A potential output candidate *[tosampi] which violates ALIGN-L by misaligning the left edge of the grammatical word (GrWd) and prosodic word (PrWd) is represented graphically in (23) below (after Kager, 2004: 111). The inserted /t/ is syllabified into the onset of the first syllable thus satisfying ONSET, but it misaligns the left edge of PrWd and GrWd.

(20) Axininca Campa: misalignment of *[tosampi]

PrWd σ σ σ t o s a m p i GrWd

84

To block epenthesis in a word-initial position, ALIGN-L must be ranked higher than ONSET and DEP-IO. The ranking which allows for word-initial onsetless syllables by respecting alignment is provided in (21) below.

(21) Ranking to generate consonant epenthesis respecting the alignment ALIGN >> ONSET >> DEP-IO

The evaluation of candidates for the input /osampi/ is given in (22) below.

(22) Axininca Campa: tableau for /osampi/

/osampi/ ALIGN-L ONSET DEP-IO

 a. osampi *

b. tosampi *! *

The misaligned candidate (22b) */tosampi/ fatally violates the undominated ALIGN-L by inserting a consonant word-initially. Candidate (22a) violates ONSET, but respects ALIGN-L and consequently wins the evaluation.

The typology of prevocalic consonant insertions presented in Chapter 2 showed that some languages do not tolerate word-initial onsetless syllables and prefer filled onsets, which are provided by epenthesis. Accounting for the prothesis in OT requires a ranking different from the one in the tableau in (22) above. Sometimes, this distinction is necessary for the dialectal varieties of a language. For instance, the ranking in (22) appropriately accounts for the lack of word-initial epenthesis in Modern Standard Ukrainian, where a glide is inserted in word-medial hiatus, but not word-initially. An evaluation of candidates for the input /izrailj/ (Ізраїль, transliterated as Izrajil) ‘Israel’ is provided in (23) below.

(23) Modern Standard Ukrainian: tableau for /izrailj/ /izrailj/ ALIGN-L ONSET DEP-IO

 a. izra.jilj * *

b. izra.ilj **!

c. jizra.jilj *! **

d. jizra.ilj *! * *

85

The evaluation of candidates in (23) above shows that ranking ALIGN-L over ONSET

results in eliminating candidates (23c) and (23d), which insert a glide word-initially.

Candidate (23b) is eliminated as a result of violating ONSET (twice). Consequently, candidate (23a) which provides word-medial onset and preserves the word-initial onsetless syllable wins the evaluation.

However, the ranking in (23) above for Modern Standard Ukrainian fails to account for word-initial epenthesis in the south-western dialects of Ukrainian, in which a word-initial syllable onset is provided by the insertion of /ɦ/, for example in the word helevator [ɦɛlɛvatɔr] ‘elevator’. The relevant examples are provided in (24) (after Czaplicki, 2007: 28). The forms from dialectal Ukrainian are juxtaposed with their Standard Ukrainian counterparts.

(24) South-western Ukrainian dialect: prothesis

SPELLING GLOSS

South-western Ukrainian Modern Standard Ukrainian

harmija armija ‘army’

hartyst artyst ‘artist’

helevator elevator ‘elevator’

hoko oko ‘eye’

Contrary to the evaluation of candidates in Modern Standard Ukrainian in (24) above, a different ranking is required to explain word-initial consonant epenthesis in dialectal Ukrainian. As ALIGN-L penalizes word-initial insertion, it needs to be ranked lower than ONSET. Hence, to arrive at the output with word-initial consonant epenthesis, the ranking in (25) below is needed. Ranking ALIGN-L higher than ONSET

would preclude word-initial epenthesis.44

(25) Ranking to generate word-initial epenthesis ONSET >>ALIGN-L, DEP-IO

The evaluation of candidates for the input /ɛlɛvatɔr/ is provided in the tableau in (26) below (after Czaplicki, 2007: 28).

44In his analysis, Czaplicki (2007: 28) uses the modified version of ALIGN-L, namely ALIGN-L (stem, σ).

However, as in the present analysis ALIGN-L brings identical results, I use ALIGN-L.

86

(26) South-western Ukrainian: tableau for /ɛlɛvatɔr/

/ɛlɛvatɔr/ ONSET ALIGN-L DEP-IO

 a. ɦɛlɛvatɔr * *

b. ɛlɛvatɔr *!

Candidate (26b) incurs a fatal violation by failing to provide the word-initial syllable with the requisite onset. Candidate (26a) violates ALIGN-L and DEP-IO, but simultaneously satisfies high-ranked ONSET and thus constitutes the optimal output.

The typological overview in Chapter 2 indicated that cross-linguistically there are languages, which unlike Axininca Campa tolerate word-medial onsetless syllables, while they require syllables to have onsets at the beginning of words. A language that exhibits such a preference is Bulgarian, as already discussed in Chapter 2. For convenience, the linguistic examples from Bulgarian are repeated in (27) below (after Rubach, 2000: 287).45

(27) Bulgarian: word-initial glottal stop epenthesis and word-medial hiatus SPELLING SURFACE FORM GLOSS

a. urnata [ʔu] ‘the urn’

Amerika [ʔa] ‘America’

Irak [ʔi] ‘Iraq’

b. dialekt [ia] ‘dialect’

patriot [io] ‘patriot’

egoist [oi] ‘egoist’

Rubach (2000: 288; citing Kenstowicz, 1994) proposes the constraint CONTIGUITY to explain the absence of epenthesis inside words. The constraint is formulated in (28) below.

(28) CONTIGUITY

‘The output form standing in correspondence forms a contiguous string’

The constraint CONTIGUITY militates against epenthesis in all environments except the string edges, so that the contiguity of input segments could be maintained. To preserve

45In transcription I follow the original convention used by Rubach (2000: 287) by transcribing only the relevant strings.

87

the word-internal hiatus, CONTIGUITY must outrank ONSET. The ranking is provided in (29) and the evaluation follows in the tableau in (30) (after Rubach, 2000: 288).

(29) Bulgarian: ranking tolerating word-internal hiatus CONTIGUITY >>ONSET >> DEP-IO

(30) Bulgarian: tableau for /dia/

/dia/ CONTIGUITY ONSET DEP-IO

 a. di.a *

b. di.ja *! *

The tableau in (30) above illustrates the evaluation of candidates for the input /dia/.

The faithful candidate (30a) wins, as it does not insert a segment and respects the undominated CONTIGUITY. Candidate (30b) commits a fatal violation by inserting a segment within a string, thus violating the contiguity of segments in the output.

The data in (27) above show that while in Bulgarian, word-internal hiatus remains unresolved, as in (27b), the word-initial syllable onset is provided with an epenthetic glottal stop (27a). This typological preference is reflected by ranking ONSET over ALIGN-L. The evaluation for the input /irak/ ‘Iraq’ is given in (31) below.

(31) Bulgarian: tableau for input /irak/

/irak/ CONTIGUITY ONSET ALIGN-L DEP-IO

 a. Ɂi.rak * *

b. i.rak *!

The evaluation is (31) is won by the candidate (31a) which violates ALIGN-L by inserting a glottal stop word-initially, but respects ONSET. Candidate (31b) incurs a fatal violation of ONSET by having a word-initial onsetless syllable.

The ranking for Bulgarian in (31) above reflects the prohibition of word-medial epenthesis and a need for filled onsets at the beginning of words. On a similar note, dialectal Ukrainian requires syllables to have onsets word-initially. However, as standard Ukrainian also shows word-internal epenthesis, ONSET must be ranked above CONTIGUITY.

88

As has been said in section 2.1.2.1, a glottal stop is sometimes inserted to mark the beginning of a metrical constituent such as the stressed syllable. This is the case, for instance, in German. For convenience, the illustrative examples are repeated in (32) below.

(32) German: glottal stop insertion in vowel-initial stressed syllables SPELLING SURFACE FORM GLOSS

a. Atem [ˈʔa:təm] ‘breath’

Opa [ˈʔo:pa] ‘grandpa’

Oase [ɁoˈɁaze] ‘oasis’

b. Theater [teˈʔatɐ] ‘theatre’

Poet [poˈʔe:t] ‘poet’

c. Thea [ˈtea] ‘proper name’

Poesie [poeˈzi:] ‘poetry’

In OT, the absence of epenthetic material in the words in (32c) may be explained, as in Bulgarian, by a high ranking of the constraint CONTIGUITY, which militates against intrusion. In other words, all segments standing in correspondence in the input must remain contiguous in the output. To generate word-medial, onsetless unstressed syllables, the ranking in (33) is required. The tableau in (34) below shows an evaluation of the candidates for the input /tao/ (Alber, 2001: 14).

(33) Ranking not generating word-medial epenthesis in unstressed syllables CONTIGUITY >>ONSET >>DEP-IO

(34) German: tableau for /tao/

/tao/ CONTIGUITY ONSET DEP-IO

 a. ˈta.o *

b. ˈta.Ɂo *! *

As in the case of Bulgarian, the faithful candidate, (34a), wins the evaluation. It violates ONSET, but at the same time respects CONTIGUITY, which is ranked higher.

Candidate (34b) incurs a fatal violation of CONTIGUITY by inserting a segment inside a word. However, the data in (32b) above show that word-medial stressed syllables do not remain onsetless. They show glottal stop insertion. Alber (2001: 14-15) argues

89

that the increased articulatory effort the stressed syllables require is accompanied by a higher subglottal pressure, which eventually leads to a glottal closure. In terms of violable constraints, Alber (2001: 13) proposes that all stressed syllables should be subject to the ONSET (STRESS) constraint.

(35) ONSET (STRESS)

‘All stressed syllables must have onsets’ (Alber, 2001: 13)

As word-medial stressed syllables must obligatorily have an onset, a revised ranking is given in (36) and the evaluation of candidates for the input /oaze/ ‘oasis’ follows in the tableau in (37).

(36) Ranking generating epenthesis at the beginning of a stressed syllable ONSET (STRESS)>>CONTIGUITY >>ONSET >>DEP-IO

(37) German: tableau for /oaze/

/oaze/ ONSET (STRESS) CONTIGUITY ONSET DEP-IO

 a. ɁoˈɁaze * **

b. Ɂoˈaze *! *

c. oˈaze *! **

d. oˈɁaze * *! *

The evaluation for the input /oase/ in the tableau in (37) above shows that candidate (37a), which provides an onset for both word-initial and word-medial stressed syllables wins the evaluation and is selected as the optimal candidate, despite violating CONTIGUITY. The remaining candidates fatally violate the undominated ONSET (STRESS) (37b), (37c) or violate ONSET in addition to CONTIGUITY (37d).

In sum, an analysis couched in Optimality Theory translates the requirement for filled onsets as the markedness constraint ONSET, which militates against onsetless syllables. Some languages ban onsetless syllables in certain environments, for instance inside words, but permit them word-initially. This typological preference can be captured with the OT machinery by imposing additional constraints which need to be ranked adequately to express the typology of a given language. For instance, word-initial insertion is banned by ranking ALIGN-Lhigh, while word-internal insertion is prohibited by ranking CONTIGUITY high. In languages such as German, which insert

90

segments prevocalically at the beginning of stressed syllables, the constraint ONSET

(STRESS) must be ranked high.

3.1.3.2 The quality of the epenthetic segment

While the constraints and their rankings discussed in the previous section can regulate the position of epenthesis, they do not govern the quality of the inserted segment. The main goal of this section is to illustrate how Optimality Theory deals with this issue.

In other words, it will be shown what mechanisms are required to ban segments of a certain quality and allow those which are required on the surface. To do so, it is necessary to introduce additional markedness constraints, which impose some restrictions with regard to the segment quality.

Markedness is one of the tenets of optimality-theoretic analyses. The notion of markedness may also refer to the place of articulation of consonants, as has already been discussed in Chapter 2. Some consonants have been argued to be more marked than others. Consonants are ranked from the most to the least marked with respect to their place of articulation. Glottals are considered the least marked segments, and consequently, they are predicted to be the most frequently used epenthetic material.

Coronals, labials, and dorsals follow in the hierarchy of the most frequently inserted consonants. In terms of OT, markedness is represented in terms of constraints that ban specific places of articulation. Candidates which contain dorsal consonants incur a violation of *DORSAL and those which insert labial consonants violate *LABIAL. Insertion of coronal segment results in violating *CORONAL, and the insertion of glottal stop violates *LARYNGEAL, a constraint that militates against pharyngeal segments. The ranking which reflects the predictions of markedness concerning the place of articulation is given in (38). Violations for each of the constraints referring to the place of articulation are depicted in the tableau in (39) below (after Vaux, 2002: 6‒

7).46

(38) Ranking reflecting the markedness of the place of articulation

*DORSAL,*LABIAL >>*CORONAL >>*LARYNGEAL

46 The tableau proposed by Vaux (2002: 6-7) does not include violations of *DORSAL. The violations of

*DORSAL in (39) are added by the author of this dissertation.

91 (39) Markedness: tableau for /gao/

/gao/ *DORSAL *LABIAL *CORONAL *LARYNGEAL

 a. ga.Ɂo * !

c. ga.to * *!

c. ga.bo * *!

d. ga.go **!

However, the sole notion of markedness with regard to the place of articulation is not sufficient to explain certain phenomena, such as glide insertion. The OT model has been used to analyze glide insertion, for instance, by Rubach (2000) for the data from colloquial Slovak. In this non-regional variety of the language, /j/ is inserted in hiatus if one of the flanking vowels is /i/, thus spreading its features (as in, for instance, Dutch and Polish). The relevant linguistic examples are given in (40) below.

(40) Colloquial Slovak: word-internal /j/ insertion (Rubach, 2000: 284) SPELLING UNDERLYING FORM SURFACE FORM GLOSS

trio /trio/ [tri.jo] ‘trio’

diéta /dieːta/ [di.jeːta] ‘diet’

intuitívny /intuitiːwni/ [in.tu.ji.tiːw.ni] ‘intuitive’

altruizmus /altruizmus/ [al.tru.jiz.mus] ‘altruism’

One may wonder why a homorganic glide, rather than a glottal stop is inserted.

Rubach (2000: 284, citing Keer, 1995) proposes a constraint *[cg] which bans glottal stops. The constraint is formulated in (41) below.

(41) *[cg]

‘Do not have feature [constricted glottis]’

In order to eliminate a candidate which resolves hiatus with glottal stop insertion, the ranking given in (42) below is required. An evaluation of the candidates follows in the tableau in (43).47

47In his analysis, Rubach (2000: 274) uses a modified version of DEP-IO, that is, DEP(SEG), which prohibits segment insertion. However, as DEP-IO brings identical results, I use DEP-IO.

92 (42) Ranking promoting epenthetic /j/ over /Ɂ/

*[cg] >> ONSET >> DEP-IO

(43) Colloquial Slovak: tableau for /dia/ (after Rubach, 2000: 285) /dia/ *[cg] ONSET DEP-IO

 a. di.ja *

b. di.a *!

c. di.Ɂa *! *

Candidate (43a) wins the evaluation, as it incurs the least costly violation of DEP-IO by inserting /j/. 48The remaining candidates are eliminated as they violate ONSET (43b), and *[cg] by inserting a glottal stop (43c).

In some languages, however, the situation is the reverse, that is, insertion of a glottal stop is preferred over the insertion of a glide. For example, this is the case in Bulgarian. As was pointed out in the previous section, word-initially, the syllable onset is provided by an inserted glottal stop in words such as Irak /Ɂirak/ ‘Iraq’. Rubach’s (2000: 288) analysis assumes that the constraint which favors a glottal stop over a glide is NO-MULTIPLE-LINK (*MULIT-LINK), which bans candidates with a single melodic segment linked to more than one X-slot (recall the discussion in 3.1.2).

(44) NO-MULTIPLE-LINK (*MULIT-LINK)(Rubach, 2000: 288)

‘Melodic segments cannot be linked to more than one X-slot’

Since the epenthetic glides are spawned by vowels, they are linked to two X-slots, and consequently, they are prohibited by *MULTI-LINK. To promote glottal stop insertion over glide insertion, *MULTI-LINK must outrank *[cg], thus the constraint which bans glottal stop insertion. Since the epenthesis occurs word-initially, ALIGN-L, already introduced in the previous section is at play. For the candidate with glottal stop prothesis to win, ALIGN-L must be ranked lower than ONSET. The ranking is provided in (45) below and the evaluation follows in the tableau in (46).

48Note that according to the markedness scale introduced in Chapter 2, coronal sounds are more marked than glottal sounds. Therefore, in (43), *CORONAL needs to outrank *[cg]. However, as will be shown in Chapter 4, epenthesis of glides which result from spreading fares better than insertions of segments such as glottal stops.

93

(45) Ranking promoting glide insertion over glottal stop insertion

*MULTI-LINK >>ONSET >> ALIGN-L, *[cg]

(46) Bulgarian: tableau for /irak/ (after Rubach, 2000: 288)

*MULTI-LINK ONSET ALIGN-L *[cg]

 a. Ɂi.rak * *

b. i.rak *!

c. ji.rak *! *

While both candidates (46a) and (46c) violate ALIGN-L by inserting a segment word-initially, candidate (46a) commits a less costly violation of*[cg], which has the lowest position in the ranking. Candidate (46b), on the other hand, fatally violates ONSET.

Another approach to the quality of the epenthetic segment has been proposed by Uffmann (2007), who bases his analysis on the notion of prominence. He argues that the choice between a glottal stop and a glide relies on the prosodic position and prominence contrast. Under this view, one should expect glottal stops at the beginning of a constituent, as the glottal stop enlarges the contrast with the neighboring vowel, and a glide intervocalically to minimize the contrast. Originally, Prince and Smolensky (1993: 149) propose two prominence scales which describe markedness with regard to the syllable position (47a) and the prominence (sonority) of individual segments.

The scale in (47a) recognizes syllable Peaks (nuclei) and Margins (onset, codas), and the scale in (47b) is based on the sonority of segments. Note that the symbol ‘>’ stands for ‘is more prominent than’.

(47) Syllabic and segmental prominence scales (Uffmann, 2007: 459, citing Prince and Smolensky, 1993)

a. Peak > Margin

b. Vowels > r > l > nasals > obstruents > laryngeals

If the scales from (47) above are combined, the result is that prominent segments from (48b) usually align with the prominent syllable position from (48a). In other words, vowels are most common as syllable peaks (nuclei) and other segments fill syllable margins. Thus, vowels are the most marked segments in syllable margins, and the least marked in syllable peaks. The cumulative scales for Margins (48a) and Peaks (48b)

94

are provided in (48) below (after Uffmann, 2007: 259, citing Prince and Smolensky, 1993).

(48) Combined syllabic and segmental harmony scales

a. *Margin/V >> *Margin/r >> *Margin/l >> *Margin/nas >> *Margin/obs >>

*Margin/lar

b. *Peak/lar >> *Peak/obs >>*Peak/nasal >>*Peak/l >> *Peak/r >> *Peak/V Uffmann (2007: 460) proposes a prominence-based analysis of glottal stop insertion in OT. He observes that while the ranking of ONSET above DEP allows epenthesis, it does not account for the quality of the epenthetic segment. Thus, basing on the harmony scales in (48) above he proposes the following ranking (49) and an evaluation of candidates for the input /ɛlç/; Elch ‘moose’ (50).

(49) German: ranking generating word-initial glottal stop insertion

ONSET >> DEP >> *Margin/V >> *Margin/nas >> *Margin/obs >> *Margin/lar (50) German: tableau for /ɛlç/

/ɛlç/ ONSET DEP *Margin/

V

*Margin/

nas

*Margin/

obs

*Margin/

lar

 a. [Ɂɛlç] * *!

b. [ɛlç] *!

c. [tɛlç] * *!

d. [nɛlç] * *!

e. [jɛlç] * *!

Candidate (50a), which inserts a glottal stop prevocalically wins the evaluation, as it violates the lowest-ranked *Margin/lar by inserting a laryngeal segment in a syllable margin. The fully-faithful candidate (50b) is eliminated as it fails to satisfy ONSET. Candidates (50c), (50d), (50e) violate *Margin/obs, *Margin/nas, and *Margin/V, respectively.

As for the intervocalic glide insertion, Uffmann (2007) proposes the following, modified analysis. As the VV sequences are prominent themselves, a more prominent separating segment is required. The glottal stop, which is preferred in syllable margins, is not an optimal epenthetic consonant in vowel hiatus. Thus, according to

95

Uffmann (2007: 461), intervocalic onsets are treated as peaks rather than margins.

This assumption is reflected by the ranking in (51). The evaluation of candidates for the input /ia/, which is a common vocalic cluster in many of the world’s languages, is given in the tableau in (52).

(51) Prominence-based ranking generating intervocalic glide insertion *V_V/lar >> *V_V/obs >> *V_V/nas >> *V_V/l >> *V_V/r >> *V_V/V (52) Tableau for /ia/

/ia/ ONSET DEP *V_V/lar *V_V/obs *V_V/nas *V_V/V

 a. ija * *

b. ia *!

c. iɁa * *!

d. ita * *!

e. ina * *!

Candidate (52a), which inserts a glide in the intervocalic position, minimally violating Dep and the lowest *V_V/V, wins the evaluation. The other candidates either fail to provide the syllable onset, (52b), or insert a consonant different than a homorganic glide, (52c‒e).

To summarize, Uffmann’s (2007) account discussed above makes correct predictions about epenthesis in the intervocalic and word-initial positions. The analysis is in line with the typological overview in Chapter 2; glides are common hiatus-breakers, while glottal stops are often found at constituent edges. The analysis, however, runs into difficulties in cases where a homorganic glide is inserted word-initially before a vowel (recall section 2.2.1 in Chapter 2 on prothesis in dialectal Polish and Kashubian). The difficulty results from the fact that the prominence based-theory does not predict glide epenthesis in syllable margins word-initially, as illustrated in the tableau in (53) below. The evaluation of candidates is done for a rural Polish pronunciation of the word inny /jinnɨ/ ‘different’. The input is /innɨ/ and the ranking applied is the one governing segment quality in syllable margins in (53).

96 (53) Rural Polish: tableau for /innɨ/

/innɨ/ ONSET DEP *Margin/

V

*Margin/

nas

*Margin/

obs

*Margin /lar

 a. Ɂinnɨ * *

 b. jinnɨ * *!

c. tinnɨ * *!

d. ninnɨ * *!

The evaluation in the tableau in (53) above shows that candidate (53b), which represents the actual surface form, is eliminated as a result of violating *Margin/V which prohibits glides in syllable margins. Under the assumed ranking, the promoted candidate is *[Ɂinnɨ], which does not find confirmation in the facts.

In this section, it has been shown how OT tackles the phenomenon of prevocalic consonant epenthesis. Crucially, accounts of the process involve an interaction of two constraints – ONSET and DEP. The former must outrank the latter to legitimize consonant insertion. The restrictions pertaining to the site of epenthesis and the quality of the epenthetic segment need to introduce additional constraints and rank them appropriately to generate the process. In each language discussed, the restrictions on the site and the quality of the epenthetic segment are reflected by different rankings of the constraints. The rankings and their consequences are summarized in Table 1 below, with the distinction into rankings dealing with the site of epenthesis given in (a) and the quality of the epenthetic segment provided in (b).

97 Table 1

CONSTRAINT RANKING RESULT

a. ONSET >> DEP-IO Prevocalic epenthesis ALIGN-L>>ONSET >> DEP-IO No word-initial epenthesis ONSET >>ALIGN-L >> DEP-IO Word-initial epenthesis

CONTIGUITY >> ONSET >> DEP-IO No epenthesis word-internally CONTIGUITY >> ONSET >>ALIGN-L>>DEP-IO No epenthesis word-internally;

word-initial epenthesis

ALIGN-L>> ONSET >>CONTIGUITY >>DEP-IO No word-initial epenthesis;

word-medial epenthesis ONSET (STRESS) >> CONTIGUITY >> ONSET >>

DEP-IO

Epenthesis in stressed syllables

b. *[cg] >> ONSET >> DEP(SEG) Insertion of /j/ and not /Ɂ/

*V_V/lar >> *V_V/obs >> *V_V/nas >>

*V_V/l >> *V_V/r >> *V_V/V

Insertion of /j/ and not /Ɂ/ in syllable peaks

*MULTI-LINK >> ONSET >> ALIGN-L >> *[cg] Insertion of /Ɂ/ and not /j/

ONSET >> DEP >> *Margin/V >>

*Margin/nas >> *Margin/obs >>

*Margin/lar

Insertion of /Ɂ/ and not /j/ in syllable margins

While the above set of rankings successfully describes epenthesis in certain conditions, the analyses within standard OT turned out to have difficulties explaining phenomena such as word-initial glide insertion. This issue has been addressed within an extension of OT, that is Derivational Optimality Theory, which is discussed in the following section.

3.1.3.3 Derivational Optimality Theory

Derivational Optimality Theory (DOT henceforth; Rubach, 1997; 2000) is an extension of standard OT which retains level distinction from Lexical Phonology (Kiparsky, 1982; Booij and Rubach, 1987). In DOT, the output of an evaluation at Leveln is an input to the evaluation at Leveln+1. The maximal number of derivational levels is 4, and each level may have a different ranking of constraints. Rubach (2011: 245) observes that the levels are not arbitrary, as they ‘need to be linked to grammatical domains, much as in Lexical Phonology. Thus, level 1 comprises the stem which is