Create an Optimality Theory tableau for English plural allomorphy showing /z/, /s/, and…
- User
Create an Optimality Theory tableau for English plural allomorphy showing /z/, /s/, and /iz/. Define MAX-IO, DEP-IO, and AGREE(voice) constraints and rank them to explain why cat+z becomes cats with /s/.
Assistant
This is a great question about Optimality Theory (OT) applied to English plural allomorphy. Let me work through this systematically.
English Plural Allomorphy
In English, the regular plural suffix has three allomorphs:
- /ɪz/ after sibilants: buses /bʌsɪz/, boxes /bɑksɪz/, judges /dʒʌdʒɪz/
- /z/ after voiced non-sibilants: dogs /dɔgz/, beds /bɛdz/, babes /beɪbz/
- /s/ after voiceless non-sibilants: cats /kæts/, mats /mæts/, books /bʊks/
The question asks specifically about cat+z → cats /s/.
The Problem
The underlying representation of the plural affix is typically /z/. So we have:
- /kæt/ + /z/ → [kæts]
This needs to be explained through constraint interaction.
Key Constraints
MAX-IO (Maximality - Input-Output)
This constraint requires that every segment in the input has a correspondent in the output. Violation occurs when a segment is deleted.
- MAX-IO: No input segment is deleted
DEP-IO (Dependency - Input-Output)
This constraint requires that every segment in the output has a correspondent in the input. Violation occurs when a segment is inserted/epenthesized.
- DEP-IO: No output segment lacks an input correspondent
AGREE(voice)
This constraint requires that adjacent obstruents agree in their voicing features.
- AGREE(voice): Adjacent obstruents must have the same voicing value
Actually, let me think about this more carefully. The traditional way to handle this in OT would be to have:
-
A faithfulness constraint MAX-IO that penalizes deletion
-
A faithfulness constraint DEP-IO that penalizes insertion
-
A markedness constraint AGREE(voice) that penalizes sequences of obstruents with different voicing
For cat+z → cats:
- The input is /kæt+z/
- We need to explain why we get [kæts] and not [kætz]
The problem is that /tz/ would violate AGREE(voice) because [t] is voiceless and [z] is voiced. So the constraint ranking would need AGREE(voice) >> MAX-IO, DEP-IO to prioritize avoiding the voicing mismatch. But this creates a puzzle: we can't simply delete the /z/ without violating MAX-IO, and we can't keep it as [z] without violating AGREE(voice). The actual output [kæts] suggests the /z/ undergoes voicing assimilation to match the preceding [t], which would satisfy AGREE(voice) while preserving the input segment. The assimilation of /z/ to [s] involves a violation of IDENT(voice), since the voicing feature changes. When the stem ends in a voiceless obstruent like /t/, the /z/ devoices to match. This could be handled through feature assimilation or by selecting different allomorphs of the plural morpheme—/s/ after voiceless sounds, /z/ elsewhere, and /ɪz/ after sibilants.
Setting up a tableau for /kæt+z/, the key constraints are MAX-IO (no deletion), DEP-IO (no insertion), AGREE(voice) (obstruents in clusters must match for voicing), and IDENT(voice) (preserve voicing from input to output). The underlying /z/ cannot surface unchanged because it violates AGREE(voice) when following the voiceless /t/. The winning candidate devoices the /z/ to [s], satisfying the agreement constraint at the cost of violating IDENT(voice). The constraint AGREE(voice) ranked high forces a resolution: devoicing the z to [s] emerges as the optimal candidate since it satisfies the voicing agreement requirement while violating only IDENT(voice).
English plurals follow a systematic pattern—voiceless obstruents trigger [s], voiced obstruents trigger [z], and sibilants trigger [ɪz]—which reflects a core principle that stem-final and suffix obstruents must align in voicing. When AGREE(voice) dominates MAX-IO, the grammar prefers maintaining both segments over allowing them to disagree, so devoicing becomes the winning strategy rather than deletion or epenthesis. The key constraints are AGREE(voice) requiring adjacent obstruents to match, IDENT(voice) preserving input voicing, MAX-IO preventing deletion, and DEP-IO blocking epenthesis. Ranking AGREE(voice) >> IDENT(voice) >> MAX-IO >> DEP-IO explains why [kæts] wins: the devoiced cluster satisfies the agreement constraint even though it violates faithfulness to the original /z/, while [kætz] fatally violates AGREE(voice) by keeping the voicing mismatch between [t] and [z]. DEP-IO, 0 MAX-IO
- [kæt]: 0 AGREE, 0 IDENT, 0 DEP-IO, 1 MAX-IO
This ranking captures the pattern. Let me verify the constraint definitions are correct:
AGREE(voice) requires adjacent obstruents to share voicing features. MAX-IO ensures all input segments appear in the output. DEP-IO prohibits epenthesis. IDENT(voice) demands that if a segment surfaces, its voicing value must match the input.
For the input /kæt+z/, I need to evaluate which candidate satisfies the hierarchy. The candidates are [kæts], [kætz], [kætɪz], and [kæt]. With AGREE(voice) ranked highest, [kætz] violates it immediately since the voiceless [t] and voiced [z] are adjacent. The winning candidate [kæts] devoices the suffix to match the stem's final obstruent, satisfying AGREE(voice) while incurring a violation of IDENT(voice) on the suffix. This ranking correctly predicts regressive voicing assimilation where the suffix adapts to the stem. The ranking AGREE(voice) >> IDENT(voice) >> DEP-IO >> MAX-IO correctly predicts (a) kæts as the winner. The /z/ assimilates to [s] to satisfy the voicing agreement constraint, eliminating (b). Epenthesis candidates like (c) lose to simpler forms, and deletion (d) violates MAX-IO.
I should consider whether [kætsǝ] with full vowel epenthesis is even a viable candidate. English only inserts [ɪz] after sibilants, suggesting there's a constraint restricting epenthetic vowels to specific contexts rather than allowing them freely.
The standard OT approach treats the three plural allomorphs (/z/, /s/, /ɪz/) as lexical entries selected by constraint ranking. Alternatively, the plural morpheme could be underspecified for voice, with AGREE(voice) forcing assimilation when it conflicts with the stem-final segment. The input would be the root /kæt/ plus an underspecified plural marker, and constraints like AGREE(voice) and FAITHFULNESS determine which form surfaces.
For a clearer presentation, I should set up a tableau using AGREE(voice) as the highest-ranked constraint, followed by IDENT(voice), then MAX-IO and DEP-IO. This shows how /kæt/ + /z/ surfaces as [kæts]: AGREE(voice) requires the obstruent cluster to share voice features, IDENT(voice) preserves input specifications where possible, while MAX-IO and DEP-IO handle segment preservation and insertion respectively.
The tableau demonstrates that [kæts] wins because it satisfies the top constraint by devoicing /z/ to match /t/, incurring only an IDENT(voice) violation. Candidate [kætz] fatally violates AGREE(voice), [kætɪz] violates DEP-IO through epenthesis, and [kæt] violates MAX-IO through deletion. The ranking AGREE(voice) >> IDENT(voice) >> DEP-IO >> MAX-IO captures why assimilation is preferred over these other repair strategies. ranking should actually favor (a) over (b) since IDENT(voice) is ranked higher than AGREE(voice). But wait—that contradicts what I established earlier about AGREE(voice) being the highest constraint. Let me reconsider the constraint hierarchy.
If AGREE(voice) >> IDENT(voice) >> DEP-IO >> MAX-IO, then (b) should be ruled out first for violating AGREE(voice). But if the data shows (a) is actually the winning candidate, then perhaps IDENT(voice) outranks AGREE(voice) instead. Each candidate violates exactly one constraint at different levels, so the ranking determines which output surfaces. At IDENT(voice), candidate (a) loses because the /z/ devoices to [s]. Candidates (c) and (d) survive this constraint—(c) preserves the voiced /z/ while incurring an epenthesis violation, and (d) avoids the constraint entirely by deleting the segment. Moving to DEP-IO, (c) is eliminated for the epenthesized vowel, leaving (d) as the winner despite its MAX-IO violation being the least costly in this ranking.
But that can't be right—(a) should win. I need to reconsider the constraint hierarchy. The ranking must prioritize IDENT(voice) violations more heavily than I'm accounting for, or perhaps the constraints need reordering so that (a)'s devoicing is actually acceptable while (d)'s deletion is penalized more severely.
Let me check AGREE(voice) first: (b) clearly violates it with adjacent voiced obstruents, so it's out. For IDENT(voice), I'm getting confused about whether (a) actually violates it—if the /z/ undergoes devoicing to [s], that's a voice change, which should violate IDENT(voice). But if (a) is the correct winner, then either IDENT(voice) isn't ranked as high as I thought, or I'm misidentifying which candidate violates it. The issue might be that I need to reconsider what counts as an IDENT violation in this context, or the constraint ranking needs adjustment so that MAX-IO violations are penalized more than IDENT(voice) violations. to the [s] in the output—it's the same segment with a feature change, so MAX-IO is satisfied.
- (d) [kæt]: the /z/ has no correspondent in the output, so MAX-IO is violated.
At IDENT(voice) (after eliminating (b) and (d)):
- (a) [kæts]: the /z/ → [s] involves a voice feature change, violating IDENT(voice)
- (c) [kætɪz]: satisfies (no feature changes)
So (c) wins. The ranking AGREE(voice) >> MAX-IO >> IDENT(voice) >> DEP-IO correctly predicts that [kætɪz] is optimal because it satisfies the top constraints while (a) violates IDENT(voice) and (d) violates MAX-IO.
But [kætɪz] isn't the actual English plural form—that's [kæts]. This suggests the analysis needs reconsideration, perhaps by treating suffix allomorphy as a lexical choice rather than a phonological derivation.
If the plural suffix exists in the lexicon with three allomorphs—/z/, /s/, and /ɪz/—then the input becomes /kæt/ + PLURAL, and the candidates are the three possible combinations with each allomorph. The constraint AGREE(voice) would require the suffix to match the voicing of the stem's final consonant, which would favor /s/ after the voiceless /t/, making [kæts] the winner without needing to derive it through phonological rules. | ✓ | | c. kætɪz | ✓ | * | ✓ |
So (b) wins. The constraint ranking selects among the three lexical allomorphs based on how well they satisfy AGREE(voice). The [z] form violates the constraint since [t] and [z] have conflicting voicing, while [s] satisfies it because both obstruents are voiceless. The epenthetic [ɪz] also works since the vowel breaks up the cluster, avoiding any voicing mismatch between obstruents.
When I apply the ranking, [z] gets eliminated first at AGREE(voice), then [ɪz] falls out at DEP-IO for introducing the epenthetic vowel, leaving [s] as the winner and producing [kæts].
I should simplify this for clarity though. The core question asks me to define the three constraints, rank them, and show why cat+z surfaces as cats with the /s/ allomorph. The fundamental issue is that English has a single underlying /z/ suffix that cannot follow the voiceless /t/, so the grammar must select the voiceless [s] variant instead. iceless obstruent followed by a voiced one violates the constraint. So [kætz] fails because [t] and [z] have mismatched voicing.
MAX-IO prevents deletion—the /z/ can't just disappear, which rules out [kæt].
DEP-IO blocks epenthesis—inserting [ɪ] to separate the obstruents violates faithfulness to the input, eliminating [kætɪz].
The constraint hierarchy AGREE(voice) >> DEP-IO >> MAX-IO >> IDENT(voice) explains why assimilation wins: voicing agreement outranks both insertion and deletion, but allows the /z/ to change its voice feature to match [t].
In the tableau, [kæts] emerges as optimal because it satisfies the top three constraints while only violating the lowest-ranked IDENT(voice). The competing candidates each violate a higher-ranked constraint, so they're successively eliminated.
I should verify this against standard analyses in the literature. Work by McCarthy, Inkelas, and Spencer typically ranks AGREE high to prevent voicing mismatches, with faithfulness constraints ranked lower. The exact ordering between MAX-IO and DEP-IO can vary, but AGREE >> IDENT(voice) >> DEP-IO >> MAX-IO works well here since English preserves segments rather than deleting them to fix voicing conflicts—it either assimilates features, inserts a vowel, or selects an alternate form. This ranking captures why [kæts] (feature assimilation) wins over [kætɪz] (epenthesis) or [kæt] (deletion). The ranking eliminates [kætz] at AGREE(voice) since it violates the constraint with its voiceless-voiced cluster. But then at IDENT(voice), [kæts] gets ruled out for changing /z/ to [s], leaving [kætɪz] and [kæt] as candidates—which means the ranking predicts the wrong winner.
This suggests IDENT(voice) isn't the right constraint here. The core issue is that I need constraints that actually prefer [kæts] over both epenthesis and deletion.
When /z/ becomes [s], it's not necessarily a faithfulness violation in the traditional sense. In some OT frameworks, the /z/ and [s] don't correspond at all—instead, the morpheme selects between allomorphs, so /z/ gets deleted and /s/ gets inserted. This would violate MAX-IO and DEP-IO rather than IDENT(voice). The question then becomes whether AGREE(voice) >> MAX-IO and DEP-IO, or if some other ranking determines whether we tolerate the voicing mismatch or resort to epenthesis and deletion.
The ranking AGREE(voice) >> DEP-IO >> MAX-IO would explain why we get [kæts] rather than [kætɪz] or [kæt]—the constraint against voicing mismatches is strong enough to force selection of the [-voice] allomorph /s/ from the lexicon. Since /s/ is already specified as [-voice] in the lexicon, IDENT(voice) isn't actually violated; it's just a matter of which allomorph gets selected.
For clarity in explaining this to someone learning, I should present a straightforward tableau showing how these constraints interact rather than getting caught up in the technical details about feature specification.
Optimality Theory Analysis of English Plural Allomorphy
Constraint Definitions
MAX-IO (Maximality - Input-Output)
Definition: Every segment in the input must have a correspondent in the output.
Violation: Deletion of input material
Example violation: Deleting the /z/ in /kæt+z/ to get [kæt] violates MAX-IODEP-IO (Dependency - Input-Output)
Definition: Every segment in the output must have a correspondent in the input.
Violation: Insertion or epenthesis of segments not in the input
Example violation: Adding an epenthetic vowel to create [kætɪz] violates DEP-IOAGREE(voice)
Definition: Adjacent obstruents must share the same [±voice] specification.
Violation: Having a voiceless obstruent immediately followed by a voiced obstruent (or vice versa) in a cluster
Example violation: The cluster [tz] violates AGREE(voice) since [t] is [-voice] and [z] is [+voice]
OT Tableau
Input: /kæt/ + /z/
Candidates AGREE(voice) DEP-IO MAX-IO IDENT(voice) a. [kæts] ✓ ✓ ✓ * b. [kætz] *! ✓ ✓ ✓ c. [kætɪz] ✓ *! ✓ ✓ d. [kæt] ✓ ✓ *! ✓ Winning Candidate: ☞ [kæts]
Constraint Ranking
AGREE(voice) >> DEP-IO >> MAX-IO >> IDENT(voice)How the Ranking Explains /z/ → [s]
-
[kætz] is eliminated first by AGREE(voice)
- The cluster [tz] is voiceless-voiced, violating the highest-ranked constraint
- English cannot tolerate this voicing mismatch
-
[kætɪz] is eliminated next by DEP-IO
- This form inserts an epenthetic [ɪ] vowel
- Epenthesis violates the second-ranked constraint
- English prefers to modify features rather than insert material
-
[kæt] is eliminated next by MAX-IO
- This form deletes the plural suffix entirely
- Deletion violates the third-ranked constraint
- English prefers feature modification over loss of segmental material
-
[kæts] wins by default
- It respects all three primary constraints
- The /z/ devoices to [s] to satisfy AGREE(voice)
- This feature change only violates low-ranked IDENT(voice), which is acceptable
The Key Insight
This ranking captures why English plurals alternate this way:
- cats [kæts] = /s/ (after voiceless: /t/)
- dogs [dɔgz] = /z/ (after voiced: /g/)
- buses [bʌsɪz] = /ɪz/ (after sibilant: /s/)
In each case, AGREE(voice) forces the suffix allomorph to match the stem's voicing, ranked above all other phonological constraints.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.