On containment and syncretism : English preterites and participles

This study explores Bobaljik’s (2012) suggestion that in English, the feature representation of the preterite contains the representation of the past participle. While containment analyses in both Distributed Morphology (DM) and Nanosyntax capture the virtual absence of ABA patterns of syncretism for the order BASEPARTICIPLE-PRETERITE, I demonstrate that they face empirical challenges when the exponence of the suffixes is considered. After evaluating an alternative feature decomposition, I show how a DM containment approach can derive the facts for both base and suffix alternations with the aid of impoverishment, which also helps to explain counterexamples to *ABA in this domain. Lastly, I offer cautionary discussion about the relationship between containment structures and deriving *ABA.

To account for the typological absence of ABA patterns, some researchers have adopted a containment approach whereby the representation of a more marked member properly contains the representation of a less marked member, as seen in ( 1) for adjectival grade.Bobaljik (2012) reduces the absence of ABA patterns in adjectival grade to the containment structure in (1) and the Elsewhere Condition.For example, in (2), a more specific Vocabulary Item be(tt)is chosen over the default good in the context of CMPR.Because the superlative contains CMPR, it will also be realized with the more specific form, producing be-st (not *good-est).More generally, all else being equal, the positive and the superlative bases cannot be realized with the same exponent to the exclusion of the comparative.
(2) Vocabulary Items √ GOOD ↔ be(tt) / CMPR √ GOOD ↔ good Bobaljik (2012;158-163) suggests that a containment relation may hold between preterites and past participles in German and English.In this domain, too, the ABA pattern appears to be virtually absent ( The objective of this short article is to consider challenges to containment approaches to English preterites and participles.After discussing issues with several approaches, I offer a refined containment analysis, couched within Distributed Morphology (DM), which invokes an impoverishment operation to account for facts that would otherwise be problematic for containment hypotheses, including counterexamples to *ABA.
I first address the morphological relatedness of participles and preterites in Section 2, and then discuss issues with containment approaches within both DM and Nanosyntax in Section 3, where I also consider alternative analyses with an 'overlapping decomposition' (cf.Caha 2017;Bobaljik & Sauerland 2018).I then provide a revised containment analysis in Section 4, and offer further remarks about deriving alternations in the base through root suppletion or morphophonological rules, and about how this fits into the discussion about deriving *ABA.Section 5 concludes.

Relating participles to preterites.
There are several well-known reasons to suspect that English preterites and participles are morphologically related to each other.First, both use a suffix -ed, which is the default suffix for both the preterite and the participle.2Second, they share most of their suffix exponents, including -ed, but also -t and -∅ (e.g.Halle & Marantz 1993).Third, they often share alternations of their base, even when the suffixes themselves differ (e.g.freeze, froze-n, froze).Lastly, the forms of participles and preterites have exhibited leveling both historically and in contemporary use in various dialects (e.g.Bybee & Slobin 1982).Given the extensive sharing of forms between the two, it seems unlikely that this sharing is accidental.
In underspecification approaches like DM, the sharing of forms is often taken to reflect the sharing of morphosyntactic features.For Bobaljik's (2012) containment approach, one feature is shared between participles and preterites -I will refer to this feature as [PTCP] -and one feature is present on preterites but not participles -I will refer to this one as [PRET].In the containment structure in (3), the participle has no feature that the preterite does not also have; put differently, the preterite builds directly on the participle. (3) There is evidence that favors (3) over the alternative containment (with participles containing preterites), namely AAB patterns such as blow, blow-n, blew.If the order is switched, then the result produces ABA patterns, and this will run afoul of the logic of the Elsewhere Condition.The comparison between the two directions of containment is represented in Tables 3 and 4, which display the same verbs with opposite orderings of participles and preterites.Other analyses in the literature that acknowledge the morphological relatedness of English participles and preterites do not involve containment.In Zwicky 1990 and related work, participles and preterites do not share features.Instead, preterites -not participles -are basic, and a default rule refers to the form of the preterite to use for the participle.However, this treats the AAB identities (Table 3) between the bare and participle forms as coincidental; it is thus an accident that e.g.grow-n shares the base of the bare form grow.
For Halle & Marantz (1993), the preterite and participle share a feature [+PAST], but the two differ in the value of the feature [±PARTICIPLE]: this is feature overlap but not containment.This approach captures syncretisms but in principle, allows a base alternation of the participle (specified for [+PAST][+PARTICIPLE]) that does not extends to preterites, thus allowing ABA.
Turning our attention to the suffixes, we do find potential evidence suggesting participles are more marked than preterites.This evidence comes from morphological markedness, which is often taken to reflect featural markedness (though only tendentially).While -∅ is a suffix for both participles and preterites (Table 5), when only one is -∅ for a given verb, it is the preterite, not the participle (Table 6, which also shows the hypothetical but non-existent pattern).
This apparent markedness is the opposite of that assumed for (3), which takes the preterite to be more marked than the participle because of their containment relation.Given that the correlation between morphological and featural markedness is a tendency, this seems not to constitute strong evidence against (3).In the next section, I evaluate containment hypotheses by looking a alternations both in the bases and in the suffixes.
3. Containment approaches to *ABA.The containment representation in (3) along with the Elsewhere Condition generates AAA, AAB, ABB, and ABC -the last three shown in Table 7 but it does not generate ABA.While the containment analysis captures *ABA in the alternations of the base, it also makes predictions for the realization of the suffix.The default suffix for both participles and preterites is -ed; it should therefore be the least specified exponent.When the suffix for the participle is specified as irregular, this should also imply that the preterite suffix is irregular also.This happens, for example, when the suffix (including -∅) is the same for the participle and the preterite; this can be thought of as an ABB pattern of suffix exponence.other when both are irregular; this can be thought of as an ABC pattern of suffix exponence. (5 [PTCP] ↔ -ed (produces e.g.participle beat-en and preterite beat-∅) However, the account faces problems for a subset of participial -en verbs, all of which should correspond to irregular preterite forms.This is not borne out, as there are some verbs with participial -en that nevertheless take -ed in the preterite; see Table 8 for a partial list.These verbs would require a homophonous entry for -ed to block -en in preterite forms (6).This can be thought of as an ABA pattern for suffix exponence.
All of the exponents of the participial and preterite suffixes are shared, with the exception of -en, which is specific to the participle.This means that the preterite suffixes are a subset of the participle suffixes: the opposite of what we should expect.Thus while the alternations of the base point to one direction of derivation (preterites build on participles), the specificity of the -en suffix seems to point to the opposite direction of derivation.Most problematically, it is unexpected for a containment analysis that there should be an irregular, participle-specific suffix whose corresponding preterite suffix is default, given the logic of the Elsewhere Condition.In the next subsection, I consider a nanosyntactic account of containment, and demonstrate that it suffers from a distinct but related problem.
3.1 CONTAINMENT IN NANOSYNTAX.Nanosyntacticians have also adopted containment structures to account for *ABA (e.g.Caha 2009;Clercq & Wyngaerd 2017), using a combination of the Superset Principle and a correspondingly different version of the Elsewhere Condition.This type of account also allows AAB and ABB to be derived, while ABA is claimed not to be (see also Caha 2018 for discussion of syncretism in English preterites/participles from a nanosyntactic perspective).In brief, nanosyntactic accounts embrace insertion of exponents at non-terminals, and derive *ABA by making the most specific tree win.The lexical entries needed for ABB and AAB patterns are simplified in ( 7)-( 8).The simplest pattern of alternation in this framework is 'whole-word' syncretisms.These exist for both AAB and ABB (Table 9).Non-terminal insertion handles such cases without reference to zero-exponence, arguably an advantage over approaches that limit realization to terminals.
Regarding suffixes, it is not entirely clear to me how the framework best handles contextual allomorphy.If we assume -en is not problematic to derive, AAB patterns in which the participle is suffixed with -en and the preterite is -∅ are generated through cyclic override (cf.Caha 2018;72).See However, when we consider ABB patterns with participial -en (e.g.freeze, froze-n, froze), a problem arises.First, the alternation from the base to the participle, in conjunction with the suffix -en should still be realized with two separate exponents: the ablauted base and the participial suffix.A solution to this type of double-marking problem is pursued by Clercq & Wyngaerd (2017), who add additional lower structure to derive a similar alternation within the domain of adjectival grade.For participles and preterites, the needed additional XP structure would not be syntactically motivated.If we set this issue to the side, the derivation of the participle could be as in (10).Now considering the preterite, we would have to posit another lexical entry to override the realization of combination of XP and the participial suffix.This entry would have to be homophonous with the XP entry for the participle in order to derive the correct result.This problem is general to ABB verbs with -en, so many homophonous entries would have to be posited to derive ABB.These homophonous entries treat the base syncretism as accidental; moreover, allowing them readily permits ABA patterns.Without homophonous entries, the nanosyntactic account derives an incorrect suffixed result such as *spoke-d (11).The issues for the nanosyntactic account are thus as follows: in order to derive ABB participles with -en, we have to i) assume a syntactically unmotivated XP, ii) grant that the contextual allomorphy of the participle can be derived somehow, and iii) create homophonous entries for participial and preterite bases.Nanosyntactic accounts thus face challenges from both alternations in the base and the suffix allomorphy.5 Explaining these problems away would involve analyzing the combination of the base and -en as monomorphemic, with separate entries for the bare form, participle, and preterite.This treats the base identities involving irregular suffixes as accidental and abandons the generalization that -en is a suffix, despite the fact that it applies to dozens of verbs.
To conclude this section so far, the containment approaches in both DM and Nanosyntax face challenges: in DM, the issue is with how the Elsewhere Condition interacts with the exponence of the suffix -en, while in Nanosyntax, the issue is with how the base alternation must be expressed in cases of double-marking with -en in the ABB pattern.In the next subsection, I consider an alternative analysis to containment that is also capable of capturing *ABA.
3.2 OVERLAPPING DECOMPOSITION.The shared base and suffix morphology of participles and preterites indicates that the two overlap in their feature representation.If the two stand in a subset/superset relation, then the specificity of participial -en seems to suggest that the participle properly contains the representation of the preterite.However, as discussed above, this runs into problems with the AAB verbs (e.g.√ BLOW).If the only feature of the preterite is shared with the participle (such as [PAST]), there is no way of singling out the preterite to the exclusion of the participle in referring to its features.( 12) One possible solution to this problem involves overlapping decomposition, which has been discussed by Caha (2017) and Bobaljik & Sauerland (2018) as another way to capture *ABA.For this domain, bare verbs and their participles will share a feature, and participles and preterites will share a feature (13).We can assume that the feature on bare verbs and participles is inserted at PF; I will call it In a DM analysis, this decomposition reflects the specificity of participial -en, and the default realization for [PAST] can remain -ed.7 One benefit of ( 13) is that it reflects the generalization for morphological markedness mentioned in Section 2 concerning the distribution of -∅.
To derive the patterns of formal identity of the bases, both nonfinite and past features are referred to; the set of derivable syncretisms is in ( 14)-( 17).Note that AAB is unusual: either i) extrinsic ordering is needed (i.e.[NONFINITE] > [PAST]) or ii) the default realization is actually the base of the preterite rather than the bare form.The second option is illustrated in (17).8 ( 14) AAA (no contextual specification) One further auxiliary assumption is needed to rule out ABA with the overlapping decomposition.As noted by Caha (2017), an overlapping decomposition could technically yield ABA in a DM approach if the element that shares one feature in common with the others (in this case, the participle) is one allomorph, with the other realization being default (18).To rule this out, we could stipulate that the Vocabulary is constrained by a principle like ( 19 How does the nanosyntacic version of the overlapping decomposition fare?The same problem with ABB verbs arises that was discussed in the previous section (speak, spoke-n, spoke).9The syntax of bare, preterite, and participle forms would be as represented in (20).Among the approaches considered so far, the containment and overlapping decomposition approaches in DM fare best, while the containment and overlapping decomposition approaches in Nanosyntax do not perform as well.In the next section, I return to the containment analysis, which, with one added impoverishment rule, derives both the alternations in the base and in the suffixes, but goes further in accounting for counterexamples to *ABA in a constrained way.

4.
Towards an analysis: Containment with impoverishment.In this section, I show how a DM containment analysis can derive the facts through impoverishment (Bonet 1991;Nevins 2011; among many others), a postsyntactic operation that deletes morphosyntactic features.The feature decomposition for containment is repeated in (23).( 23) [PRET] V VERB I assume postsyntactic realization in the process called Vocabulary Insertion, and crucially, that impoverishment operations delete morphosyntactic features prior to Vocabulary Insertion (e.g.Embick & Noyer 2007).My proposed impoverishment rule deletes a diacritic class feature on a root (cf.Noyer 2005; Calabrese 2015): in this case, the diacritic feature [α] defines a declension class of all verbs that take the -en suffix.Recall that many of these verbs take -∅ in the preterite.This is reflected in the Vocabulary Items in (25), with the first two specified to be inserted in the context of [α].If impoverishment of [α] is triggered, the default -ed will be chosen instead.
The impoverishment rule will apply only to a subset of -en verbs: namely, only those that take -ed in the preterite.The impoverishment rule is given in (24).architecturally impossible for the exponence of roots to be affected by other exponents.
The difference in the conditions for MP and contextual allomorphy may lead us more generally to expect divergent patterns for containment structures.That is, if MP changes to a root's realization can be triggered by other exponents, but root suppletion cannot, then we expect ABA patterns to be possible with containment structures for the former but not the latter (except in the case of impoverishment).If the alternations of the base are driven by MP, the near absence of ABA patterns in English for the order BASE-PARTICIPLE-PRETERITE does not follow solely from containment and the Elsewhere Condition; the particular formulation of the MP rules as being sensitive to morphosyntactic features is also relevant.

Conclusion.
In this paper, I have evaluated several approaches to the syncretisms and alternations of English participles and preterites.Examining the patterns of both stems and suffixes led me to adopt a DM containment analysis supplemented by one impoverishment rule.
This paper is somewhat unusual in that it examines an ABA generalization without discussing typology, focusing instead on the morphology of a single language.The utility of this type of investigation is that it can be used to compare different theories of *ABA that make predictions beyond the syncretisms and alternations of one morpheme.Future research should carry out this type of investigation on other languages and in other domains in which ABA generalizations have been observed.
I have said very little about the relationship between syntax, semantics, and feature representation.For example, while it has been convenient to refer to the past participle as a monolithic syntactic entity, this is a gross oversimplification.I have also referred to features with labels such as PRETERITE without identifying their syntacticosemantic contributions.Abstracting over such details leaves many questions open.But I hope that in looking at how feature representations map to exponents, I have moved the discussion forward on how to relate participles to preterites. 4 the non-trivial issue of contextual allomorphy, -en could lexicalize [PAST [ NONFINITE ] ].We can derive AAB straightforwardly, such as give, give-n, gave, as shown in simplified form in (21).
ABB participles suffixed with -en, such as spoke-n.Here, we derive the incorrect form of the base (*speak-en) as in (22).If the lexical entry for spoke is enlarged to include a phrase with both PAST and NONFINITE, then we incorrectly derive spoke without a suffix as the participle.

Table 2 :
Table 2).Patterns of Syncretism for English Participles and Preterites

Table 7 :
Deriving Alternations and Syncretisms with Containment

Table 10 :
Table 10 for relevant examples, and (9) for a simplified derivation.-en/-∅ AAB Verbs ), stated informally here.)appealsto a similar principle to rule out AAB patterns in adjectival gradation.However, I cannot confirm that (19) is true in the general case.That being said, with this assumption in place, the overlapping decomposition analysis can work mechanically.However, the insertion of the feature [NONFINITE] does not seem well-motivated.Moreover, it would have to apply in various contexts, such as the present participle.Otherwise, the Vocabulary Items in (17) would produce the incorrect *took-ing, an unwelcome result.