Syntactic OCP Effects

Ad Neeleman & Hans van de Koot - UCL

1. The Repeated Morph Constraint

2. Four strategies

   1. Tolerance and avoidance

   2. Deletion

   3. Suppletion

3. Repeated features

   1. Syntactic identity

   2. Deletion

   3. Suppletion

   4. Portmanteau morphemes

   5. Tolerance and avoidance

4. Conclusion

5. References

1. The Repeated Morph Constraint

A number of authors have observed that languages exhibit a resistance against accidental repetition of morphemes (Perlmutter 1971, Stemberger 1981, Menn & MacWhinney 1983, Yip 1998, Plag 1998, and references cited there). This resistance is reminiscent of the Obligatory Contour Principle in phonology (Leben 1973, Goldsmith 1979, and much subsequent work), which requires adjacent phonemes to be contrastive. In this paper we will explore which phenomena at the sentence level can be understood in terms of the avoidance of repeated morphs, an issue first addressed within a generative context by Perlmutter (1971). This will require that we establish which strings can be taken to involve accidental repetition of morphemes and which strategies are available for avoiding them.

Suppose M₁ and M₂ are identical in the relevant sense and happen to be adjacent, as in (1a). Perhaps in the majority of cases, grammars simply live with this repetition and allow M₁and M₂ to surface unchanged. In cases where the string is ruled out, four possibilities arise. (i) If the grammar does not provide an alternative realization for either M₁ or M₂, structures in which these morphemes surface in adjacent positions are ungrammatical (the language is characterized by (1b)). (ii) Repetition of identical morphemes can be circumvented if the grammar allows haplology: deletion of one of the offending morphemes (the language is characterized by a combination of (1b) and (1c)). (iii) A further option is to associate either morpheme with a different realization, typically based on a subset or a superset of its features (the language is characterized by a combination of (1b) and (1d)). (iv) Finally, the grammar may make available a so-called ‘portmanteau’ morpheme, which spells out the features of both M₁ and M₂ (the language is characterized by a combination of (1b) and (1e)). Repair strategies always affect functional elements; repetition of lexical heads is tolerated.

Which of these strategies is exploited varies from language to language and from case to case. We should also note that there are other conceivable strategies, such as the insertion of a dummy morpheme between M₁ and M₂. Few, if any, cases of this type are attested. An anonymous reviewer suggests that there is a tendency in Spanish to introduce an expletive negation when the second term in a comparative construction begins with que (see (2a), where this tendency is marked by a question mark). This might be case in point, although insertion of expletive negation is not obligatory and can also appear without morpheme iteration (see (2b)). We will not investigate this issue any further here.

From the fact that many grammars are tolerant of repeated morphemes we can conclude that an inviolable condition blocking accidental repetition cannot exist. On the other hand, there seem to be clear cases in which structures are ruled out precisely because they show accidental repetition. This tension can be resolved in at least three ways.

First, one could develop a ‘conspiracy’ theory in which the rules of syntax and morphology are such that in specific cases accidental repetition happens to be avoided, even though the grammar does not contain any conditions ruling it out. At best, grammars that avoid accidental repetition could be favoured by an extra-grammatical, psycholinguistic, repeated morph constraint (see Menn & MacWhinney 1983 and Bonet 1995 for examples of this approach).

Second, one could assume that the grammar contains conditions that rule out fairly specific cases of accidental repetition and that trigger adjustments of the type in (1c) through (1e). These ‘low-level’ filters could also be seen as psycholinguistically motivated (see Radford 1977 for an approach in this spirit).

Third, one could adopt an optimality-theoretic approach to grammar (see Prince & Smolensky 1993). On this view, a maximally general condition against repeated morphemes would be part of the grammar, but its effects would depend on its interaction with other constraints. In other words, if a particular language allows structures that violate the constraint, these structures would be favoured by constraints ranked higher in the grammar of that language (see Grimshaw 1997 and Ackema 2001 for examples of such an approach).

In the sections that follow, we will state a number of descriptive generalizations involving the repeated morph constraint. These will be formulated in terms of the second approach for reasons of presentational convenience. No theoretical claims should be attached to this choice.

What constitutes an accidental repetition of morphemes depends on what we mean by identity of morphemes, ‘repetition’ and ‘accidental’. (i) Many authors assume that identity is to be understood in terms of phonological form, although it could be that similarity in form rather than strict identity may trigger haplology or suppletion. We will argue that, in addition to this, a sequence of morphemes expressing identical features may do so. (ii) The question about repetition concerns the context in which haplology and suppletion are triggered. The standard assumption is that M₁ and M₂ must be adjacent, which is compatible with all cases discussed here. (iii) Accidental repetition is to be distinguished from intentional repetition. In the case of reduplication in morphology, for example, two adjacent morphemes may end up having the same form by rule, thus exemplifying intentional repetition (the reduplicative morpheme is phonologically incomplete and takes its form, partially or wholly, from its host). Haplology and suppletion, however, involve accidental repetition: the relevant morphemes happen to have the same form and happen to surface next to each other. This raises the nontrivial question why grammars should distinguish the two cases. Menn & MacWhinney suggest a partial psycholinguistic answer, but a more comprehensive solution remains elusive.

In section 2 we discuss phenomena from several languages that illustrate strategies by which repetition of phonologically identical morphemes is avoided. In section 3 we turn to cases of haplology and suppletion involving morphemes that express identical features but are different in form. We offer some conclusions in section 4.

2. Four strategies

2.1. Tolerance and avoidance

It is easy to demonstrate that repetition of identical morphemes is not ruled out across the board. For example, in felicitous contexts, prepositional phrases may be used as ‘honorary NPs’. If so, they can be the complement of a preposition that is identical to their head, as shown by the Dutch example below:

Dutch also has a preposition that seems to combine freely with prepositional complements, even if these are not honorary NPs. It turns that the selected PP can be headed by the same preposition, voor. This is illustrated in (4a). Repetition of prepositions is not restricted to complementation. Dutch has verbs that are accompanied by a prepositional particle. If such a particle is stranded by verb movement, it can end up adjacent to the head of an extraposed PP, giving rise to examples like (4b,c).

However, in other cases repetition of phonologically identical morphemes leads to ungrammaticality. Thus, the Dutch demonstrative pronouns die and dat cannot be followed by a relative clause introduced by the same pronoun. It seems unlikely that this is due to syntactic or semantic factors, given that relative clauses introduced by die and dat may follow the non-homophonous demonstratives degene and hetgene, while the demonstratives die and dat allow modification by prepositional phrases with a comparable semantics. The full paradigm is given in (5) and (6). A similar point is made for German and English by Pullum & Zwicky, as cited in Menn & MacWhinney 1984).

The hypothesis that (5c) and (6c) are ruled out because of repetition of identical morphemes is corroborated by the grammaticality of relative clauses that modify the WH operators wie and wat:

Note that the ban on repetition of identical morphemes is conditioned by adjacency. No ungrammaticality results if the demonstrative is separated from the homophonous relative pronoun by a nominal complement.

Interestingly, the filter that excludes repetition must be sensitive to syntactic environment as well as phonological form, since the demonstratives dat and die may freely follow homophonous relative operators:

Indeed, in a carefully constructed example involving the complementizer dat, the demonstrative dat, the relative dat and an autoreferential DP, it is possible to create a sequence of six repeated morphemes:

The situation can be summed up by the following descriptive rule, where small capitals indicate syntactic elements, while material between forward slashes indicates their phonological realization:

2.2. Deletion

The strategy chosen in the Dutch examples above is one of avoidance of repetition. As Menn & MacWhinney point out an alternative strategy would be deletion. Thus, if two free adjacent morphemes normally receive the same phonological realization, one of them would fail to be spelled out. In Mandarin Chinese, for example, the perfective marker le and the particle le which marks a ‘new situation’ trigger deletion when adjacent (Yip 1998 and references cited there):

A similar rule of deletion is triggered in French by adjacency of comparative que and the homophonous complementizer (Radford 1977, 1979):

A further example from French involves the deletion of superlative le in a position immediately following the homophonous determiner (Radford 1977, 1979). (An anonymous reviewer suggests that (14b) may not involve haplology, but rather raising of the comparative expression plus beau, which lacks an article, to a prenominal position where it acquires superlative semantics. It would take us too far afield to explore this possibility here.)

The phenomenon is also found in Spanish clitic clusters. Grimshaw (1997) points out that at least in some dialects a sequence of a reflexive and an impersonal clitic, expected to surface as se se, in fact surfaces as a single clitic. Although the empirical claim is somewhat controversial, it would – if correct – show that haplology extends to clitic clusters.

We will discuss one notorious case of syntactic haplology in some more detail. The Dutch weak pronoun er can have four different functions. It can be an expletive, the fronted argument of a preposition, a locative meaning ‘there’ and a fronted noun associated with a numeral or a quantificational modifier. We illustrate these uses of er below, where we adopt Odijk’s (1993) marking of er’s functions by X, P, L and Q, respectively. (See Bech 1952, Van Riemsdijk 1978, Bennis 1980, 1986 and Model 1991 for additional discussion.)

Since weak pronouns are not generally in complementary distribution in Dutch, one would expect sentences with up to four occurrences of er to be possible. In fact, however, in most sentences of the relevant type only a single er is spelled out, which somehow manages to combine multiple roles. The following examples, taken from Odijk (1993), illustrate sentences in which er combines two functions:

There are in principle two approaches to these data, both based on the fact that weak pronouns tend to cluster in a position between the regular subject position and the VP. A syntactic approach could postulate a single position for er in the cluster, which could then bind several empty pronominal elements in the clause. The various functions of er would then be associated with these elements. The alternative, haplological, approach assumes that the cluster can contain multiple occurrences of er, each with a single function. However, adjacent occurrences of er are subject to deletion, giving rise to the appearance of shared functions in the remaining pronoun.

The first approach cannot be correct as it stands, since it does not allow any multiple occurrences of er, whether adjacent or not. This is clearly too strong, given that quantificational er can cooccur with expletive, locative or prepositional er (although other co-occurrences are excluded):

The second, haplological, approach is advocated in Bennis (1980, 1986). The main advantage of this proposal is that it allows multiple occurrences of er as long as they are nonadjacent. However, a deletion rule is by itself not sufficient to accommodate all the data. In particular, a syntactic analysis seems required to explain the ungrammaticality of any structure that combines locative and preposition er. This is illustrated in (20a), which tries to spell out er_L and er_P separately, and (20b), which tries to combine the two functions in a single pronoun.

As argued by Odijk (1993) on the basis of largely comparable data, a hybrid analysis of er is required, which combines syntactic restrictions with haplology. Here we will briefly present a variant of his proposal.

Data like (18) and (19) show that the position occupied by er_Q in the pronominal cluster must be distinguished from the one occupied by er_X, er_L and er_P. In particular, it seems to be located to the right of this more general position. This conclusion makes it necessary to adopt a rule of deletion triggered by adjacent occurrences of er. The point is that, whenever the two positions are filled in the absence of intervening material, only a single er surfaces:

The co-occurrence restriction on er_L and er_P can be understood if these compete for a single position. But why does this explanation not affect the expletive function? In other words, why can a single er realize er_X, as well as er_L or er_P, despite the fact that separate realization of er_X and er_L or er_P is prohibited? The ungrammaticality of (22a) and (22c) demonstrates that haplology cannot be involved in the explanation of the grammatical examples.

We suggest that the key to understanding the distribution of expletive er lies in an interpretive rule which states that an occurrence of er is assigned the expletive function if it occurs in the IP or CP domain (we remain neutral on what this function exactly entails). Thus, er is never inherently expletive, but acquires this property on the basis of its context. This explains why an expletive function is unavailable in (23a), where er can only be locative. By contrast, (23b) is ruled out because er must be exclusively locative (given the choice of subject), while the interpretive rule forces an additional expletive function.

On this view, the lexicon contains two entries for er, quantificational er_Q and a referential variant that subsumes er_P and er_L. That this is a reasonable assumption is supported by the fact that er_P and er_L have a strong counterpart, daar, but er_Q does not:

Pure expletive er is merged in a position to which the interpretational rule applies. However, since this merged expletive is not related to an interpretable empty position, its semantic content must be suppressed, much like the semantics of do in do-support constructions. The data in (22) now fall into place. In general, suppression of semantic content is avoided where possible. Thus, in constructions that allow verb movement, do-support is ruled out. Similarly, where a variant of er is available that allows movement into a position to which the interpretational rule applies, merger of er is blocked, since the movement option avoids suppression of semantics. In the resulting structure er fulfills both the expletive function and the function it is associated with in its base position.

There is one remaining issue. As it turns out, the expletive and quantificational functions of er are incompatible:

This restriction may be related to the fact that cross-linguistically expletives are derived from referential rather than quantificational elements. Apparently, quantificational semantics is incompatible with the expletive function.

In summary, the distribution of Dutch er can be captured by the following generalizations:

The generalizations in (26a) and (26c,c’) are directly taken from Odijk (1993), while (26b), (26d) and (26e) provide an alternative to his analysis of expletive er, which circumvents some technical difficulties. For our present purposes, it is important to acknowledge the properties of the rule of haplology in (26c,c’). As the earlier rule in (10), it is conditioned by adjacency and must refer, not only to phonological identity, but also to syntactic category. The latter is easily demonstrated by the examples in (27), in which no deletion takes place, despite the adjacency of phonologically identical morphemes er:

Another property shared with the earlier rule is that the deletion applies to specific lexical items only. Identical weak pronouns are generally allowed to surface in adjacent positions:

2.3. Suppletion

A fourth strategy in dealing with cases of accidental repetition is to map one of the morphemes to a phonological form that can normally not be used to realize it. Menn & MacWhinney (1983) list a number of cases of this type that suggest the phenomenon is particularly common in clitic clusters. For instance, a sequence of an indirect and a direct object third person proclitic in Spanish is expected to surface as le lo but does in fact take the form of se lo. The example below is taken from Bonet 1995 (see also Perlmutter 1971):

Notice that, if these cases are to be conditioned by adjacency of identical forms, identity must be defined in terms of onset identity, a considerably looser notion than the strict identity assumed for the cases discussed so far. An alternative account, explored in the next section, could rely on le and lo sharing the phi-features 3rd person singular.

The phenomenon is not restricted to clitic clusters, however. Radford (1977, 1979) notices that in German a sequence of comparative als and temporal als is realized as denn als.

In many cases the alternative realization of the affected morpheme cannot reasonably be derived from the expected phonological form. Rather, it is an independently existing word in the language with a feature make-up that is either a subset or a superset of that expressed by the expected form. Below we will discuss two examples of suppletion in a little more detail. The first involves non-realization of a feature, while the second is a case in which the phonological form used spells out more features than are motivated by the syntax/semantics.

Ackema (2001) observes that the Dutch coordinating complementizer of ‘or’ and the homophonous complementizer used to introduce embedded questions cannot surface adjacently. This is illustrated by the case of coordinate reduction in (31b). As shown by (31c) and (31d), the strategy employed in this case of accidental repetition is not haplology but suppletion. The second complementizer surfaces as indicative dat ‘that’. ((31b) is also acceptable for some speakers.)

Ackema argues that these data fall out from a ban on adjacent identical complementizers (or rather functional heads). For our present purposes we can formulate the relevant filter as in (32).

In addition, the grammar of Dutch is such that violation of this principle is avoided by a rule that suppresses realization of the Q feature in the second complementizer. The result of this is that only the feature indicating subordination will be realized, yielding the phonological form dat:

Several aspects of this analysis are independently motivated. First, the adjacency condition is corroborated by the fact that material stranded by coordination reduction blocks suppletion (see (34a)). Second, the relevance of the syntactic category of the adjacent morphemes can be inferred from the grammaticality of (34b), in which the second of functions as an honorary NP (it has auto-referential semantics). Third, phonological identity is a precondition for suppletion: if the coordinating complementizer is focussed, suppletion cannot take place (see (34c)).

The data in (35) provide an interesting complication not discussed by Ackema. Although these examples have been claimed to be ungrammatical by Koster (1978), many speakers of Dutch find them acceptable and, moreover, do not distinguish between the cases of accidental repetition in (35a,b) and the cases of non-repetition in (35c,d).

(35a) can easily be accommodated as a case of tolerance (see section 2.1), since dat cannot be input to the rule in (33). (35b), however, is predicted to be ungrammatical. It should be realized as (36), which is ungrammatical on the intended reading.

A potential solution would be to say that suppression of features may not obliterate semantic contrasts. The subject position is not selected and can therefore either contain an indicative or interrogative CP, which are of course associated with different readings (‘whether ...’ versus ‘the fact that ...’). Ackema’s data involve selected interrogatives, which apparently reconciles suppression of the interrogative feature with recoverability. That selection does not do so generally raises questions about the nature of recoverability, which, however, are beyond the scope of this paper.

Ackema derives various apparently stipulative properties of (32) and (33) from optimality-theoretic competition. In this respect, the paper builds on Grimshaw 1997, which provides an account of suppletion in certain Italian clitic clusters. Bonet (1995) observes that a sequence of an impersonal and a reflexive clitic, expected to surface as si si or si se, is in fact realized as ci si, where ci is a form that normally functions as the first person plural pronoun:

It seems, then, that Italian has a ban on adjacent identical clitics:

The same filter could be responsible for the fact that the dative clitics lo and la become glie before accusative clitics beginning with l. Of course, this would require that identity is defined as onset identity.

As in the case of dative clitics, the strategy adopted for avoiding violation of (38) by si si sequences is one of suppletion. What distinguishes replacement of si by ci in Italian from replacement of of by dat in Dutch is that the suppletive form in Italian is richer in content than the form it replaces. Bonet (1995) argues that the impersonal clitic si is highly underspecified: it only contains the feature plural, as indicated by the fact that it triggers plural adjectival agreement (Cinque 1988). If this analysis is correct, whatever form replaces si will be associated with a more specific feature content. In particular, Bonet captures the situation in Italian by a rule that inserts the feature first person, prior to phonological realization of the clitic. The resulting form is ci, which normally expresses first person plural:

Three remarks are in order.

First, the first person feature in ci should not feed semantic interpretation. This leaves open two options: either the rule is syntactic and relies on subsequent suppression of the relevant feature at LF, or the rule applies in the PF branch of the grammar (as suggested by Bonet).

Second, it remains to be explained why the suppletive form is ci rather than any other of the available clitics. Bonet gives a partial answer to this question, based on economy. Grimshaw (1997) discusses the problem in detail, providing an optimality-theoretic analysis. Ci is chosen because it is minimally marked, while at the same time realizing as much of the target semantics as possible.

Third, it is an open question why the first rather than the second si receives a suppletive realization. Although this is expressed by Bonet’s rule, it remains unexplained.

3. Repeated features

3.1. Syntactic identity

The rules we have formulated so far must operate at the PF interface: while their effects are phonological, the environment in which they are triggered is characterized in both morpho-syntactic and phonological terms. For example, Dutch does not exclude the sequence /er/ /er/ in general, but only if it corresponds to a repetition of the pronoun er in syntax. Thus, the sequence /duiker/ /er/ in (27) is well-formed.

This state of affairs raises the question what exactly triggers haplology in this case: repetition of phonological forms or of syntactic features. In some cases phonology seems the only possible trigger (for example, the case of Dutch complementizer suppletion), but in others the situation is arguably less clear. A complicating factor is that elements often have the same form because they express the same features.

Given that the spell-out rules relate morpho-syntactic objects to their phonological form, one would expect to find cases in which deletion or suppletion is triggered by syntactic features, even though the morphemes affected are not phonologically identical in isolation. Thus, even if M₁ and M₂ in (40a) are usually spelled out differently, the fact that they both express the feature F may trigger avoidance (40b), haplology (40c), suppletion (40d) or insertion of a portmanteau form (40e).

In this section we will discuss some phenomena that can be analyzed in these terms. Of course, the existence of such cases does not imply that the cases discussed so far must be reanalyzed in terms of repetition of syntactic features.

3.2. Deletion

A suitable phenomenon for introducing the issue is provided by the deletion of possessive markers in Romanian in the context of preceding definite determiners (see Dobrovie-Sorin 1987, Grosu 1988, Corlinescu 1995 and others). Indefinite articles in Romanian precede the noun, while definite articles follow it. This implies that definite, but not indefinite, articles will be adjacent to the particle that introduces post-nominal possessives. As it turns out, the possessive marker is deleted if it surfaces next to the definite article, as in (41b). No deletion occurs if the article and possessive marker are separated by an adjectival modifier (cf. 41c) or a demonstrative (cf. 41d).

Although this phenomenon is often analyzed in purely syntactic terms, Ortmann and Popescu (2001) argue convincingly that it is in fact conditioned by adjacency rather than hierarchical structure. A particularly strong argument to this effect is provided by coordinated possessives, which, in the relevant context, require deletion of the possessive marker that introduces the leftmost conjunct but not of any other possessive markers. This is particularly hard to understand in terms of syntactic structure.

Ortmann and Popescu analyze possessive deletion in terms of phonologically conditioned haplology. Indeed, as they point out, the definite determiner and the possessive share at least one phoneme in contexts that trigger deletion. In Romanian, as in French, possessives agree in number and gender with the nominal constituent they modify. As a result, a noun ending in a masculine singular definite determiner will take a possessive phrase starting with a particle expressing the same features, etc. In other words, the contexts that require deletion of the possessive marker are presented in the table below:

The deletion of the possessive marker cannot have a purely phonological trigger, since indefinites whose stem ends in ul do not require deletion of a following al possessive. Ortmann and Popescu therefore propose a ban on any shared phonological material in adjacent functional heads.

Although the determiner and possessive forms consistently share a phoneme, they are clearly not generally phonologically identical and consequently it is not immediately clear that we can generalize from the cases discussed in the previous section to Romanian possessive deletion. Indeed, given that the determiner and possessive marker share phi-features, it seems more plausible to attribute the deletion to a ban on accidentally repeated syntactic features:

This ban triggers the rule below:

In other words, Romanian possessive deletion takes place under systematic syntactic identity but only partial phonological identity, which favours an account in terms of repeated syntactic features.

As Ortmann and Popescu’s work demonstrates, there is a reasonable interpretation of the Romanian data based on accidental phonological repetition. A case in which such an approach seems much less plausible, if not downright impossible, is presented by Old French pro drop. The facts are as follows. Old French is a verb second language and pro drop is sensitive to word order. It is possible in exactly one context, namely in main clauses in which subject-verb inversion would occur if the subject were overt, as pointed out by Foulet (1928). Thus, it is possible in declarative main clauses in which a constituent other than the subject is fronted and in yes/no questions, but it is unattested in embedded clauses and subject-initial declaratives. This is illustrated by the data in (46a,b), from Adams 1987. (46c,d) are constructed examples of unattested patterns.

The simplest generalization that describes these data is that a weak pronoun is omitted precisely when it is adjacent to the inflectional ending it agrees with (for obvious reasons focussed pronouns cannot be deleted). This suggests an account in terms of accidentally repeated syntactic features. The relevant filter is given in (47).

Violation of this filter is avoided by deletion of the pronoun:

This is a reformulation of the analysis offered in Ackema & Neeleman 2003, who characterize the domain of rule application somewhat differently. Crucially, the conditioning environment cannot be purely phonological, given that pronouns and inflectional endings in Old French do not show any similarity in form. Hence, if haplology is at stake here, it is best captured in terms of repeated syntactic features. (As shown by Ackema & Neeleman, the argument carries over to Arabic pro drop.)

The main earlier treatment of Old French pro drop (Adams 1987) is based on government and relies on the following assumptions. (i) The position of pro must be identified by a governing head. (ii) The content of pro must be identified by coindexation with the proper features. (iii) The proper governor for pro is I⁰ but not C⁰. (iv) I⁰ governs to the right. (v) I⁰ can govern from its derived position in C⁰ in verb-second structures. (vi) pro cannot be fronted to satisfy the verb second constraint. Several of these conditions are unexpected, even in GB theory. More importantly, however, they amount to a rather complex way of deriving what is in essence a simple generalization, namely that pro drop relies on adjacency of an agreeing affix.

3.3. Suppletion

Having seen a possible case of deletion in order to avoid accidental repetition of syntactic features, one may expect the same avoidance mechanism to give rise to suppletion. A possible case of this is found Dutch verb-second structures. In general agreement in Dutch is insensitive to word order, but second person singular verb forms surface as first person singular whenever verb second gives rise to inversion:

As in the case of Old French pro drop, the obvious generalization is that spell-out is sensitive to adjacency of the second person singular affix and the agreeing subject. Ackema & Neeleman (2003) develop an analysis roughly along these lines, based on a characterization of the Dutch verbal paradigm in terms of the features PRT, ADD and PL, which represent participant in the speech act, addressee and plural (see also Kerstens 1993, Harley & Ritter 1998). The paradigm may then be characterized in terms of the following spell-out rules:

Given this verbal paradigm, the condition that rules out (49c) can be stated in terms of accidental repetition of the ADD feature:

Violation of this filter is avoided by suppression of the ADD in spell-out:

The resulting phonology only realizes the PRT feature and therefore has the form of the first person singular. The analysis is supported by the fact that, although there is in general no adjacency requirement on a subject and a preceding fronted verb, the reduced form of the second person singular must immediately be followed by the subject. (See Paardekoper 1961 and E. Hoekstra 1996 for related discussion and observations. We should acknowledge here that there is some variation among speakers with respect to the relevant judgements. Some speakers reject (53b), where the subject contains a DP-internal modifier, but no speaker accepts (53b’).):

Indeed, if the fronted verb and the following second person singular subject are not adjacent, the preferred form of the verb ends in t:

The example is not perfect, which Ackema & Neeleman attribute to a parsing difficulty.

Although we have presented the loss of second person marking under inversion as a possible case of suppletion, it could also be analyzed in terms of deletion, given that the suppletive first person singular form does not carry an ending. However, in realizational models of morphology deletion corresponds to non-spell-out of all features. Hence, if the syntactic agreement affix were deleted, the resulting phonological form would be the elsewhere form, which ends in t. This is obviously not the correct result.

Be this as it may, there are comparable phenomena in other languages that must involve suppletion rather than deletion, because there is an overt form that unambiguously spells out the features unaffected by suppression. A well-known case is the agreement alternation with preverbal and postverbal subjects in Modern Standard Arabic (Fassi Fehri 1993, Benmamoun 1997, 2000, Ackema & Neeleman 2003). As the data below show, plural agreement is suppressed in VSO order, but gender agreement is maintained. Only in VSO order is the subject adjacent to the verbal ending (data from Fassi Fehri):

Of course, both the Dutch and the Modern Standard Arabic data have received syntactic analyses (see Zwart 1997 and references mentioned there for a syntactic approach to the Dutch data and Huybregts 1991 and Fassi Fehri 1993 for a syntactic approach to agreement weakening in Arabic). It would take us to far afield to compare these to the suggestions made here.

3.4. Portmanteau morphemes

In the case of haplology based on phonological identity one can either say that one of the two morphemes fails to be realized phonologically (deletion) or that the remaining phonological form spells out two morphemes (coalescence). It is not easy to find empirical arguments distinguishing the two approaches and it may even be the case that different instances of phonologically motivated haplology require different analyses (see De Lacy 1998 for related discussion in morpho-phonology).

If we turn to avoidance of accidental repetition of syntactic features, however, the two strategies can be told apart. In the case of deletion, the remaining phonological form will be identical to the spell-out one of the two morphemes receives in isolation. In the case of coalescence, it may well be that the form that surfaces differs from the form of either morpheme in isolation. In other words, avoidance of accidental repetition of syntactic features may give rise to a portmanteau morpheme.

The case we will look at involves Huastec weak pronouns, as discussed Edmonson 1987. Huastec is an absolutive-ergative language with a separate paradigm for absolutive and ergative pronominals. This language displays the following phenomenon. In intransitive structures, pronouns show up in the absolutive (cf. 56a). In transitive structures, a third person singular object pronoun may be omitted, in which case a subject pronoun shows up in the ergative form (cf. 56b). However, when both pronouns should be overtly realized, we do not find the expected repetition of pronouns, but rather a portmanteau morpheme indexing both the ergative and absolutive arguments (cf. 56c,d).

The analysis of the Huastec data could involve a filter ruling out adjacent weak pronouns. Violation of this filter is avoided by assigning the two morphemes a single phonological realization:

It is not immediately clear from the Huastec data that what Edmonson analyzes as weak pronouns could not instead be agreement affixes, given that they tend to surface adjacent to the verb. However, that we are dealing with a syntactic portmanteau morpheme is supported by two facts. First, pronouns can be separated from the verb by certain modifiers (cf. (58)). Second, in sentences containing full DP arguments, the pronominals can be omitted (although doubling is also allowed).

3.5. Tolerance and Avoidance

Before we summarize the main findings of this paper, let us briefly discuss the two remaining strategies of dealing with accidental repetition of syntactic features, namely tolerance and avoidance. It will be clear that tolerance is by far the most common strategy. Agreement, for example, is not normally affected by adjacency. As regards avoidance, we have not found any examples of this. In cases where repetition is not tolerated, languages consistently seem to provide an alternative realization. We have no explanation for this, but speculate that functional factors may disfavour avoidance strategies in such cases.

4. Conclusion

Strategies for avoidance of accidental repetition involving free morphemes can invariably be formulated as filters that mention both phonological and syntactic properties of the relevant structure. The minimal phonological condition is adjacency, but often partial or complete identity of form plays a role as well. The minimal syntactic condition specifies the syntactic category of the elements in question, but often shared features are mentioned as well. Given these characteristics, the conclusion is unavoidable that avoidance strategies form part of the PF interface. Moreover, this interface must provide a mapping from morpho-syntactic to morpho-phonological representations. It cannot be the case that the phonological form of syntactic terminals is jointly merged with their syntactic and semantic properties.

It is highly unpredictable in which cases avoidance strategies are employed and, where they are employed, which strategy is chosen. The fact that avoidance is language-specific, construction-specific and morpheme-specific suggests to us that it must be seen as part of the lexicon of the language. In the view of PF sketched above, it is natural to view the lexicon as a list of minimal mapping rules associating, among other things, syntactic terminals with phonological forms (see Jackendoff 1997, Halle and Marantz 1993, and many others). If so, some lexical mapping rules may be context-sensitive, giving rise to avoidance of accidental repetition in specific cases.

For most cases of haplology involving free forms a purely syntactic analysis is unsatisfactory, primarily because it cannot express the generalizations that seem to underlie the data. It cannot but treat the relevance of adjacency as an epiphenomenon.

References

Ackema, P. 2001. Colliding complementizers in Dutch: Another OCP effect. Linguistic Inquiry 32, 717-727.

Ackema, P. & A. Neeleman 2003. Context-sensitive spell-out and adjacency. Natural Language and Linguistic Theory 21, 681-735.

Adams, M. 1987. From Old French to the theory of pro drop. Natural Language & Linguistic Theory 5, 1-31.

Bech, G. 1952. Über das niederländische Adverbialpronomen er. In Traveaux du Cercle Linguistique de Copenhague 8, 5-32. Copenhague/Amsterdam. (also appeared in Hoogteijling, 1969, 147-174).

Benmamoun, E. 1996. Agreement asymmetries and the PF interface. SOAS working papers in linguistics 6, 106-128.

Benmamoun, E. 2000. The feature structure of functional categories. Oxford: Oxford University Press.

Bennis, H. 1980. Er-deletion in modular grammar. In S. Daalder and M. Gerritsen, eds., Linguistics in the Netherlands 1980, 58-68. Amsterdam: North-Holland.

Bennis, H. 1986. Gaps and dummies. Dordrecht: Foris.

Bonet, E. 1995. Feature structure of Romance clitics. Natural Language and Linguistics Theory 13, 607-647.

Cinque, G. 1988. On si constructions and the theory of arb. Linguistic Inquiry 19, 521-582.

Corlinescu, A. 1995. Romanian genetive constructions. In G. Cinque & G. Guisti, eds., Advances in Romanian Linguistics, 1-54. Amsterdam: John Benjamins.

Dobrovie-Sorin, C. 1987. A propos de la structure du groupe nominal en roumain. Rivista di Grammatica Generativa 12, 123-152.

Dobrovie-Sorin, C. 1999. Spec, DP and (in)definiteness spread: From Romanian genetives to Hebrew construct state nominals. To appear in V. Motapayane, ed., Comparative Studies in Romanian Syntax. The Hague: Holland Academic Graphics.

Edmonson, B. 1987. Huastec pronouns. Anthropological Linguistics 29, 342-361.

Fassi Fehri, A. 1993. Issues in the structure of Arabic clauses and words. Dordrecht: Kluwer.

Foulet, L. 1928. Petite syntaxe de l’ancien français. Paris: Champion.

Goldsmith, J. 1979. Autosegmental phonology. New York: Garland.

Grimshaw, J. 1997. The best clitic: Constraint conflict in morphosyntax. In L. Haegeman, ed., Elements of grammar. Kluwer Academic Publishers.

Grosu, A. 1988. On the distribution of genetive phrases in Romanian. Linguistics 26, 931-949.

Halle, M. and A. Marantz. 1993. Distributed morphology and the pieces of inflection. In The view from building 20, eds. K. Hale and S. Keyser, 111-176. Cambridge, Mass.: MIT Press.

Harley, H. & A. Ritter 1998. Meaning in morphology: Motivating a feature-geometric analysis of person and number. Ms. University of Pennsylvania and University of Calgary.

Hoekstra, E. 1996. On the interaction between agreement and pronouns in the 2sg present tense in Dutch. Ms. P.J. Meertens Institute. Amsterdam.

Huybregts, R. 1991. Allosteric agreement in VSO languages. In F. Drijkoningen & A. van Kemenade, eds, Linguistics in the Netherlands 1991, 81-90. Amsterdam: John Benjamins.

Jackendoff, R. 1997. The architecture of the language faculty. Cambridge: MIT Press.

Kerstens, J. 1993. The syntax of number, person and gender. Berlin: Mouton de Gruyter.

Koster, J. 1978. Why subject sentences don’t exist. In S. Keyser, ed., Recent transformational studies in European languages. Cambridge: MIT Press.

Leben, W. 1973. Suprasegmental phonology. PhD dissertation. MIT.

Menn, L. & B. MacWhinney 1983. The repeated morph constraint: Toward an explanation. Language 60, 519-541.

Model, J. 1991. Gramatische analyse: Syntactische verschijnselen in het Nederlands en het Engels. Dordrecht: ICG Publications.

Odijk, J. 1993. Compositionality and syntactic generalizations. PhD dissertation. Catholic University of Brabant (KUB).

Ortmann A. & A. Popescu 2001. Haplology involving morphologically bound and free elements: Evidence from Romanian. In J. van Marle and G. Booij, eds., Yearbook of morphology 2000, 43-70. Dordrecht: Kluwer.

Paardekoper, P. 1961. Persoonsvorm en voegwoord. De nieuwe taalgids 54, 296-301.

Perlmutter, D. 1971. Deep and surface structure constraints in syntax. Holt, Reinhart and Winston.

Prince, A. & P. Smolensky 1993. Optimality theory. Ms. Rutgers University and University of Colorado at Boulder.

Radford, A. 1977. Counter-filtering rules. York papers in linguistics 7, 7-45.

Radford, A. 1979. The functional basis of transformations. Transactions of the Philological Society.

Stemberger, J. P. 1981. Morphological haplology. Language 57, 791-817.

Van Riemsdijk, H. 1978. A case study in syntactic markedness. Dordrecht: Foris.

Yip, M. 1998. Identity avoidance in phonology and morphology. In S. LaPointe, D. Brentari & P. Farrell, eds., Morphology and its relation to phonology and syntax, 216-246. Stanford: CSLI Publications.

Zwart, C. J.-W. 1997. The morphosyntax of verb movement. Dordrecht: Kluwer.

6 September 2004