Two clauses that are in a syntactic relation may remain isolated or may merge to different degrees. The fusion, or interlacing (Lehmann 1988), may take the following forms:
- omission of material from one clause under identity with material of the other clause: coordination reduction, backward coordination reduction, gapping
- dependence of grammatical categories of one clause on corresponding categories of the other clause: empty place formation, control, switch-reference, cosubordination, consecutio temporum
- syntagmatic interweaving of the clauses: prolepsis, raising, clitic climbing.
Omission under identity
As said elsewhere, coordination is an operation that may apply at all syntactic levels from the sentence down to the morpheme. We are here primarily concerned with relations between clauses. However, since omission (ellipsis) of material may produce expressions that actually or apparently belong to lower levels, lower-level coordination will be mentioned, too.
Starting from the coordination of clauses, an initial observation is that material (i.e. expressions that may have lexical content) of one clause may be omitted under identity with material of the other clause. This is illustrated by – (from Kempen 1991:357).
|.||John stole a bike and ... sold it immediately.|
|.||John stole a bike and Peter ... a car.|
|.||John stole ... and Peter bought a bike.|
The three dots symbolize an omitted expression, underlining marks the expression that is understood as missing. The three examples illustrate three forms of omission in coordinative constructions which have received different names in transformational accounts:
- In , the subject is omitted from the second clause and understood to be identical to the first clause's subject. This is coordination reduction (also called “conjunction reduction”).
- In , the verb is omitted from the second clause and understood to be identical to the first clause's verb. This is called gapping.
- In , the direct object is omitted from the first clause and understood to be identical to the second clause's object. This is called backward coordination reduction.
If we assume some kind of zero anaphora, then that works rightwards in and , but leftwards in . However, it is not certain that such a process is at work in all of these cases. We will examine each in turn.
In constructions like , what remains from the elliptic clause is a syntagma, viz. its verb phrase. Ignoring the three dots as an artifact of a certain theoretical approach, we can see that two verb phrases are being coordinated, the structure then being [ John [[stole a bike]VP and [sold it immediately]VP]VP]. In other words, instead of assuming a reduction operation and zero anaphora, one might as well assume an expansion of a basic coordination of verb phrases into a coordination of clauses.
Coordination of syntagmas of lower levels may always be expanded into coordination of clauses which are copies of each other except that where one contains one of the coordinate constituents, the other contains the other one. – show this paradigmatic relationship for coordination of noun phrases in subject and direct object function, of adverbials and of verbs.
|.||a.||Linda arrived and Irvin arrived.|
|b.||Linda and Irvin arrived.|
|.||a.||Linda drank a beer and Linda drank a coffee.|
|b.||Linda drank a beer and a coffee.|
|.||a.||Linda will come today and Linda will come tomorrow.|
|b.||Linda will come today and tomorrow.|
|.||a.||Linda printed the dissertation and Linda copied the dissertation.|
|b.||Linda printed and copied the dissertation.|
As usual, this syntactic paradigm may be described by transformations which either reduce a set of clauses to one clause containing a coordinate phrase or else expand a clause containing a coordinate phrase into a complex sentence consisting of parallel coordinate clauses.
Such data do not warrant an analysis that considers lower-level coordination as a derived phenomenon based on coordination of clauses plus coordination reduction. Instead, the #a and #b versions of - are to be regarded as expansions and condensations of each other.
Things are different for . There the remainder of the reduced clause is Peter a car, which is not a syntagma. Consequently, the alternative of analyzing the construction as the coordination of two lower-level syntagmas is not available. Therefore the construction is analyzed as a coordination of two clauses, with the formation of a gap in that position of the second clause which would be occupied by the verb.
Pairs such as those of – illustrate the paradigmatic relationship between the coordination of two full clauses and the construction with verb ellipsis in the second conjunct.
|.||a.||Linda had a beer and Irvin had a coffee.|
|b.||Linda had a beer and Irvin a coffee.|
|.||a.||Linda had more beers than Irvin had coffees.|
|b.||Linda had more beers than Irvin coffees.|
|.||a.||Linda cooked ravioli today and Linda cooked pumkin soup day before yesterday.|
|b.||Linda cooked ravioli today and pumkin soup day before yesterday.|
In all such cases, a reduction transformation converting the #a versions into the #b versions is a straightforward descriptive solution. The transformation that applies to a series of syntactically parallel clauses containing the same verb and which drops this verb in all but one clauses is called gapping. Depending on whether word order in the language has the verb at the beginning or at the end of the clause, gapping works “forward”, i.e. deletes subsequent copies of the verb, or “backwards”, i.e. deletes preceding copies. Forward gapping works like zero anaphora and therefore seems to be the default (Gaeta & Luraghi 2001).
Backward coordination reduction
is a coordination of two clauses which only share their direct object while the subjects and verbs are different. Here, the object is omitted from the first clause. Again, what remains from this clause is not a syntagma, so that no lower-level coordination (something like [[John stole] and [Peter bought]]) is possible. Instead, something like zero cataphora appears to be at work here. Observe that this construction works the better the more obligatory the direct object of the first verb is, since the gap becomes all the more noticeable.
Forward conjunction reduction of the direct object, as in , is ungrammatical in English and some other languages:
|.'||John stole a bike and Peter bought ....|
Again, in a verb-last language such as Turkish or Japanese, this construction would be entirely unproblematic.
Interdependence of categories
Zero anaphora and empty-place formation
If a referent is kept constant over a series of predications, rules of anaphora apply. These are least constrained at the text level. The explicitness of resumption of a referent essentially depends on the freshness of its status in the awareness of the interlocutors (s. Lambrecht 1994). At lower levels and following the Penthouse Principle, such considerations are increasingly replaced by rules of grammar which demand less explicit anaphors.
For example, an identical referent appearing in non-first coordinate clauses such as .a and .a may be represented by an anaphoric pronoun, as in the #b versions, or may be omitted altogether, as in the c-versions. The latter case is called ‘zero anaphora’.
|.||a.||Linda ate a doughnut and Linda drank a coffee.|
|b.||Linda ate a doughnut and she drank a coffee.|
|c.||Linda ate a doughnut and drank a coffee.|
|.||a.||Linda cheated Irvin and Jill abused Irvin.|
|b.||Linda cheated Irvin and Jill abused him.|
|c.||Linda cheated and Jill abused Irvin.|
As may be seen, zero anaphora yields results that are similar to coordination reduction and gapping.
Zero anaphora plays an important role in subordinate clauses, where it contributes to their desententialization. In English, the complement of want may be a non-finite clause, as in . If the subject of the subordinate clause differs from the main clause subject, then it is expressed, as in .a. If they are the same, then the subject of the infinitive must not be expressed, as in .b.
|.||a.||Linda wants [Irvin to play tennis].|
|b.||Linda wants [to play tennis].|
This process was called “Equi-NP Deletion” in the times of transformationalism. However, the operation performed on such clauses as .b is better conceived of as empty place formation. The referent of the empty place is filled in by syntax, semantics or by inference. In , it is part of the meaning of the verb persuade that the subject of the dependent infinitive is understood to be the object of persuade.
|1.||Linda persuaded Irvin [to marry her].|
The (empty) subject position of the dependent infinitive is under phoric control of the direct object of the superordinate verb.
Consecutio temporum is a constraint on the choice of tenses in a type of complex sentence.1 The traditional paradigm case is a constraint on the tense of a Latin temporal clause introduced by postquam ‘after’. In classical Latin,2 the conjunction postquam requires that perfect tense in the subordinate clause which is anterior to the tense of the main clause, as illustrated by .
|.||a.||Postquam cibum praeparaveram, edi.|
|Latin||After I had prepared food, I ate it.|
|b.||Postquam cibum praeparavi, edo.|
|After I have prepared food, I eat it.|
|c.||Postquam cibum praeparavero, edam.|
|After I will have prepared food, I will eat it.|
In .a, the main verb is in a past tense, so the verb of the anterior clause is in pluperfect. In #b, the main verb is in present tense, so the subordinate verb is in the (present) perfect. In #c, the main verb is in the future, so the appropriate anterior tense is the future perfect. The point here is that the constraint is an ingredient of the tightness of the bond between the two clauses.
Cosubordination (term introduced in Foley & Van Valin 1984; see also Van Valin & LaPolla 1997) is a grammatical relation between verbal constructions which may be clauses, clause cores or predicates. The relation of adjunction may be considered as its analog at the sentence level, but is here treated as a kind of subordination. As its name indicates, cosubordination shares properties both with coordination and with subordination and is half-way between them.
(Franklin 1971 apud Van Valin & LaPolla 1997:450) is from Kewa (New Guinea), a language that has the following three complex sentence constructions:
|He is coming, but I am not afraid.|
|I whistled while I came.’ or: ‘I came whistling.|
|[ 1.SG||whistle||say-PRS.1.SG-CAUSAL ]||dog||come-FUT.3.SG|
|Because I am whistling, the dog will come.||(Van Valin & LaPolla 1997:450)|
- .a is a paratactic combination of two independent clauses into one sentence. The properties of either are independent of the properties of the other.
- #b illustrates cosubordination. The first clause is non-finite, as the verb does not bear the final tense-person-number suffix appearing on finite verb forms as in #a and #c, but instead only shows relative tense and relative subject identity. More specifically, instead of tense we have a binary contrast between simultaneous and anterior with respect to the tense of the following clause. Similarly, instead of identifying person and number of the subject, the morpheme only marks whether the subject of the next clause is identical or not to the subject of the current clause.
- The construction of .c is a hypotactic combination of a causal subordinate clause with a following main clause. The subordinate clause depends on its final suffixal conjunction, but in contrast to #b, its verb is finite.
The strategy of marking, at the end of a clause, the identity or otherwise of its subject with the subject of the following clause is called switch-reference.
The strategy of switch-reference is typically bound up with a complex sentence structure roughly as follows (and illustrated by .b):
[ medial clause1 ... medial clausen final clause ]S
The main clause is the final clause. It is finite and, thus, specified for all verbal categories. The so-called medial clauses are non-finite and dependent, i.e. they can neither be final in a chain nor can they constitute a sentence.
Verbs in a series are in the relationship of cosubordination. This relationship is tighter than coordination of verbs. This is shown by tests which apply certain clause-level operators such as tense, aspect, mood and negation separately to the would-be conjuncts and fail for verb series.
Despite their name, verb series do not have a serial structure:
- The valencies of the member verbs combine in a regular way so that the complex as a whole can take nominal dependents.
- The complex as a whole is marked for categories such as tense, mood, aspect and negation. In accordance with word order patterns, this is done either at the start or at the end of the series.
- A verb series may be a construction consisting of definite syntactic slots (“attractor positions” according to Bisang 1992). The verbs occupying these slots fall into semanto-syntactic classes such as verbs of body position, movement, of transport, of object manipulation etc.
- Feature #2 is a presupposition for the lexicalization of a verb series. Feature #3 is a presupposition for the grammaticalization of a verb series. This is treated in the section on reduction.
Prolepsis is the anticipation of material in a clause where it does not belong semantosyntactically, from a subsequent clause where it does belong. is an example.
|.||Viden me [ ut rapior ]?|
|Latin||Don't you see (me) how I am being kidnapped?||(Pl.Rud.869)|
A specific variety of prolepsis which is grammaticalized in many languages is clitic climbing, illustrated in .
|.||a.||Quise [ decirtelo ayer ].|
|Spanish||I wanted to tell you yesterday.|
|b.||Te lo quise decir ayer.|
The examples in consist of a main clause and a dependent infinitival some of whose dependents are clitic pronouns. In the #a version these clitics attach to the infinitive, thus reflecting their semantic relation in structure. The #b version has the clitics instead on the higher verb, where they do not belong from a semanto-syntactic point of view. This raising of clitics from a dependent clause to the main clause predicate is clitic climbing (Aissen & Perlmutter 1983). In certain varieties of Spanish, the #b version is preferred. It is nevertheless easily described with reference to the #a version, e.g. by a transformation that raises the clitics into the main clause. The functional effect of this transformation may be described as tightening the bond between the two predicates. I.e. the #b version contains a desiderative verb querer decir “to want to say”. Similar observations can be made for causative constructions in many languages.
1 This definition is stricter than those commonly offered in Latin coursebooks. There, consecutio temporum is usually considered as the choice of tense in a subordinate clause to code a certain temporal relation, viz. anteriority, simultaneity or posteriority, to its main clause. This, however, does not in general obey any grammatical constraint, but just follows the meaning to be expressed.
2 The example in the section on syndesis which violates this constraint is from Old Latin.
Aissen, Judith L. & Perlmutter, David M. 1983, "Clause reduction in Spanish." Perlmutter, David M. (ed.), Studies in relational grammar 1. Chicago & London: Chicago University Press; 360-403.