Intonation constitutes a first difference between backward and forward conjunction reduction. Backward conjunction reduction triggers a marked intonation pattern: (8b) shows that the direct object Els/haar preceding the elided material in (6b) above, as well as its correlate in the antecedent clause, must be contrastively stressed, and therefore cannot be replaced by a reduced pronoun. However, example (8a) shows that such a marked intonation pattern is not necessary in the case of forward conjunction reduction, as is clear from the fact that the indirect object in (6a) above can easily be replaced by a reduced pronoun. Backward, but not forward, conjunction reduction also requires a distinct intonation break preceding the coordinator enand, which is indicated by a comma in (8b); in later examples we will usually not indicate this break, for ease of presentation.

A second difference is that (presumed) forward conjunction reduction generally affects full clausal constituents (arguments, adverbials and complementives), whereas backward conjunction reduction can also affect parts of clausal constituents. In the two (a)-examples in (9), for instance, the subject of the second coordinand can be affected as a whole, but it is impossible to omit only a part of it: the surface string Oude mannen zijn dom en vrouwen hebben geen gevoel voor humor can only be used to express that women in general have no sense of humor. Example (9b), on the other hand, shows that backward conjunction reduction can omit a part of a clausal constituent (here: the nominal head of the direct object oude mannen).

Backward conjunction reduction can omit not only parts of phrases, but even parts of words; the only conditions for allowing this are (i) that the omitted part has the same phonetic form as its antecedent and (ii) that the remnant can be used contrastively. The examples in (10) illustrating this are taken from Royen (1941). Although Royen objects to this kind of reduction on aesthetic grounds and considers it a Germanism, it is frequently used. We take the construction to be part of the core syntax because we see no compelling reason for assuming that it should be considered part of the formal register. Note in passing that in writing the omitted part is usually represented by a hyphen; e.g. schrijf- en spreektaal written and spoken language.

Similar forward conjunction reduction targeting parts of words is hard to find and constructed examples like those in (11) are unacceptable (under the intended reading).

A third difference is that while the target and its antecedent must have the same reference in the case of forward conjunction reduction of nominal arguments, this does not hold in the case of backward conjunction reduction; cf. Dik (1968:81ff.) and Van Oirsouw (1987:30ff.). The case of forward conjunction reduction in (12a') differs from its non-reduced counterpart in (12a) in that the existential quantifier iemandsomeone must refer to the one and the same person: the non-reduced form is generally interpreted as meaning that there are two persons, one buying a box of cigars and one buying a bottle of gin, while the reduced form refers to a single person buying both a box of cigars and a bottle of gin. The case of backward conjunction reduction in (12b), on the other hand, does not differ in this way from its non-reduced counterpart (not given here): both forms are preferably interpreted in such a way that the book bought and the book sold refer to different entities.

Subsection II will show that the syntactic/morphological and semantic differences between forward and backward conjunction reduction follow naturally if we assume that the former differs from the latter in that it does not result from deletion, but instead involves the coordination of phrases smaller than clauses.

The differences between backward and forward conjunction reduction discussed in Subsection I suggest that the two phenomena cannot be regarded as instantiations of a single reduction process. Neijt (1979:50ff.) claims that all acceptable structures supposedly derived by forward conjunction reduction can also be derived without reduction if we allow the coordination of smaller, non-clausal phrases. For instance, an example such as Mannen zijn dom en gewelddadig does not require forward conjunction reduction, as in (13a), but can also be derived without elision by coordinating the two APs, as in (13b). In fact, Neijt claims that the representation in (13a) is ungrammatical, which we have indicated here with an asterisk (although it should be noted that the structure can be derived by gapping but then the adjectives would usually receive contrastive accent; cf. Section 39.2, sub I). The remainder of this subsection will show that this allows us to derive the differences between forward and backward conjunction reduction discussed in Subsection I.

Neijt (1979:53) supports her claim that forward conjunction reduction does not exist by pointing out that it also accounts for the fact that correlative coordinators cannot be used to coordinate the putative reduced clause with its full antecedent clause in the case of forward conjunction reduction: example (14a) does not give rise to an acceptable surface form, while the surface form associated with (14b) is impeccable. For completeness’ sake, note that correlative coordinators can be used to coordinate the reduced clause with its full antecedent clause in the case of backward conjunction reduction: cf. Of Jan heeft Els bezocht of Els heeft Jan bezocht Either Jan has visited Els or Els has visited Jan.

If forward conjunction reduction is impossible, the contrast between the two (a)-examples in (9) under the intended readings follows from the fact that they must be assigned the structures in (15). The attributive adjective oudeold must be combined with the head noun of the subject before the latter can be combined with the verbal predicate. The intended interpretation of the second clausal coordinand in (15b) “old women have no sense of humor” can thus only arise if the attributive modifier is overtly present: the number sign indicates that without it, the second clausal coordinand can only be interpreted generically as “women have no sense of humor”.

Neijt’s claim can also be used to account for the unacceptability of the examples in (11): if forward conjunction reduction is truly impossible, these forms should be directly derived in morphology. However, on the assumption that the embedded structures in (16) cannot be formed by morphology, they cannot serve as input to the morphological rule that derives the surface form either. The asterisks again indicate that we are dealing with ungrammatical morphological structures.

This account of the unacceptability of these forms seems perfectly sound for the two (b)-examples, given the standard assumption that affixes must be attached to free morphemes. It is also plausible for (16c), given that, at least synchronically, the embedded structure consists of word parts that do not function as morphemes. However, the compound in (16a) may require some further justification, given the acceptability of the frequently used compound zonsopgang-en-ondergang to which we will now turn. Example (17a) can be straightforwardly derived by backward conjunction reduction, as expected, but if forward conjunction reduction is excluded in (17b), we must conclude that at least some coordinate structures with free morphemes can be the input for compounding, as in (17b'). That this is indeed possible can be independently motivated by the fact that a morphological process is also needed to derive diminutive and plural forms of compounds of the type kop-en-schotelcup and saucer: [kop-en-schotel]-tje and [kop-en-schotel]-s.

To account for the acceptability contrast between huisdeur-en-bel in (16) and zonsopgang-en-ondergang in (17), we must also assume that the coordinate structures deur en bel and opgang en ondergang differ in that the latter but not the former can be analyzed as a morphologically complex form that can be the input for further compounding. That opgang en ondergang and the backwardly reduced form op- en ondergang can indeed be analyzed as morphologically complex forms is clear from the fact that (i) they usually occur in the order given here and (ii) that they can easily be preceded by a single determiner. A Google search (August 12, 2017) on the two strings [de opgang en ondergang van] and [de op- en ondergang van] yielded in 51 and 149 hits, respectively. That deur en bel cannot be analyzed in this way is clear from the fact that our search on the string [de deur en bel] yielded only one relevant hit.

Since forward conjunction reduction is not necessary for providing a descriptively adequate description of the core data that has been put forward in favor of this rule, we may as well dispense with it in favor of the independently needed assumption that coordination of smaller phrases is possible. This assumption has been extensively motivated on semantic grounds in Section 38.1, sub IV. Here we repeat just two simple cases: if the primed examples in (18) were derived from the primeless ones by forward conjunction reduction, we would wrongly expect these examples to be synonymous, but this is not the case.

The two (a)- examples differ in that (18a') is true only if at least some boys were engaged in both the singing and the dancing, whereas this is not needed for (18a) to be true, since the weaker condition that both activities were performed by at least some boys is sufficient. If (18a') is derived by the coordination of two verbal projections, sommige jongens hebben [gezongen en gedanst], this interpretative contrast follows without further ado. The (b)-examples differ in that (18b) is true only if at least one of the propositions “all men are stupid” and “all men are violent’ is true, whereas (18b') allows both propositions to be false as long as all men are included in the union of the two sets of entities denoted by domstupid and gewelddadigviolent. This interpretative contrast follows without further ado if (18b') is derived by coordinating two adjectival projections, alle mannen zijn [dom of gewelddadig].

The same line of reasoning also explains the contrast between the examples in (12) derived by backward and (putative) forward conjunction reduction. If the “reduced” counterpart of (19a) is not derived by conjunction reduction but by coordination of two noun phrases, as in (19b), the fact that (19a) is interpreted as involving different persons while (19a') is interpreted as involving a single person simply follows from the fact that the former structure involves two clauses and hence two subjects, whereas in the latter structure there is only one clause and hence only one subject.

The argument can be repeated for referential pronouns such as hijhe in (20); whereas (20a) would normally be interpreted as referring to two separate events involving different agents, (20) can only be interpreted as a single event involving a particular person buying both a box of cigars and a bottle of gin. This would again be unexpected if (20a) were a reduced version of (20b); see De Vries (1992:68ff.) for a more complex version of this argument based on the different uses of hetit (as argument, quasi-argument and anticipatory pronoun).

More arguments in favor of the assumption that coordination of smaller phrases is possible were given in Section 38.4.2 on correlative coordinators. Here we repeat only one simple case. Since example (21a) shows that the correlative zowel ... als ...both ... and ... cannot be used to coordinate main clauses, this a priori dooms to futility any attempt to derive the (b)-examples by conjunction reduction; however, the acceptability of these examples follows immediately if we assume that we are dealing with coordinated noun phrases and APs.

Note that although this subsection has argued that the term forward conjunction reduction should be considered a misnomer, we will occasionally use this notion as a purely descriptive term in what follows.

This subsection discusses backward conjunction reduction in more detail, and will argue that this form of reduction cannot be given a syntactic account. We have already seen that backward conjunction reduction affects the right periphery of the target coordinand. That omission of material located elsewhere is not allowed is clear from the acceptability contrast between the simple past tense examples in (22b&c) and their perfect tense counterparts in the corresponding primed examples: (22b') is unacceptable and (22c') cannot be interpreted in the intended way (as indicated by the number sign) because backward conjunction reduction is blocked by the clause-final past participle in the first coordinand.

The primeless examples in (22) involve the omission of a single main verb or a single clausal constituent, but the elided sequence can also be larger. The examples in (23), for instance, show that it is also possible to elide more than one clausal constituent: (23a) shows this for two arguments, (23b) for two adverbial phrases, and (23c) for one argument and one adverbial phrase.

The embedded counterpart of (22a) in (24a) further shows that backward conjunction reduction can also affect a verb cluster, and the cases in (24b&c) show that it is also possible to elide the clause-final verb (cluster) together with a clausal constituent, a direct object in (24b) and a temporal adverbial in (24c). Note in passing that the complementizer datthat in the second coordinand of (24a) can also be omitted, in which case we may be dealing with the coordination of smaller (non-clausal) verbal projections.

The facts in (23) and (24) are potentially problematic for providing a syntactic account of backward conjunction, since syntactic operations usually target only single clausal constituents. One might want to defend a syntactic approach by claiming that the omitted constituents in (23) and (24) form a constituent of some higher order, e.g. a verbal projection with the main verb extracted from it by verb-second in (23), or a verbal projection with one or more objects extracted from it by A-scrambling in (24). Such a proposal, however, would run afoul of examples like those in (25). The primeless examples first show that backward conjunction reduction can also target parts of clausal constituents: the elided part is the nominal part of a PP-complement of the verb in (25a) and the nominal part of a postnominal PP-modifier of the object in (25b). The embedded counterparts of these sentences in the primed examples further show that conjunction reduction can also affect the clause-final verb (cluster) plus a phrase embedded in a clausal constituent.

Since the omitted strings in the primed examples of (25) do not count as constituents in any current theory, we cannot adopt the account proposed for (23) and (24) that we are dealing with the deletion of some higher-order verbal constituent. This argument against a syntactic account of backward conjunction reduction gains even more strength from the fact that the elided string may be even more deeply embedded. Besides (25a), in which the elided string is the nominal complement of the preposition, we also have (26a), in which the elided string is the head of that nominal complement. And besides (25b), in which the elided string is the nominal complement of a postnominal modifier of the direct object, we have (26b), in which the elided string is a postnominal modifier of the nominal complement of a postnominal modifier. In short, it seems that the depth of embedding of the elided element knows no principled bounds.

For completeness’ sake, the examples in (27) show that the elided part can be preceded by different kinds of elements; it can be the finite verb, or (a part of) an argument, a complementive, or an adverbial phrase; nor do there seem to be any independently motivated syntactic operations that could affect the strings preceding the elision site.

The facts discussed in the previous subsection suggest that backward conjunction reduction is not a syntactic rule; cf. Neijt (1979:§2.2) and Zwarts (1986:16). Since syntactic rules usually affect only a single clausal constituent, the fact illustrated in (23) that conjunction reduction can affect more than one clausal constituent raises a warning flag. The fact illustrated in (25) that conjunction reduction can affect parts of clausal constituents is also telling, since such parts cannot normally be affected by syntactic rules applying at the clausal level. The same is shown by the examples in (28) where the elided strings are located within syntactic islands: an embedded wh-question in (28a), a relative clause in (28b), and a coordinate structure in (28c), although it should be noted that some speakers object to the last case.

The claim that backward conjunction reduction is not a syntactic rule is in agreement with the fact that there can be no established syntactic relations between elements within a coordinate structure if they are located in different coordinands. We are therefore led to conclude that we are dealing with a post-syntactic phonological rule that deletes a right-peripheral string in the target coordinand under identity with a string in the antecedent clause. Note in passing that the conclusion that backward conjunction reduction is a post-syntactic rule is inevitable in theories that follow Kayne (1994) and, especially, Chomsky (1995:§4.8) in assuming that information about linear order is not available in syntax, as this makes it impossible to formulate the condition that the elided string is right-peripheral in its clause in syntactic terms.

The conclusion that backward conjunction reduction is a phonological rule that deletes a right-peripheral string in the target coordinand under identity with a string in the antecedent clause is still not fully descriptively adequate. Recall from Subsection I that the element preceding the elided material, as well as its correlate in the antecedent coordinand, must be contrastively stressed. This means that elements that cannot be stressed cannot precede the gap. This is illustrated for reduced pronouns in (29a), for R-pronouns in (29b), and for the indefinite article eena in (29c). The contrastive accent is indicated by small caps. This shows that backward conjunction reduction should also be made sensitive to the prosodic properties of the element preceding the elided string.

Another complication is that the marked intonation pattern of backward conjunction reduction is reflected in the fact that it always implies some semantic contrast: in (29a), for instance, two (ordered) pairs <Jan, Marie> and <Els, Peter> are contrasted. Note that the fact that the contrast in this example involves ordered pairs is accidental: in contexts such as (30) it can also involve singletons.

The main conclusion therefore seems to be that backward conjunction reduction requires not only that the elided string is recoverable from the antecedent coordinand, but also that the rightmost part of the remnant and its correlate in the antecedent coordinand can be contrasted both phonetically and semantically. This suggests that backward conjunction reduction cannot be described in purely phonological terms.

The previous subsections have mainly discussed coordinate structures with clausal coordinands. However, the cases in (31) show that backward conjunction reduction can also apply to coordinands of other categorial types: we have instances of conjunction reduction within nominal, adjectival and adpositional coordinate structures. We refer the reader to the examples in (10) in Subsection I for examples showing that backward conjunction reduction applies not only within phrases but even within words.

The fact that backward conjunction reduction is possible in non-clausal coordinands is important in the light of our earlier conclusion that forward conjunction reduction should be reanalyzed in terms of coordination of phrases smaller than clauses: it makes it possible to account for the fact that backward and forward conjunction reduction can co-occur by assuming that backward conjunction reduction can apply in smaller (non-clausal) verbal projections. We illustrate this here with example (6c) from the introduction: the original analysis of this example, repeated as (32a), can now be replaced by the analysis in (32b) with backward conjunction reduction in a coordinated structure with two smaller, non-clausal verbal projections (which we have labeled as VP, for convenience).

Combinations of forward and backward conjunction reduction can also be found in nominal coordinate structures, as is illustrated in (33a). Note that the string oude mannen en vrouwen uit Duitsland is ambiguous in various ways and can also be analyzed as in e.g. (33b). The intonation patterns associated with the two cases seem to be different: as always, backward conjunction reduction requires that the element preceding the elided string in the target coordinand as well as its correlate in the antecedent coordinand to be contrastively accented, while examples such as (33b) allow a more neutral intonation pattern. Since we are not aware of any systematic investigation of this issue, we will leave it to future research.

Similar ambiguities can be found in coordinated APs, as shown in (34). Again, the two cases given have different intonation patterns: the backward conjunction reduction case in (34a) has the by now familiar contrastive intonation pattern, while (34b) can easily be pronounced in a more neutral way.

Similar cases as in (32) to (34) are more difficult to construct for PPs, but it seems that the interpretation of the two sentences in (35) is what we would expect: the contrastive intonation pattern in (35a) triggers a reading according to which each exercise is preceded and followed by the ringing of a bell; under a more neutral intonation pattern, (35b) seems compatible with a reading according to which the bell rings at the beginning of a contextually determined event (e.g. a training session) as well as after each exercise.

We conclude with a discussion of the more special case involving R-extraction in (36a). The most likely interpretation of this example is that the degree modifier ergvery has both adjectives in its scope, suggesting the structure given in (36b). If this structure is correct, we must conclude that backward conjunction reduction applies after R-extraction, since only then is the elided preposition in the right periphery of the first coordinand. This finding is of particular theoretical interest because, under the standard assumption that syntactic rules apply cyclically, it supports our earlier conclusion that backward conjunction reduction is a phonological rule that applies after the syntactic derivation has been completed. For completeness, note that the alternative structure in (36b') is probably ungrammatical (which is indicated by an asterisk), because R-extraction would then violate the coordinate structure constraint in Section 38.3, sub II.

The conclusion of Subsection II that forward conjunction reduction does not exist and should now be reanalyzed in terms of the coordination of non-clausal phrases raises several empirical (as well as theory-internal) problems that still need to be solved, and which in fact play an important role in Wilder’s (1997) plea for the reintroduction of forward conjunction reduction. The most important empirical problem concerns the status of the coordinands: is coordination restricted to maximal projections, or is it also possible to coordinate (lexical) heads or intermediate projections?