Traditionally it was assumed that coordinate structures can only conjoin elements of identical syntactic categories (Law of the Coordination of Likes, LCL, Chomsky 1957; Williams 1981). The debate on the validity of the LCL (recently Przepiórkowski 2022; Patejuk & Przepiórkowski 2023; Bruening 2025) seems to have come to the agreement that, when both conjuncts are independently selected, the LCL is too strict and mismatches in category are in fact allowed, (2). Instead, these structures have to obey Wasow’s generalization, (1) (see also Huddleston & Pullum 2002). Thus, while become in (2) selects both NP and AP, it does not select a PP, and does not allow PPs in coordinations.

But apparently Wasow’s generalization is still too strict for structures like (3). A CP that is not independently selected by the predicate, (3b), can appear in a coordination with a selected nominal, (3a). The order of the DP- and CP-conjuncts plays a role: reversal of the order leads to ungrammaticality, (3c).

These selection-violating coordinations have received analyses either as proper DP&CP coordinations violating (1) (e.g., Munn 1993; Peterson 2004; Zhang 2010), or as structures where a clause may be headed by nominal functional structure, [ DP & [ DP [ CP ]]], which technically obey Wasow’s generalization, (e.g., Sag et al. 1985; Neeleman et al. 2022). Recently, two new proposals take the analysis in different directions: Al Khalaf 2015; Bruening & Al Khalaf 2020; Bruening 2025 argue that linear adjacency to the selector is crucial in mismatching coordinations and propose a left-to-right derivation for linear order effects. Kim & Lu (2024) argue that selection-violating coordinations are not a real syntactic phenomenon at all, but instead an effect of processing: high processing costs make selection failures harder to detect, which gives rise to high acceptability ratings.

I present observations from German that argue against both the linear adjacency account and the processing account. German is an OV language, where a coordination of objects linearly precedes the selecting verb. Both of these accounts make the prediction that the rightmost conjunct should be the one selected by the verb, because it is linearly or temporally closer to it. Empirical evidence from a forced-choice experiment shows that this is not borne out. Speakers strongly prefer the order of conjuncts in which the selected DP is the leftmost conjunct, and the unselected CP is the rightmost one. The results are predicted in bottom-up derivations in which coordinations have an asymmetric structure, and raise questions about cross-linguistic variation.

This squib is structured as follows: section 2 gives an overview of the analyses of mismatching coordination. Section 3 presents some background on German and two experiments. Section 4 discusses theoretical implications of the results.

Mismatching DP-CP coordinations still pose a theoretical puzzle. There are three groups of analyses for these coordinations, all of which face empirical challenges. In one group, sentences such as (3a) constitute real DP&CP coordinations, i.e., Wasow’s generalization can be violated. Grammaticality is attributed to the more prominent position of the initial conjunct: only the higher, c-commanding DP-conjunct is licensed in this position, while the other (CP-)conjunct is too deeply embedded to be accessible to the verb, and can thus be ‘smuggled’ into a complement position where it otherwise cannot surface. This approach is illustrated in (4). ‘Licensing’ here can be understood either in terms of c-selection (Peterson 2004; Zhang 2010) or in terms of case assignment (Munn 1993). These analyses cannot explain why these coordinations are as restricted as they are (CPs can occur in DP-positions, but almost no other violation of selection requirements seems to be possible, see Bruening & Al Khalaf 2020; Przepiórkowski & Patejuk 2025). A related approach posits a nominal shell on top of the CP-conjunct, such that we are actually dealing with a DP&DP structure, (5), e.g., Sag et al. (1985). The challenge for this type of account is to adequately restrict the availability of null DP shells such that they can derive the ungrammaticality of (3b). I will discuss both of these analyses under the label bottom-up group. They both assume some type of asymmetric structure for coordinations.1 The possibility of a mismatch in these accounts comes from the structural asymmetry between the conjuncts. The first/left conjunct will always be more prominent for the selecting element than more deeply embedded ones. This group predicts that, all else being equal, mismatching coordinations in an OV-language like German should pattern exactly like in (3a), i.e., only a non-initial conjunct can be unselected.

In the second group of analyses, Bruening & Al Khalaf (2020); Bruening (2025) also postulate an empty noun on top of CPs, but they assume a left-to-right derivation. The motivation for this is their observation that it is not always the final, deeper embedded conjunct that can violate selectional requirements. In (6a), the linearly first conjunct is the non-selected one.

I call this type of approach the linear closeness group. Derivations in these analyses proceed not bottom-up but left-to-right (following e.g., Phillips 1996). In this analysis, selection has a syntactic and a semantic component. Only the conjunct that is linearly closest to the selector has to meet both syntactic and semantic selection requirements. Conjuncts further away only have to fit the syntactic requirement. If the noun phrase is closer to the selecting verb, i.e., the first conjunct in (3a) and the last conjunct in (6a), both the syntactic and semantic selection features can be checked, (7). If the closest conjunct is the that-clause, the semantic features will be unchecked, leading to a crash. This is derived in a left-to-right derivation by the order of Agree operations that dictate the accessibility of features in the highest projection of the coordination at the point when it merges with the selector. Space restrictions forbid going into further detail, but I refer the reader to Bruening (2025); Bruening & Al Khalaf (2020).

These analyses have similar problems as the bottom-up group in deriving the distribution of NP-shelled CP: they propose that CPs are disallowed in (3b) because they are semantically empty, i.e., not sentient or animate (Bruening 2025: 29, 31). However, this can certainly not be the whole story, since predicates like depend on can also combine with inanimate NPs and inanimate non-NPs, like small clauses, (8).

Crucially, in the linear closeness accounts the often observed selection of only the initial conjunct as in sentences like (3a) is an accident due to the verb-initial word order in English: the first conjunct in (3a) happens to be linearly closer to the verb than the second conjunct. In different contexts, English apparently also allows the reverse order, as in (6). The linear closeness accounts predict that OV languages should show the opposite pattern: the linearly final conjunct should generally be the selected one.

The last group of analyses is the temporal closeness one. Kim & Lu (2024) propose a processing-based account of mismatching coordinations in which they are a grammaticality illusion. They are actually ungrammatical, i.e., Wasow’s generalization is real and cannot be violated, but are rated as acceptable because the parser gradually ‘forgets’ the selectional requirements of the predicate as the sentence proceeds (Gibson & Thomas 1999; Futrell et al. 2020). Kim & Lu (2024) argue that the more time passes between parsing of the selector and parsing of the mismatching CP conjunct, the less likely the mismatch is to be detected. The verb is encountered first, and the parser registers its selectional features. When the first conjunct matches the features, parsing proceeds, and with passing time, the verb’s requirements fade in memory. When the parser encounters the second conjunct, a possible mismatch cannot be detected anymore.2 Their evidence comes from an acceptability judgment experiment in English that found that the closer a CP-conjunct is to the selecting verb in a three-way coordination, the more acceptability decreases: (9b) is rated significantly worse than (9a).

In an OV language, the selectees are encountered before the selector, but the same logic should hold: increased dependency length (of feature checking) leads to decreased mismatch detection. The first conjunct prompts the parser to expect a selecting predicate. While waiting for the predicate, the representation and features of the conjuncts have to be held in working memory, incurring storage costs. The oldest information is the first conjunct. It will be forgotten first. Thus, when the predicate is encountered, the features of the second conjunct will still be in storage and can be checked against the selectional requirements. The expectation is that a mismatch here is more likely to be detected than mismatch with the first conjunct. This means that speakers should (erroneously) accept structures in which the mismatching conjunct is the leftmost one, i.e., they predict the same pattern as the linear closeness accounts: a non-selected CP should only be able to surface in the position that is further away from the selecting verb, (10a). This contrasts with the bottom-up group that predict (10b), where the unselected CP should be the more deeply embedded conjunct.

This squib contributes to the theoretical debate of mismatching coordination structures by evaluating competing analyses. The group of approaches that makes use of structural asymmetries between conjuncts makes the right predictions for the acceptability of DP-CP-coordinations in German, whereas approaches that make use of linear or temporal asymmetries between conjuncts make exactly the wrong predictions. The effect of linear adjacency in (6), i.e., the motivation for the linear closeness accounts, cannot be observed in German. As far as I know, the order effect has never been demonstrated experimentally in English. (See also Patejuk & Przepiórkowski 2023; Przepiórkowski & Patejuk 2025 for criticism of the generality of data like (6)). Similarly, there was no effect of temporal adjacency such that greater temporal distance between mismatching conjunct and selector leads to greater acceptability.

The results suggest that adjacency, either in terms of linear order or temporal sequence, does not play a role for selection-violating coordinations in German. Instead, they are compatible with bottom-up analyses that exploit the asymmetry between conjuncts illustrated in (4) and (5) above. As an anonymous reviewer points out, the results presented here are not straightforwardly an argument for this approach, but only against the temporal/linear closeness approaches. There could be a general, independent preference for DP-first order, even with predicates that could select both the DP-conjunct and the CP-conjunct. But even if that were the case, the possibility of DP-CP orders in preverbal position is completely unexpected in linear/temporal closeness accounts. The bottom-up approach is therefore only supported indirectly, by weakening the motivation for linear/temporal approaches. As pointed out above, the bottom-up accounts also face problems. However, a full analysis of mismatching coordinations that overcomes the shortcomings of the bottom-up approaches is beyond the scope of this squib.

At the least, we can conclude that the data described for English mismatching coordinations do not constitute a general pattern. If there is a real difference between English and German, this raises the question of how a system looks like that is flexible enough to derive both types of patterns, or alternatively, whether the direction of derivation is something that could be parameterized. Symmetric coordination structures (Neeleman et al. 2022; Przepiórkowski 2024, see also fn. 1) could provide a direction for future research: it is possible that linear order effects are correlated with flat coordinate structures. There are arguments that English coordinate structures are symmetric, but these arguments don’t seem to carry over to German (Schwarzer & Weisser 2024). An exploration of this correlation could provide insights.