Before delving into the analysis, let me spell out my theoretical assumptions. First, I assume an asymmetric structure for coordination of the kind proposed by Johannessen (1998), given in (17) (see also Munn 1993; Zoerner 1995; Progovac 1998a; b; and the references therein for other variants of such an asymmetric structure).

Second, I assume that agreement is the result of an Agree operation, indicated with a dotted line in what follows, between a Probe (T or C in cases considered here) and a Goal (the subject DP), resulting in the valuation of uninterpretable features.11 In cases of complementizer agreement, the verb agrees with the subject in number and gender (as shown in (18b)) and the complementizer agrees with the subject in person and number (as shown in (18c)). Verbal agreement is mediated by T, which does not necessarily imply that the verb itself is in T.

Unlike some of the earlier approaches to First Conjunct Agreement (such as Aoun, Benmamoun & Sportiche 1994 or Citko 2004), I assume that all the agreement patterns under consideration here involve the same structure; there is no structural ambiguity.12 While structural ambiguity approaches could in principle handle Resolved Agreement on both the complementizer and the verb and sandwiched agreement, the “mixed agreement” pattern, where the complementizer agrees with a single conjunct and the verb shows Resolved Agreement, is problematic, since it would require two different structures simultaneously; one to get Resolved Agreement on the verb, and another one to get First Conjunct Agreement on the complementizer.1314

When the subject is coordinate, well-studied feature resolution principles determine agreement (see Corbett 1983; 1991; Dalrymple & Kaplan 2000; Wechsler & Zlatić 2003; Van Koppen & Rooryck 2008; Ruda 2010; Prażmowska 2016, among others). Table 3 summarizes these principles for Polish.

These are the principles that in (19), for example, yield first person plural nonvirile agreement on verb:

Feature resolution is typically taken to be something that is mediated by the Conjunction Phrase (&P) or its head; either all phi-features are present on &P or a subset thereof. In (20), for example, all phi-features are present on &P, and in (21), due to Bošković (2009); Marušič, Nevins & Badecker (2015), among others, number but not gender is present on &P, presumably due to feature percolation and feature resolution on &P.16

Instead, I propose that neither &P nor its & head have any phi-features whatsoever; only individual conjuncts do. The structure is thus (22) instead of (20) or (21).17

This is what distinguishes my proposal from many alternatives, which also rely on a single structure, but allow the Probe to agree either with the entire Conjunction Phrase or just with the first conjunct. Some, such as Babyonyshev (1996) or Van Koppen (2007), take the first conjunct and the Conjunction Phrase to be equidistant from a higher Probe, which is what allows it to agree with either.18 Other researchers allow &P to be underspecified with respect to some (or all) phi- features and/or probe differently for different types of phi-features (e.g. Bošković 2009; Ruda 2011; Marušič, Nevins & Badecker 2015). Bošković, for example, takes &P to be specified for number but not for gender. First Conjunct Agreement for him is thus the result of the Probe agreeing with the entire &P for number and with the first conjunct for gender.19 Marušič, Nevins & Badecker (2015) also rely on the ability of the Probe to probe independently for number and gender. They further assume that the conjunction head has only number (and no gender) features, and that Agree is a two-step process, consisting of Agree-Link (which establishes the relevant relationship between the Probe and Goal) and Agree-Copy (which copies the actual values), the latter being post-syntactic.20 Agree-Copy is relativized to the linearization procedure which turns hierarchical constituency to a flat string. If it happens before the linearization procedure (“flattening of the two conjuncts”), the result is agreement with the hierarchically closest conjunct. If it happens after the linearization procedure, the result is agreement with the linearly closest conjunct. The difference between Single Conjunct Agreement and Resolved (or default) Agreement lies in whether agreement targets the Conjunction Phrase only or “peeks” into the Conjunction Phrase. The “No Peeking Grammar”, in which it does not peek into the conjuncts and the & head has a value for number but not gender, results in default agreement.21 The mixed agreement pattern is problematic, as it would require a conflicting feature specification on the Conjunction Phrase: the presence of number features on the Conjunction Phrase to get Resolved Agreement (“No Peeking Grammar” in Marušič, Nevins & Badecker’s terms) but the absence thereof in order to get First Conjunct Agreement on the complementizer (“Peeking Grammar”).2223

Returning to the lack of phi-features on &P, an important aspect of my proposal, there seem to be no independent reasons for positing phi-features on the &P or the & head, other than capturing agreement with coordinate subjects. Furthermore, there are other more concrete reasons against positing phi-features on &P. First, it is not clear how they get there.24 Second, we do not see overt realizations of these phi-features on conjunction heads or conjunction phrases, as shown by the ungrammaticality of the examples in (23a–b).25

And third, Resolved Agreement is also possible with larger (non-DP) conjuncts, where the agreeing verb occupies a position above the coordination level, as shown schematically in (24). It is unlikely that in this scenario, phi-features would percolate all the way up to &P.

This is the configuration involved in a construction Kazenin (2002) assimilates to gapping, illustrated in (25), which is modeled on Kazenin’s Russian example (see also Citko 2018).

This is the so-called summative agreement (aka cumulative agreement), also found in Right Node Raising (RNR), illustrated in (26a–b) for English (see Postal 1998; Yatabe 2003; Chaves 2014; Grosz 2015; Citko 2018). In both (25) and (26a-b), the verb can be plural in spite of the fact that the two singular subjects are buried inside their respective conjuncts.26

I follow Grosz (2015), who treats agreement with coordinate singular DPs as “completely analogous” to summative agreement in RNR and takes both to be a result of Multiple (Simultaneous) Agree between a single Probe (T/Participle) and two Goals (the two conjuncts), as shown in (27).

Importantly, feature resolution happens on the Probe (T in this case) not on the Goal (the Conjunction Phrase).27 In Grosz’s (2015) account, each DP enters the derivation with referential features, which he represents as numerical indices, with distinct indices encoding distinct reference. As part of the Agree operation, these indices are copied onto T. If two distinct indices are copied (as is the case when T undergoes Multiple Agree with two DPs), the result is Resolved Agreement. However, I depart from Grosz in arguing that there is another option, which is the option that will yield agreement with a single conjunct: First or Last Conjunct Agreement. T, instead of undergoing Multiple Agree with both conjuncts simultaneously, could instead agree with DP₁ first and DP₂ next. As a result of these two “Singular” Agree operations, T ends up with two distinct agreement values, as shown in (28a–b).2829 The two types of agreement are similar in spirit to what Shen (this volume) refers to as distributive and summative agreement, respectively.

The presence of two conflicting feature values on T is resolved at PF. Resolution can be achieved in two ways; either via syncretism (if the Lexicon contains a single form that can realize these two features) or via proximity whereby the value of the linearly closest conjunct is the one that is realized. Which one it is will thus depend on the position of the Conjunction Phrase relative to the agreeing head; with preverbal conjoined subjects, it is going to be the last conjunct (as shown in (29a)), whereas with postverbal conjoined subjects, it is going to be the first conjunct (as shown in (29b)).30 As one of the reviewer points out, T has to be able to “know” which set of features came from which conjunct, and suggests that the mechanism of index copying that Grosz proposes for summative agreement can capture it. I follow this suggestion; for Grosz referential features of both conjuncts are encoded as numerical indices. Agree between T and a conjoined subject DP copies these two different sets of indices. These indexes will then identify which conjunct the relevant features came from.

An important question, brought to my attention by one of the reviewers, is what determines the choice between different types of resolution strategies I have employed in this paper. I have linked the two types of resolution to different type of Agree; Multiple Agree leads to resolution by the principles in Table 3, and Singular Agree leads to resolution by syncretism or proximity. A related question concerns the point in the derivation where these resolution rules apply. The view I take here is that Multiple Agree, which is a syntactic process, is what gives rise to feature resolution. This in turns suggests that feature resolution can take place in the syntax. The reviewer further asks “why repair via syncretism/spell-out of a single feature set is not available with Multiple Agree, and why resolution rules cannot be used to repair a feature set that results from Singular Agree.” I take the answer to be related to the fact that two instances of Singular Agree copy two distinct feature sets at two different points in the derivation, whereas Multiple Agree copies them simultaneously.

Another potential question that the mechanism of multiple instances of Singular Agree proposed here raises concerns intervention effects. I assume that T cannot agree with DP₂ without agreeing with DP₁ first; DP₁ does not act as a defective intervener (see, for example, Broekhuis 2007; Bruening 2014; Moreno & Petersen 2017, for a discussion of the issues the concept of Defective Intervention raises, and a reanalysis of the relevant facts in a way that does not rely on Defective Intervention).3132 I will thus proceed on the assumption that two separate instances of Agree are in principle possible and result in two separate sets of values on T, without incurring a violation of Defective Intervention.

An alternative that avoids the issue altogether would be for T to always undergo Multiple Agree with both conjuncts, which would always yield two distinct sets of phi-feature specifications on T. The actual PF realization of this multiple feature specification could then be determined in any of the following ways: via insertion of a syncretic form if the lexicon contains such a form, via feature resolution rules (yielding Resolved Agreement) or via proximity (resulting in the spell-out of the agreement value of the closer conjunct).33

While such an account would be simpler in that it would only need a single Agree mechanism (i.e. Multiple Agree), it would not be able to exclude all the patterns we want to exclude.

With this as background, we turn to deriving the possible patterns, and ruling out the impossible ones. This is the topic of the next section.

Resolved Agreement on both the complementizer and the verb, illustrated in (31), is the most straightforward one to account for.

Its derivation is schematized in (32a–c). First, T undergoes Multiple Agree with both conjuncts, which will yield Resolved Agreement on the verb. Next, &P moves to [Spec, TP]. This is possible because T has agreed with both conjuncts. And, finally, C undergoes Multiple Agree with both conjuncts, which will yield Resolved Agreement on the complementizer.

However, the complementizer, instead of showing Resolved Agreement, can agree with the first conjunct instead, as shown in (33).

The only difference between this case and the case we just considered is that the complementizer, instead of undergoing Multiple Agree with both conjuncts simultaneously, agrees only with the first conjunct. This derivation is given in (34a–c); Multiple Agree between T and the two conjuncts yields Resolved Agreement on the verb, and Singular Agree between C and the first conjunct yields First Conjunct Agreement on the complementizer. This is essentially the sequence of operations Van Koppen (2007) proposes for Dutch; however, Van Koppen allows T to agree with the entire &P. As pointed out by two of the reviewers, C agreeing with both conjuncts (with feature resolution determined by proximity) would yield the same result. Since agreement with only one conjunct is sufficient, I assume this is all that needs to take place.

The sandwiched agreement pattern, where the complementizer agrees with the first conjunct and the verb agrees with the last conjunct, is illustrated in (35).

Its derivation is schematized in (36a–e). Since both T and C undergo two instances of Singular Agree (with DP₁ first and with DP₂ next), both T and C end up with two sets of conflicting values.

Since the lexicon does not contain an appropriate syncretic form for either the verb (i.e. there is no past tense verb form that would be simultaneously feminine and masculine) or the complementizer (i.e. there is no complementizer form that would be simultaneously first and third person), resolution is sensitive to linear proximity. This yields the sandwiched pattern; the verb agrees with the last conjunct and the complementizer with the first one:

This concludes the discussion of the possible patterns; we have seen that they can all be derived without phi-features being present on the Conjunction Phrase or the Conjunction head. In the next section, I turn to the discussion of the impossible agreement patterns.

The first pattern we want to exclude is one in which both the complementizer and the verb agrees with a single conjunct, but, crucially, with the same conjunct. In (38) below, both agree with the first conjunct.

As we saw in the previous section, First Conjunct Agreement could be the result of the relevant Probes (C and T in this case) undergoing Singular Agree with just the first conjunct or Singular Agree with both conjuncts. Let us consider these two possibilities in turn. If T agrees with the first conjunct without agreeing with the second conjunct, as shown in (39), the derivation stops. Given that Agree is a prerequisite for movement in that only elements that have undergone Agree can move, the entire Conjunction Phrase cannot move to [Spec, TP].34

If, instead, T undergoes Agree with both conjuncts (as in (40a–b)), the Conjunction Phrase can move to [Spec, TP], but this will not yield First Conjunct agreement on the verb. Now the Conjunction Phrase is preverbal (see (40c)), which means that the resolution of the conflicting feature specification on T (T_{uφ:F.SG, M.SG}) will get determined by the conjunct that is closer to T, which is the last conjunct not the first one. This will yield First Conjunct Agreement on the complementizer but not on the verb.

The next agreement pattern we want to exclude is the one illustrated in (41), where both the complementizer and the verb agree with the last conjunct.

Here, we face the opposite problem. The sequence of operations given in (40a–e) will yield Last Conjunct Agreement on the verb but not on the complementizer. Thus, there does not seem to be any way for both the verb and the complementizer to agree with the same conjunct.

The next two patterns we want to exclude involve Resolved Agreement on the complementizer and single conjunct agreement on the verb. In (42), the complementizer has Resolved Agreement, and the verb agrees with the first conjunct.

The problem with this example is that there is no way to get First Conjunct Agreement on the verb with the conjoined subject being preverbal; if T agrees with DP₁, the Conjunction Phrase cannot move to [Spec, TP], as shown in (43a–b).

If T agrees with both conjuncts instead, as shown in (44a–b) below, the entire &P can move, but this can only yield Last Conjunct Agreement on the verb, since the last conjunct is the one closer to the verb. Resolved Agreement on the complementizer is not a problem; it is the result of Multiple Agree between C and the two conjuncts, illustrated in (44d).

However, this sequence of operations yields Resolved Agreement on the complementizer and Last Conjunct Agreement on the verb, which is the ungrammatical pattern in (45a), which we also want to exclude. In (44a–c), T undergoes Singular Agree with DP₁ and DP₂, which will lead to Last Conjunct Agreement on the verb, and C undergoes Multiple Agree with DP₁ and DP₂, which will lead to Resolved Agreement on the complementizer. The issue with (45a) cannot be the issue of Resolved Agreement co-occurring with Closest Conjunct Agreement, since we saw that the opposite, Resolved Agreement on the verb and Closest Conjunct Agreement on the complementizer, is possible. The relevant example is repeated in (45b).

The question then is how to exclude the derivation in (44a–d). In essence, we want to allow a sequence of operations in which Singular Agree follows Multiple Agree, but disallow one in which Multiple Agree follows Singular Agree. General economy principles provide one plausible solution: the possible sequence yielding Resolved Agreement on the verb and First Conjunct Agreement on the complementizer involves two Agree operations (Multiple Agree between T and the two conjunct DPs, and Singular Agree between C and DP₁), whereas the impossible sequence yielding Resolved Agreement on the complementizer and Last Conjunct Agreement on the verb involves three Agree operations (Singular Agree between T and DP₁, Singular Agree between T and DP₂ and Multiple Agree between C and the two conjunct DPs). This in turn raises the question of why a sequence of two Singular Agree operations is ever possible, given the availability of Multiple Agree. What I have argued for here is that both have to be in principle possible. Their interaction, however, is constrained in that Singular Agree on the lower Probe cannot be followed by Multiple Agree on the higher Probe. At this point, I can only offer a speculation on why this should be the case. Given the general economy considerations, I take the ungrammaticality of the pattern in (46d) to mean that the choice of the less economical Singular Agree option on the lower Probe means that the more economical Multiple Agree option for some reason is not available. This could be linked to Feature Inheritance if we assume that in addition to phi-features, T also inherits the ability to probe in a Multiple vs. Singular fashion from C. If C “chooses” the more economical Multiple Agree option, T has to inherit this option. The reverse is not necessarily the case; the availability of the less economical Singular Agree option on C does not preclude the availability of the more economical Multiple Agree option. This is what allows the pattern in (46b).