2.1 Ditransitives in French and the PCC

In French, the PCC is observed in ditransitive sentences with two internal-argument clitics. The restriction is such that a first person (1P) or second person (2P) accusative clitic is banned in the presence of a co-occurring dative clitic (Perlmutter Reference Perlmutter1971, Kayne Reference Kayne1975, Bonet Reference Bonet1991, among others), as shown in (1) and (2).

The accusative clitic represents the theme argument, and the dative clitic represents the goal argument (goal > theme). As shown in (1), whenever the accusative clitic is third person (3P), the sentences are grammatical no matter which person appears as the dative clitic. However, when the accusative clitic is either 1P or 2P as in (2), we obtain ungrammaticality irrespective of the person of the dative clitic. I refer to the grammatical sentences in (1) as PCC-compliant sentences, and those barred by the PCC in (2) as PCC-violating sentences.

2.2 Split Agree and the Person Licencing Condition

Béjar and Rezac (Reference Béjar, Rezac, Perez-Leroux and Roberge2003) attribute the ineffability of certain clitic combinations to an interface condition called the Person Licensing Condition (PLC) defined as in (3).

Under their approach, ungrammaticality results if the marked person feature (1P/2P) of the structurally lower argument (accusative/direct object) is not licensed as schematized in (4). Following Taraldsen (Reference Taraldsen, Haider, Olsen and Vikner1995), Béjar and Rezac propose that the phi-probe is split into two components, a person probe (uπ) and a number probe (u#), which can independently establish an Agree relation with an argument.Footnote ² According to their analysis, the person probe on v° probes first and Agrees with the dative argument, the closer potential goal. The dative argument, if it is a clitic, moves to a position higher than v° (=cliticization). The number probe subsequently initiates its search and finds the accusative argument – the dative argument, a potential intervener, having moved out of the search path – but the number probe cannot license the person feature of the argument.

Under this account, if the accusative argument is either 1P or 2P (1/2/3 > 1/2), then the marked person feature of that argument cannot be licensed, which leads to a crash at the interface. This analysis is useful in explaining the availability of an alternative construction, referred to as a “repair”, seemingly dependent on the PLC violation (Rezac Reference Rezac2011). The next section shows the apparent dependency between the PCC construction and the “repair” construction.

2.3 The PCC “repair”

For the ungrammatical, or PCC-violating, sentences in (2), there is an alternative construction, shown in (5), in which the dative argument is expressed as an independent prepositional phrase with a strong pronoun (á + strong pronoun) instead of a clitic. This alternative construction is referred to as a PCC “repair” (Bonet Reference Bonet1991, Reference Bonet, Harley and Phillips1994, Reference Bonet, D'Alessandro, Fischer and Hrafnbjargarson2008; Rezac Reference Rezac2011). The interesting fact that Rezac (Reference Rezac2011) brought to our attention is that no “repair” is available for the PCC-compliant sentences in (1), as is evident from the ungrammaticality of the sentences in (6).

As shown in (5) and (6), the “repair” construction is used only to fill the interpretational gap created by the PCC.Footnote ³ In other words, the PCC-compliant sentences with two clitics and the “repairs” with an à-phrase are in complementary distribution with respect to person combinations of the arguments, as shown in Table 1.

Table 1: The PCC and its “repair” in French (elicited judgements in brackets)

While Rezac (Reference Rezac2011) reports that the sentences in (6) are ungrammatical, some native French speakers I consulted with (marginally) accepted them without any emphasis on the pronoun. As will be shown in section 5, my analysis accommodates these judgements as well. The next section presents Rezac's (Reference Rezac2011) analysis of the PCC repair, and identifies some of the challenges for the analysis.

3.1 Full PPs vs. Defective PPs

An important component of Rezac's (Reference Rezac2011) analysis is that there are two types of prepositional phrases, defective PPs and full PPs. Because dative clitics in French behave differently from their à-phrase counterparts with a strong pronoun, Rezac (Reference Rezac2011) takes these distinctive properties of dative clitics and prepositional datives to reflect a structural difference: dative clitics are defective PPs, and prepositional datives are full PPs. Dative clitics are defective in that the argument inside is visible to an external probe and, therefore, may act as an intervener. Prepositional datives or à-phrases, on the other hand, are full PPs and make the argument inside inaccessible to an external probe. The next section explains how this classification of PPs plays a role in Rezac's (Reference Rezac2011) analysis of the PCC repair.

3.2 Global mechanism to derive the PCC “repair”

Rezac (Reference Rezac2011) proposes that the PCC repair is not generated through a single derivation within the syntactic component. Rather, a dependency relation is established between two separate derivations by means of the interface algorithm. Specifically, the Interface Algorithm ℜ introduces an additional person probe to a non-convergent derivation, which makes the PCC “repair” available. Given a PLC violation as discussed in section 2.2 and schematically shown again in (8a), ℜ is activated, and inserts an uninterpretable person feature on the preposition of the defective dative PP, as in (9). The additional uninterpretable person feature strengthens the dative clitic to a full PP.

This newly inserted probe licenses the goal argument within the PP. The higher argument, being in a full PP, is no longer an intervener for the person probe,Footnote ⁴ and that probe can now license the lower argument (theme) as shown in (8b). This analysis explains the complementarity between the PCC and its “repair”, because the repair arises as the result of a PCC violation. In other words, under this analysis, the PCC “repair” is truly a repair since its only purpose is to rescue a defective derivation.

3.3 Systemic problems with the Interface Algorithm ℜ

The global mechanism discussed in the previous section seems to explain why the PCC “repair” is only available when the sentence violates the PCC. However, the analysis is not without problems. First, there are some systemic issues with the proposal. As pointed out by Rezac (Reference Rezac2011) himself, it is not clear why a rescue mechanism similar to the Interface Algorithm ℜ is not available for other types of ungrammaticality, like that in (9), which Rezac (Reference Rezac2011: 19, (34)–(37)) calls hard ungrammaticality.

1. (9) *They showed her_ACC (*for) seven to be a prime.

In fact, the Interface Algorithm leads us to predict that all ungrammaticality caused by an unsatisfied feature should be reparable, which is clearly not the case. Second, it is also unclear what determines where the additional uninterpretable feature will be inserted. In the analysis presented in the previous section, the additional person probe was inserted on the preposition of the dative argument. However, the dative argument is successfully licensed in the original derivation (8a). The Interface Algorithm ℜ could potentially have repaired the accusative argument, which was the source of the PLC violation, instead.Footnote ⁵ At least in French, the option of repairing the accusative clitic is not available, and the reason is not clear.

3.4 The weak PCC: An empirical challenge for ℜ

Rezac's (Reference Rezac2011) proposal also faces an empirical challenge. There are languages that allow 1/2.dat > 2/1.acc combinations but still ban 3.dat > 1/2.acc combinations (referred to as the ‘weak’ PCC, as opposed to the ‘strong’ version presented in section 2.1; see Heger Reference Heger1966; Ashby Reference Ashby1977, Simpson and Withgott Reference Simpson, Withgott and Borer1986, Schwegler Reference Schwegler1990, Laenzlinger Reference Laenzlinger1993, and Nicol Reference Nicol, Heggie and Ordóñez2005). Some varieties of French have the weak PCC pattern as shown in (10).Footnote ⁶

Weak PCC varieties are problematic for the Interface Algorithm ℜ because these varieties also allow the “repair” construction for 1/2.dat > 2/1.acc clitic combinations as shown in (11) despite the fact that these are licit clitic combinations in these varieties.

The notion of “repair” is inapplicable here, as the double-clitic construction is also available for these person combinations; the sentences in (11) with the goal in a prepositional phrase are simply alternatives. On the surface at least, the PLC violation that should give rise to the “repair” construction is missing. Therefore, the Interface Algorithm ℜ cannot explain why there is an overlap between the PCC and its “repair,” as shown in the box in Table 2.

Table 2: The weak PCC and its “repair”

Rezac (Reference Rezac2011) acknowledges that the weak PCC pattern is potentially problematic for his analysis. The explanation he provides for the weak PCC is that the syntax only sees the strong PCC. Therefore, 1/2 combinations will activate the algorithm ℜ and the repair is available for these combinations. He further suggests that 1/2.dat > 2/1.acc clitic combinations in (10a) are available because the dative argument in these sentences is something like a non-argumental ‘ethical dative.’ As Jouitteau and Rezac (Reference Jouitteau and Rezac2008) explain, ethical datives are optional, are not arguments, and do not contribute any truth-conditional meaning to the sentence. They identify the speaker or addressee as a “witness or vaguely affected party” (Jouitteau and Rezac Reference Jouitteau and Rezac2008:Sec.1). Rezac's (Reference Rezac2011) claim is that in weak PCC languages, the sentences with the 1/2 clitic combinations in (10a) have a completely different configuration. These sentences escape the PCC because the dative argument, being non-argumental, does not participate in agreement with the probe on v. He also suggests that this non-argumental dative receives an idiomatic interpretation as an argument at LF.

Rezac's (Reference Rezac2011) explanation makes incorrect predictions regarding participant argument combinations. If the 1/2 clitic combinations are allowed because the dative argument is actually non-argumental, then all the varieties with non-argumental/ethical datives, including the strong PCC varieties (see Jouitteau and Rezac Reference Jouitteau and Rezac2008), should allow these clitic combinations. This falsely predicts a weak-PCC pattern for strong PCC languages. Furthermore, if something that is not argumental can be interpreted as an argument, then we expect that the me lui combination (3 > 1), which is ruled out by the PCC, should be available. There is no morphological case distinction for 1P and 2P clitics in French, so the 1P clitic me could well be a non-argumental ‘dative’ which is idiomatically interpreted as the theme. However, to my knowledge, 3 > 1 clitic combinations are rejected in all varieties of French. Therefore, while Rezac's (Reference Rezac2011) Interface Algorithm analysis successfully captures the complementarity between the PCC and its repair found in the strong PCC varieties, it runs into problems with the weak PCC varieties, in which the “repair” construction is available for PCC-compliant clitic combinations.

4.1 Featurally constrained Merge

I make two specific assumptions about the Merge operation. First, I assume that there is a featural constraint on Merge. Wurmbrand (Reference Wurmbrand, Kosta, Franks, Radeva-Bork and Schürcks2014) diverges from the traditional view that Merge is free (Chomsky Reference Chomsky, Martin, Michaels and Uriagereka2000) and proposes that it is constrained, as in (12).

This means that the two syntactic objects participating in a Merge operation must be featurally asymmetric: one bears a valued feature ([F:val]) and the other bears an unvalued feature of the same type ([F:_]). I follow Wurmbrand (Reference Wurmbrand, Kosta, Franks, Radeva-Bork and Schürcks2014) and assume that Merge is triggered and constrained by feature valuation.

The second assumption comes from the idea of selection. Adger (Reference Adger2003) proposes that Merge has the property in (13), which is motivated by (c-)selection.

In Adger's system, the selector has an uninterpretable (categorial) feature, which needs to be checked by an interpretable counterpart of the same feature. For example, a transitive verb like press has an uninterpretable D-feature. That feature is checked and deleted upon merging with a DP like the button, which carries an interpretable D-feature. As stated in (13), the selecting verb projects as a result of this Merge operation. Although I do not assume c-selection as Adger (Reference Adger2003) does, I follow his general idea, and assume that a syntactic object whose feature is satisfied (or given a value) via a Merge operation projects along with all of its features.

In addition to these two assumptions, I further assume that all the features of a projecting syntactic object (the selector) are carried up to the root of the constituent created by a Merge operation. The feature that triggered Merge will be replaced by a valued counterpart of that feature. The properties of Merge assumed in this article are summarized in (14).Footnote ⁷

1. (14) Valuation-based Merge

   1. i. α and β Merge if and only if α has a feature that can value a feature of β

   2. ii. Valuation takes place between a valued feature [F:val] and an unvalued feature of the same type/attribute [F:____]

   3. iii. The participating features must be at the root of α and β

   4. iv. The result of the operation is β projecting with a valued counterpart of the participating feature [F:val] along with other features that β bears

These properties are schematically illustrated in (15).

1. (15) Valuation-based Merge (α, β)

   

Even though the PCC cases that are under consideration here do not require it, I assume that only one type of feature can trigger a particular instance of Merge. The next subsection discusses the second component of the proposed feature system.

4.2 Articulated person features

The second component of the proposed syntactic mechanism is articulated person features. Building on the φ-feature system developed by Harley and Ritter (Reference Harley and Ritter2002), Béjar (Reference Béjar2003) takes morphological φ-features to be visible to Agree (see also Béjar and Rezac Reference Béjar and Rezac2009). Béjar proposes the person specifications in (16).

In this system, the person features are “articulated,” meaning that they are composed of dependent features in an entailment relation. [Addressee] ([Ad]) entails [Participant] ([Part]), which in turn entails [Person] ([π]). The person probe in Béjar's (Reference Béjar2003) system consists of uninterpretable counterparts of [π] and [Part], as in (16d). I adopt the idea that person features are structured as in (16), and are active in the syntax. However, the φ specifications I employ differ from (16) in two respects. First, there is an additional feature [Speaker] ([Sp]) that differentiates 1P from 2P (McGinnis Reference McGinnis2005).Footnote ⁸ Second, there is a featural animacy distinction in the third person. The proposed feature system employs the feature specifications in (17) for the strong PCC. As shown in (17), I make an animacy distinction in the 3P arguments, following Ormazábal and Romero (Reference Ormazábal and Romero2007), Richards (Reference Richards2008), Adger and Harbour (Reference Adger and Harbour2007), and Pancheva and Zubizarreta (Reference Pancheva and Zubizarreta2018).

The 3P inanimate is specified only for [#] (number), while the 3P animate is also specified for [π] (person).Footnote ¹⁰ 2P is inherently specified up to [Ad], and 1P is specified up to [Sp]. A crucial aspect of the proposed feature system is that it makes use of unvalued features instead of uninterpretable features. The difference between uninterpretable features and unvalued features is trivial in Chomsky's (Reference Chomsky and Kenstowicz2001) system, since uninterpretable features are necessarily unvalued. However, under the non-generation approach, which takes ungrammaticality to be non-generability (Preminger Reference Preminger, Hornstein, Lasnik, Patel-Grosz and Yang2018), the distinction is important: uninterpretability is taken to cause a crash at the interface, while unvaluedness is simply the absence of a value. An unvalued feature is still legible for external systems (e.g., the insertion of a default agreement morpheme). Furthermore, as with the probe in Béjar's (Reference Béjar2003) system, an unvalued φ-feature is “active” until all of its component features are saturated. Using both valuation-based Merge in (14) and articulated person features in (17) has an interesting consequence, which I discuss in the next subsection.

4.3 Incremental Valuation

An advantage of using articulated person features in the Cyclic Agree system (Béjar and Rezac Reference Béjar and Rezac2009) is that a single probe can enter into an Agree relation with two arguments. Although similarities have been drawn between the PCC and direct/inverse patterns, which Cyclic Agree was originally proposed for (see, for example, Rezac Reference Rezac2011), Cyclic Agree is not designed to explain ungrammaticality and cannot simply be extended to the PCC cases. The syntactic system proposed here retains the above-mentioned advantage of Cyclic Agree, but also explains ungrammaticality in terms of non-generability. It allows arguments to interact with each other through the mediation of a single unvalued feature. With the additional mechanism of valuation-based Merge, this essentially means that arguments featurally engage with each other through Merge. This subsection spells out the exact mechanism, which I call Incremental Valuation.Footnote ¹¹

An unvalued articulated φ-feature, as shown in (17e), is composed of dependent unvalued features. Valuation-based Merge, as described in (14), projects a syntactic object with a feature that has received a value, making the valued counterpart of the feature visible at the root. This means that the same unvalued φ-feature can be used more than once to trigger a Merge operation if the first instance of Merge does not fully saturate the composite φ-feature. In other words, one can have multiple arguments valuing the same unvalued φ-feature.Footnote ¹² If, for example, a verb or a functional element with an unvalued φ-feature complex merges with a 3P inanimate argument, the argument values the number feature but leaves other features unvalued, as shown in Figure 1. These unvalued dependent features can later be valued by merging another argument that is more highly specified (i.e., 1P, 2P, or 3P animate). An example with a 1P argument is shown in Figure 2. Figures 1 and 2 show that the unvalued φ-feature can be incrementally valued by two arguments. What is perhaps already apparent is that this structure-building system prevents an argument from valuing an unvalued φ-feature that has already been valued by an equally or more highly specified argument.

Figure 1: Initial Merge (Appl, 3ia.sg)

Figure 2: Subsequent Merge (1sg, Applʹ)

For example, if the order of Merge operations in Figures 1 and 2 is reversed (i.e., 1sg → 3ia.sg), the second instance of Merge cannot take place as the unvalued φ-feature will have been saturated by the 1P argument. More specifically, all the dependent features of the unvalued φ-feature will be valued in the first instance of Merge. There is thus no value for the 3P inanimate argument to contribute to the derivation, making it impossible for the argument to come into the structure. It should be noted that it is not the exhaustion of the unvalued feature that makes the subsequent Merge impossible; rather, it is the lack of additional value to be given to the unvalued feature that makes Merge nonexecutable. For example, once an unvalued φ-feature is valued by a 3P animate argument, subsequent valuation by a 3P inanimate argument is impossible, not because the unvalued feature has been saturated by that point but because the second argument, being less specified, cannot provide any additional value to the unvalued feature. Three pieces of the proposed syntactic mechanism are summarized in (18).

1. (18) Proposed syntactic mechanism

   1. i. Articulated person features

      Any of the dependent features of an unvalued φ-feature can trigger Merge.

   2. ii. Valuation-based Merge

      Merge is necessarily triggered by valuation between the two uniting syntactic objects (SOs); no Merge operation is executable if no value can be transmitted between the two SOs.

   3. iii. Incremental Valuation

      A particular unvalued φ-feature complex can be responsible for multiple instances of Merge.

Before presenting the details of how the proposed analysis accounts for the PCC and its “repair,” there are two additional assumptions that are crucial for the analysis, which I discuss in the next subsection.

4.4 Additional assumptions

The proposed analysis contrasts with that of Rezac (Reference Rezac2011) in that I assume two independently available structures – that is, neither is derived from the other – for the PCC sentences and their “repairs.” For the double-clitic construction, I assume a structure where the goal is higher than the theme. Specifically, I assume Pylkkänen's (Reference Pylkkänen2008) low applicative phrase for this construction, where an Appl head with an unvalued φ-feature mediates the two arguments as shown in (19a). This is the structure proposed for the English double-object construction (e.g., give goal theme) and is compatible with the scopal/superiority facts in English noted by Barss and Lasnik (Reference Barss and Lasnik1986) and Larson (Reference Larson1988).Footnote ¹³ For the “repair” construction, I assume the Larsonian structure for the prepositional dative construction (e.g., give theme to goal) where the theme DP is higher than the goal argument, the goal is in a PP, and the theme DP and the PP are mediated by the verb (with an unvalued φ-feature) as shown in (19b).

Another crucial assumption is that à in French is a preposition inherently specified as 3P inanimate singular. This is motivated by the fact that locational PPs trigger 3P singular agreement both in English (20) and in French (21).Footnote ¹⁴ Note that these sentences are intended to be predicational (‘The place under the bed is good for hiding’) and not existential (‘There is a good place to hide under the bed’).

While one cannot rule out the possibility of default agreement in (20) and (21), resulting from failure of agreement (Béjar Reference Béjar2003, Preminger Reference Preminger2009), prepositions establish a relation between entities (they are of type 〈e, 〈e, t〉〉 in terms of Heim and Kratzer Reference Heim and Kratzer1998). The relation might be locational (in, on), temporal (at), or directional (to, from) among other classes. Therefore, they do not denote a specific entity (e) or a set of entities (〈e, t〉), and it is reasonable to assume that prepositions, which denote relations, are specified as inanimate. Finally, I also assume that French clitics are generated in argument positions and undergo cliticization (Anagnostopoulou Reference Anagnostopoulou, Stepanov, Fanselow and Vogel2004).

With all the machinery and assumptions laid out, the next section explains how Incremental Valuation explains the PCC pattern and the PCC “repair” pattern. The proposed account differs from that of Rezac (Reference Rezac2011), in that the double-clitic construction and the “repair” construction are treated as independent. As will be shown, the present analysis does not run into the same problem as Rezac's (Reference Rezac2011) account faces.

5.1 PCC-compliant sentences

As discussed in section 2.1, in French ditransitive sentences, when the theme is 3P, both internal arguments can be expressed with co-occurring clitics as shown in (1). Partial derivations of (1a) and (1b) are shown in (22) and (23), respectively. The full representation of the resulting feature of each instance of Merge is shown on the right-hand side.

In the case of the 1 > 3 combination in (1a) and (22), the 3P inanimate theme provides a number value to the unvalued φ-feature on Appl, and the 1P goal later values all the remaining person features. Since both arguments can contribute at least one value to the unvalued feature, the derivation is successful. The same is true for the 3an>3ia combination in (1c) and (23),Footnote ¹⁶ where the 3P animate goal passes the [π] value to the unvalued φ-feature on Appl. In all cases (1a,b), there is at least one value that the second argument (goal) can provide to the unvalued φ-feature on the mediating head, which allows the argument to enter the derivation. Therefore, these sentences are generable with the proposed syntactic mechanism. As a natural consequence of the Incremental Valuation system proposed here, the 3ia>3ia, 3ia>3an, and 3an>3an combinations are non-derivable. This explains the general tendency for an applicative argument to be animate without stipulating that an argument introduced by Appl is necessarily animate or specified for [Participant] (Adger and Harbour Reference Adger and Harbour2007: 21,25–6).

5.2 PCC-violating sentences

When the theme is a local person (1P/2P) in a double-clitic construction, on the other hand, the structure is ungrammatical as shown in (2). The theme argument in (2a,b) is either 1P or 2P, and as was shown in (17c,d), 1P and 2P arguments are fully specified. Since the theme is the first argument to enter the derivation, when it merges with Appl, it saturates the unvalued φ-feature on the head, leaving no unvalued dependent features. Since valuation-based Merge can take place only between two syntactic objects with appropriate features for valuation, the goal argument, regardless of its person, cannot enter the derivation after Appl's unvalued φ-feature has been fully saturated. A partial derivation of (2b) is shown in (24).

The unvalued φ-feature in (24) is exhausted by the first instance of Merge; therefore, this structure (i.e., Applʹ) cannot merge with the goal. The PCC-violating sentences therefore cannot be generated with the system at hand.

5.3 PCC “repairs”

For the ungrammatical double-clitic sentences in (2a,b), it has been established that there is an alternative construction with the goal realized as a PP instead of a preverbal clitic. As shown in (5a-c), the 1P/2P theme argument is still a clitic, but the goal is in a postverbal PP. With the assumed structure for the PCC “repair” construction (19b), the PP merges first with the verb bearing an unvalued φ-feature, and the theme argument comes in later. The preposition à is inherently specified as 3P inanimate under the present assumptions, which conceals the phi-features of its complement (i.e., the goal argument).Footnote ¹⁷ The PP merges with the verb and values only the number feature. Since the person features are still unvalued, 1P or 2P arguments can come into the derivation as the theme to value the person features, as shown in the partial derivation of (5b) in (25). As a result, regardless of the person of the goal argument, 1P and 2P can be the theme argument.Footnote ¹⁸

As was the case with PCC-compliant sentences, the “repair” sentences in (5) can be generated, because both internal arguments have values to contribute to the unvalued φ-feature on the verb.

5.4 “Repairs” for PCC-compliant sentences

Finally, the “repair” construction is not available for the PCC-compliant sentences, as shown in (6). The inherent specification of à as 3P inanimate is crucial again, as it prevents the phi-features of the goal argument from valuing the unvalued φ-feature on V. The φ-feature of the preposition, specified as 3P inanimate, is instead carried up to the PP node through valuation-based Merge,Footnote ¹⁹ and values the unvalued φ-feature on V first. Since the number feature has been valued by the PP, a 3P inanimate theme cannot then enter the derivation, as shown by the failed derivation of (6a) in (27). No matter what the goal’s person is, a 3P inanimate theme is impossible in this structure; this explains the previously reported unavailability of repairs for 1/2/3 > 3 argument combinations.

A question remains as to why the theme must be featurally inanimate – that is, only specified for [#], not [π] – if it is 3P, for those speakers who reject the à-phrase “repairs” for PCC-compliant argument combinations in (6).Footnote ²⁰ I attribute this to the fact that the default interpretation of the theme argument is inanimate. Although this does not fully explain why there is such a restriction, it seems that the restriction is relaxed for some speakers. As noted in section 2.3, there are at least some French speakers that accept the sentences in (6). I claim that for these speakers, the theme argument can be animate (i.e., specified for both [π] and [#]), and this makes it possible for the theme to value the unvalued φ-feature on the verb after it has been valued by a PP specified as 3P inanimate, as shown in (26). Therefore, the present analysis of the PCC and its “repair” accommodates these judgements. This new observation is problematic, however, for Rezac's (Reference Rezac2011) global mechanism, as the double-clitic construction and the “repair” construction are not in complementary distribution for these speakers.

To summarize, in the present analysis, we can attribute all the cases of ungrammaticality to non-generability of the derivation. The double-clitic construction and the “repair” construction are treated independently, but it can still explain the apparent complementarity between them in strong PCC varieties. The “repairs” are not dependent on the presence of a PCC violation in the proposed account, and the apparent complementarity is a coincidence that derives from the feature specifications of the arguments. The following subsection shows that the proposed syntactic mechanism explains even the weak PCC pattern.

5.5 The weak PCC

Section 3.4 introduced the weak PCC varieties of French, in which both the double-clitic construction and the “repair” construction are available for 1P/2P combinations of internal arguments, as shown in (10) and (11). This overlap in availability between the two types of constructions is a challenge for Rezac's (Reference Rezac2011) analysis, as it requires a PLC violation in the double-clitic structure in order to generate the repair structure. Given the grammatical sentences in (10), it appears that there is no PLC violation in these varieties. In this subsection, I show that a simple modification of the unvalued φ-feature used for the strong PCC captures the weak PCC pattern without having to stipulate non-argumental status for one of the arguments, or an additional syntactic operation or condition.

As mentioned above, the difference between the strong PCC and the weak PCC can be explained by a structural difference in the unvalued φ-feature of the verb/Appl. If the unvalued φ-feature can receive both [Addressee] and [Speaker] values as shown in (27),Footnote ²¹ then we can explain the availability of 1/2 > 2/1 clitic combinations in the weak PCC varieties.

For the strong PCC varieties, the unvalued φ-feature (17e) has only one unvalued feature for the two participant features, so it could be exhausted either by a 1P argument or a 2P argument. As shown in (27e), the unvalued φ-feature in the weak PCC languages has both unvalued [Ad] and [Sp] features, which allows the feature to be valued by both 1P and 2P arguments. A partial derivation of the double-clitic construction with the 1 > 2 argument combination is shown in (28). The first Merge operation in (28) results in all dependent features of the unvalued φ-feature of Appl valued except for the [Sp] feature, which can subsequently be valued by the 1P argument through the second instance of Merge. The 3 > 1/2 combinations would still be ruled out as 1P and 2P are more highly specified than 3P, as was shown in (24).

The “repairs” for 1/2 > 2/1 would still be available since the preposition à conceals the person feature of the goal in the “repair” construction, as shown in (25). Therefore, with the syntactic machinery at hand, we can explain the weak PCC pattern without introducing a new syntactic operation.Footnote ²²