Historical Changes in Basque Dative Alternations: Evidence for a P-based (neo)derivational analysis

The properties and internal chronology of various dative changes in the history of the Lapurdian dialect of Basque are shown to be fully incompatible with the basic tenets of standard non-derivational approaches to dative alternations (both Phave and “Low Applicative projection” types), and support the presence of an underlying P in applicative constructions. A neo-derivational approach based on the incorporation of an adpositional head accounts naturally for important generalizations on the distribution of the changes and conforms to the properties of dative variation crosslinguistically.


Introduction
The expression of VP-internal arguments in general and applicative ones in particular shows, at least superficially, a great amount of crosslinguistic, dialectal and historic variation affecting virtually all areas of the syntactic derivation.Naturally, they have always been in the center of much theoretical debate.In Lapurdian Basque (a North-Eastern dialect spoken in the French side of the Basque Country), dative constructions undergo a series of changes in historical times that emphasize the difference between their semantic extension and the syntactic nature of the alternation itself, two questions that standard works on dative alternations usually mix up.The chronology of diachronic changes shows that, although interacting, these questions are very different both with respect to the way they occur and in their extension in time.Semantic spreading is a slow and steady extension of the dative's conceptual space with no syntactic changes associated to it: applicative semantics extends in this dialect way beyond standardly assumed change-ofpossession contexts to a variety of structures covered under the umbrella of unbounded path relations (Etxepare & Oyharçabal 2013).On the other hand, a cluster of very sudden and general changes that occurred later, during the second half of XIX Century, result in the birth of an agreementless dative PP construction with virtually the same semantic extension as the agreement one.The result are triplets like (1a-c) and (2a-c) in Lapurdian that do not exist in the other dialects of Basque, as will be discussed in detail: (1) a. Ama-k semea-ri ogia ekarri dio.mother-erg son-dat bread.absbring aux.(3sA).3sD.3sE'The mother brought the son (the/some) bread.' Glossa general linguistics a journal of Ormazabal, Javier and Juan Romero. 2017. Historical Changes in Basque Dative Alternations: Evidence for a P-based (neo)derivational analysis.Glossa: a journal of general linguistics 2(1): 78.1-39, DOI: https://doi.org/10.5334/gjgl.103b.Ama-k ogia ekarri du semea-ri.mother-erg bread.absbring aux.(3sA).3sEson-dat 'The mother brought (the) bread to the son.' c.Ama-k ogia ekarri du parke-ra.mother-erg bread.absbring aux.(3sA).3sEpark-all 'The mother brought (the) bread to the park.' (2) a. Ate horr-i hurbil-tzen nintzaion.door that-dat approach-hab aux.1sA.3sD.past'I approached that door.' b.Ate horr-i hurbil-tzen nintzen.door that-dat approach-hab aux.1sg.A.past 'I approached that door.' c.Ate horre-tara hurbil-tzen nintzen.door that-all approach-hab aux.1sg.A.past 'I approached that door.' In this paper we analyze the diachronic emergence of the alternation and its theoretical consequences for the different hypotheses on dative alternations.Although the intricacies of the historical data might complicate the discussion, the logic of the paper is rather simple.We are careful to show that in (1) and ( 2) we are dealing with a genuine dative alternation, where agreement dative constructions corresponds structurally to the applicative construction (the structural equivalent of the Double Object Construction in English and other dative constructions in many languages) and both the agreementless dative and the allative are PP structures (English to-construction, etc.).Originally, only (1a) and the allative constructions (1c) and (2c) existed.As we will show, the emergence of structures like (2a) shows that the (DOC-type) dative agreement (applicative) construction is not semantically restricted to change of possession contexts.On the other hand, the late emergence of (1b) and (2b) shows that neither the applicative construction nor the dative PP one are semantically dedicated structures: the (b) sentences systematically show the semantics of (1a) and (2a) but the syntactic structure of (1c) and (2c) respectively; moreover, the details of their emergence indicate that each (a-b) pair in (1) and ( 2) are derivationally connected.
The paper is organized as follows: the next two sections present a description of the phenomena to be analyzed.Section 2 briefly presents the general properties of dative constructions in Standard and Western dialects of Basque (the dialects spoken in the Iberian Peninsula, in the Spanish side of the border).Since Lapurdian shared these common properties in earlier stages, the general description presented in that section is also to be considered, minimal details aside, as the initial stage of the dialects on which the diachronic changes have operated.Section 3 presents the main changes that occurred in the Lapurdian dialect; we base our description on Etxepare (2014) and, especially, Ormazabal (2017).Section 4 shows the impossibility of accounting for the linguistic changes from a non-derivational approach to dative alternation.We also show that these explanations run into serious problems precisely because of the general assumption that applicative constructions are basic, non-derived ones, which makes the proposal fail to explain many important correlations.In section 5 we argue that the distribution of changes in Lapurdian favors a derivational connection between agreementless dative PPs and agreement dative DPs, and we present the details of our analysis following Ormazabal & Romero's (2017a) general proposal, framed within a cross-linguistic perspective on b.Bidaia-n zehar hainbat lagun batu zaizkie.trip-ines through many friend.absjoin aux.(3pl.A).3plD 'Many friends joined them through the trip.' (5) a. Jon-ek Mikel-i besoa hautsi dio.Jon-erg Mikel-dat arm.absbreak aux.(3sA).3sD.3sE'Jon broke Mikel's arm.' b.Jon-ek Mikel-i liburua irakur-arazi dio.Jon-erg Mikel-dat book.absread-cause aux.(3sA).3sD.3sE'Jon has made Mikel read the book.' c.Jon-i liburuak gusta-tzen zaizkio.Jon-dat books.abslike-hab aux.(3pl.A).3sD 'Jon likes books.' d.Semea joan zait Ameriketa-ra.son.abs go aux.(3sA).3sDAmerica-all 'My son went away to America (and it affected me).' As illustrated in (3)-( 5), in Western dialects of Basque datives trigger obligatory verbal agreement.

Structural and semantic properties: Dative constructions
There is plenty of syntactic and morphological evidence that the dative argument agreeing with the verbal complex is a DP and not a PP (Hualde 1986;Elordieta 2001;Albizu 2001;Oyharçabal 2010;Etxepare 2014;Pineda 2014, and references there).There is also general consensus that the canonical hierarchy among the three arguments agreeing with the verb in ditransitive constructions is the one observed in (6a).In particular, quite a lot of arguments have been presented in the literature showing that the dative indirect object in ditransitive constructions c-commands the absolutive direct object (Fernández 1997;Montoya 1998;Elordieta 2001;Arregi 2003a;Oyharçabal 2010; among others).Thus, for instance, the anaphoric direct object in the subordinated clause may be bound by the silent pronominal IO in (6b) but not the reverse (Oyharçabal 2010): (6) a. Subject ergative > Indirect Object dative > Direct Object absolutive b.Jon i ez zen ohartu pro i bere.buruai aipatzen niola.Jon.abs not aux realize reflex.absmention aux.(3sA).3sD.1sE-comp'Jon i didn't realize that I was mentioning himi himself i .' c. *Jon i ez zen ohartu pro i bere.burua-rii aipatzen niola.Jon.abs not aux realize reflex-dat mention aux.(3sA).3sD.1sE-comp'Jon i didn't realize that I was mentioning himselfi him i .' Concerning semantic effects traditionally discussed in the literature associated to applicative constructions, with verbs of the send-(7a), throw-(7b), give-(7c) and teach-types (7d) dative DPs show animacy effects in Basque, as in other languages.(7) a. *Jon-ek Kutxi kalea-ri eskutitz bat bidali dio.Jon-erg Kutxi street-dat letter one.abssent aux(3sA However, as we have argued elsewhere (see Ormazabal & Romero 2010, 2017b and references there for a more extensive discussion), this animacy restriction is a property of a certain subset of dative-taking predicates only.Typically the beneficiary and the goal must be animate, but languages with a richer range of applicative constructions than English, including Basque, often include predicates that allow non-animate datives: Jon-erg chair-dat leg.abs fix aux.(3sA).3sD.3sE'Jon fixed the chair's leg.' c.Jon-ek liburua-ri hitzaurrea kendu dio.Jon-erg book-dat preface.abstake.outaux.(3sA).3sD.3sE'Jon took away the preface from the book.'

Morphological properties: Applicative constructions
As described in many previous works (Trask 1997;Albizu 1998;Hualde & Ortiz de Urbina 2004;Rezac 2008;Etxepare & Oyharçabal 2013;Ariztimuño 2013;Etxepare 2014;among others), the presence of a dative argument in a Basque finite sentence is associated to three morphological characteristics: (i) a dative suffix shows up in the agreeing DP; (ii) person and number agreement appears in the auxiliary or the synthetic verb, and (iii) in the case of synthetic verbs a dative flag is inserted, a morpheme in the position immediately preceding dative agreement that indicates the presence of an applied argument.That is the general situation in Western dialects of Basque, as well as in Standard Basque, and it is also the basic state of affairs in previous stages of the Lapurdian dialect in the Northeast of the Basque Country we analyze in the following section.These dialects undergo a series of syntactic changes that have brought some amount of attention in recent Basque studies.Specifically, Lapurdian makes four basic innovations: i) new semantic relations are assigned to applicative constructions, ii) dative agreement becomes optional in most contexts, iii) there is a concomitant change in c-command relations and, up to some point, word order, and (iv) there is a process of specialization pragmatically driven.

Datives in the Lapurdian dialect
In this section we present the historical changes related to dative constructions that occurred in the Lapurdian dialect. 3Etxepare & Oyharçabal (2013) and Etxepare (2014) describe the main contexts where the changes occur in detail, but they mistakenly analyze these changes as chronologically simultaneous processes that constitute a single change.The reason is that most of their sources are late examples from a period -late XIX and XX Lapurdian authors and Oyharçabal's native speaker judgements of the same dialectwhen all the main changes have already taken place.Other works do make some more accurate observations concerning historical aspects of the different processes involved but, as far as we know, none of them tries to make the relative chronology of the different (micro)changes explicit.As we argue in sections 4 and 5, the details of that chronology are central to the theoretical understanding of what dative alternations stand for crosslinguistically.
In this section, we follow Ormazabal (2017), where a detailed dating of the changes is made, and both the progressive spreading of the dative in locative contexts and the relative chronology in connection with the rest of the changes are analyzed.Ormazabal (2017) systematically studies all the relevant verbs and verb-classes mentioned by Etxepare & Oyharçabal (2013), and manually analyzes all their occurences in all Lapurdian texts included in the Euskal Klasikoen Corpusa (Classic Basque Corpus, http://www.ehu.eus/ehg/kc/) -the most complete corpus on Basque classical texts to date-.We complete our analysis with examples from the references mentioned in footnote 3, as well as some observations in Mitxelena/Sarasola's (1987-2005) Diccionario General Vasco-Orotariko Euskal Hiztegia, the closest to a Basque historical dictionary to our date. 4ection 3.1 describes the expansion of the dative suffix to include the marking of spatial functions of different sorts that are not possible in the other dialects of Basque.Contrary to what has been assumed in the literature (Pikabea 1993;Etxepare & Oyharçabal 2013); this expansion is shown to be a slow process that initiated earlier than the other changes discussed in this paper and extends up to our days.As a result of this expansion, there is a semantic reorganization of the space between locative adpositions, mostly the allative, and new locative (unbounded path) and aspectual uses of the dative.But this new distribution of the cognitive space has no associated effect on the morphological or syntactic behavior of the applicative construction (the agreeing dative construction) and the locative postpositions of the language.
In section 3.2, we discuss the other changes, which behave as a cluster of which the optional loss of dative agreement in the auxiliary is the most salient one.As will be shown, this change occurs very fast, extending to the old dative contexts, including change of possession ones, as well as to the new unbounded path datives that resulted from the previous semantic expansion.Concomitantly, the hierarchical structure, as well as the categorial status of dative arguments also change.The resulting picture is a dative alternation where the agreeing applicative and the PP dative constructions have the same semantic extension but different structural properties: agreeing datives are clear DPs and show the standard unmarked order and the IO dative > DO absolutive c-command relations; in contrast, in the agreementless construction the absolutive DO c-commands, and tends to precede, the dative-marked element, as in other (spatial, path, etc.) PP constructions.
Section 3.3 briefly discusses a more recent innovation in these dialects that reduce the appearance of applicative constructions in some information structure driven contexts, and section 3.4 wraps up the main descriptive results.

Extensions of the dative to new contexts
In Lapurdian the dative expands to mark spatial goal of the event where only spatial postpositions or complex postpositional phrases are possible in Western dialects (Etxepare & Oyharçabal 2013;Etxepare 2014).Compare, for instance the use of dative in Lapurdian examples like (10) with the use of allative or complex postpositions in Western Basque in the same contexts ( 11): (10) a. Leizarraga (1571: Lukas VII 12) eta hiri-ko portalea-ri hurbildu zaion bezala and city-gen door-dat approach aux.(3sA).3sDas 'and as he approached the door of the city' b.Larzabal (1991-98: VII 53) Aleman-en tankak oldartzen zire-la Maginot harresia-ri.German-gen tank.abscharge aux.(3pl.A)-comp Maginot fence-dat 'As the German tanks charged against the Maginot line.' (11) a. eta hiri-ko portale-ra hurbildu den bezala and city-gen door-all approach aux.(3sA) as 'and as he approached the door of the city' b.Aleman-en tankeak oldar-tzen zire-la Maginot German-gen tank.abscharge-hab aux.(3pl.A)-comp Maginot harresia-ren kontra.fence-gen against 'As the German tanks charged against the Maginot line.' The change consists in a reorganization of the semantic field of spatial adpositions and the dative construction, a process that is common to many other languages.In particular, Etxepare (2014) proposes a partition in the set of Path exponents in Lapurdian Basque dialect between the allative and this directional dative, roughly as in (12):5  Jainkoa-ren nahi saindua-ri erortzen diren arima jenerosak.God-gen will holly-dat fall-hab aux(3pA).compspirit generous.pl'Those generous spirits who are inclined towards god's holly will.'A careful analysis of the changes shows that this extension of the dative to new semantic contexts is not associated to any change in the syntactic and morphological properties of the dative construction in these dialects: i) dative agreement with the auxiliary is still obligatory: in "old" applicative constructions (3) and in "new" unbounded path ones (10) agreement is treated alike in all Lapurdian texts from XVI to middle XIX centuries.Moreover, (ii) the dative morphology of the language, in the nominal argument or in the auxiliary, is not affected by the extension of the dative to new semantic contexts.(iii) The category of the new unbounded path is the same as in the "older" ones: they are agreeing DPs.And (iv) the general Case and agreement relations in the language are not affected by the new lexical extensions either.In conclusion, they all uniformly behave as applicative constructions.On the other hand, allative -ra (bounded path 'to') continues to show the same postpositional properties as in previous stages of the language.
Before providing a chronology of the changes, it has to be noted that, contrary to what has been suggested in the literature (Pikabea 1993;Etxepare & Oyharçabal 2013;Etxepare 2014), the extension of the dative does not co-occur with the optional loss of agreement, a change that will be introduced in the next section.Dative extension begins much earlier and constitutes a slow and progressive process of lexical/semantic spreading of the applicative construction.It was initiated in the XVI century, and it continued its expansion to new semantic contexts (almost) until our days (see Ormazabal 2017 for a detailed chronology and discussion).Thus, lexical directional verbs like hurbildu ('approach'), itzuli ('turn towards') (13), and atelic verbs in their aspectual use with event-denoting dative complements lotu ('start'); jarraiki ('continue to'), abiatu ('begin') (14), are already attested with inanimate dative arguments in XVI and XVII century texts respectively.

Optional agreement, hierarchical relations and word order
Unlike the applicative's semantic extension, the other changes observed in the Lapurdian dialect come together, and very rapidly extend virtually through the entire range of dative constructions.Contrary to Western dialects, where agreement was -and still isobligatory, in Lapurdian dative agreement becomes optional.Consequently, while (17a) is grammatical in all dialects, (17b) is only available in North-Eastern (NE) ones. 66 In Low-Navarrese and Souletin dialects, also in the French side of the Basque Country, agreementless datives appear since the beginning of the written records.There is broad consensus among both diachronists and generative grammarians that the Lapurdian agreementless system and the Low-Navarrese and Souletine ones constitute two different processes, both structurally and chronologically (see Ortiz de Urbina 1995; Mounole 2011; Ormazabal 2017; and references there).For instance, an anonymous reviewer reports example (i) to us, in which c-command relations in early Low-Navarrese agreementless constructions are the opposite to the ones of Lapurdian to be discussed next in the text: (i) Etchepare (1545: 265) (reviewer's reference and translation) Eman dezan iujiak nor-i beria give aux.subjunct.(3sA).3sEjudge-erg whom-dat his.abs 'As for the Judge to give each one his due' The chronological gap and structural differences, as well as Ulibarri's (2015) report of agreementless dative cases in early Araban/Biscayan texts (in the other extreme of the Basque Country), and Blanca Urgell's suggestion (p.c.) to treat them as syntactic archaisms, strongly support the existence of two independent systems.If all this is correct, the Low Navarrese case might be considered the remains of an older general system where no dative agreement morphology was present, in accordance with most diachronists' view nowadays that consider the development of agreement morphology in the verbal complex as a relatively late phenomenon in the prehistory of the language.
(17) a. Ama-k semea-ri ogia igorri dio.mother-erg son-dat bread.abssent aux.(3sA).3sD.3sE'The mother sent (the) bread to the son.' b.Ama-k ogia igorri du semea-ri.mother-erg bread.abssent aux.(3sA).3sEson-dat 'The mother sent (the) bread to the son.' Morphologically, agreementless constructions involve loss of the entire dative-marking system: verbal dative agreement, dative flag in synthetic verbs, and selection of the nondative auxiliary.Agreementless constructions (17b) resort to the same auxiliary form du as regular transitive verbs such as ikusi ('see') in ( 18 As some examples will illustrate later, quite often this hierarchical change also correlates with a change in the linear order with the indirect object in postverbal position. 8In sum, the new structure has all the properties of a PP construction.
As for the animacy effects, the same group of predicates that show animacy effects with agreeing datives also show the same effects with agreementless ones, an important observation: (21) a. *Ama-k merkatua-ri ogia eman dio.mother-erg market-dat bread.absgive aux.3sE.3sD.(3sA)'The mother gave the market (the) bread.' b. *Ama-k ogia eman du merkatua-ri.mother-erg bread.absgive aux.3sE.(3sA)market-dat 'The mother gave bread to the market.' From a diachronic perspective, some of the properties of this change will also be important for our discussion: the new agreement/agreementless alternation initiates much later than the reorganization of the semantic fields, and systematically expands to the entire spectrum of dative constructions, old as well as new ones, in a very short period of time.Concerning "old" datives, ditransitive and unaccusative constructions appear in agreementless contexts from the second half of XIX century on.The pairs in (22a-b) and (23a-b) show that both agreement and agreementless constructions coexist, a situation maintained until our days: Ez diote soldadu-ek min-ik egin-en.not aux.(3sA).3sD.3pEsoldiers-erg pain-part do-fut 'The soldiers will not cause pain to her.' 8 Descriptively speaking, Basque is a "free word order" language and, as could be expected, both agreement and agreementless dative constructions show all possible S-DO-IO orders.In particular, an anonymous reviewer reports to us examples like (ia-b), both from Larzabal, where the word orders revert the ones corresponding to the ones in ( 22)-( 23 Ez uste … baitezpada gorrieri emango dudala ene botza Not think… just.in.case red.datgive.futaux[(3sA)-1sE] my vote 'Do not think I will give my vote to the reds just in case' However, it has often been observed that agreementless datives do have a tendency that agreement datives do not show to appear in postverbal position (Albizu 2001;Etxepare andOyharçabal 2009, 2013).It is not obvious how much could be concluded from that fact, because there are too many factors to control for and a systematic analysis would be needed to yield more substantial conclusions, but the word order tendency is consistent with all the other properties that show that the DO is structurally higher than the IO in these constructions.
The other group of exceptions are obligatory agreement contexts, very specially, experiencer (28a) and possessor (28b) datives, which never ever show up in the agreementless construction (Fernández & Landa 2009 For most authors, the structures in (28) correspond to "high applicatives", in the sense of Pylkkänen (2008).Following some authors (McFadden 2004; Fernandez & Ortiz de Urbina 2010 and refereces there) we will use the more neutral term "high dative" as a descriptive umbrella, and postpone the discussion of these structures until the last section of the paper.

Information-structure oriented specialization
This cluster of changes is completed with a process of specialization in Etxepare's term, the well observed fact that whenever the agreement/agreementless alternation holds -and only in those contexts -dative constructions show important restrictions constraining the presence of low referentiality elements (e,g., anaphora, Negative Polarity Items) in applicative structures.As Etxepare observes, these elements "tend (overwhelmingly) not to agree" (examples from Etxepare (2014:  This specialization mechanism, which Etxepare attributes to a third factor in language design, has been observed to affect dative alternations in a large range of languages, and has important theoretical consequences.Two important properties must be observed in relation with this condition.First, that it only restricts applicative constructions (agreement datives in Basque, DOCs in English, etc.) when there is a genuine alternating dative PP construction in the language.This is clearly shown by the fact that in those dialects of Basque where agreement is obligatory, including in stages of Lapurdian previous to the appearance of agreementless dative PPs, NPs or anaphoric elements show up in agreeing dative position naturally.Moreover, Lapurdian NPIs, etc. have no problem to show in the subject and direct object positions, where agreement is obligatory and, consequently, no alternation exists.The second important fact is that the condition does not induce ungrammaticality, and anaphora and NPIs may show up as agreement dative DPs also in Lapurdian.However, the referentiality effect affects speakers' preferences depending on pragmatic contexts, with drastic statistical consequences in the number of occurrences each alternate presents, an important factor that might be the trigger to new diachronic changes in the future.See, among many others, op.cit.as well as Etxepare & Oyharçabal (2013); and Bresnan et al. (2004); Bresnan & Nikitina (2008); Rappaport- Hovav & Levin (2008); Ormazabal & Romero (2010;2017b); Anttila, Adams & Speriosu (2010); Levin (2015), and references there for details and discussion.

Summary
Summarizing the properties of the changes: i) The reorganization of the semantic field of paths with the resulting spreading of the datives is a slow process that extends the dative domain to locative, semelfactive and aspectual contexts, semantically very distant from the initial change of possession contexts and the like.This semantic change is not accompanied with structural differences.The extension to the new contexts has no effect on the inflectional properties, the syntax or the morphology of the dative construction itself, which remains structurally the same: an applicative construction.ii) The rest of the phenomena constitute a cluster of changes that occur later and extend very fast to all dative contexts (with the notorious exception of high datives).iii) The result of these changes is a genuine dative alternation, an applicative/PP alternation similar to the DOC/to-construction of English and their equivalents in other languages.iv) Like DOC/to-constructions and other dative alternations in many other languages, the agreement/agreementless distinction correlates with two different categorial and c-command structural relations.v) Crucially, the syntactic distribution of the two alternates does not correspond in any way to any semantic difference: both structures cover virtually all the same semantic contexts, and the same animacy effects show up with the same predicates in both agreement and agreementless constructions equally.
In the next two sections, we argue that these results strongly support a derivational analysis of dative alternations over a non-derivational one.Section 4 is dedicated to argue that this state of affairs is at odds with the two most prominent non-derivational analyses of dative alternations proposed in current linguistic research in general and in Basque syntax in particular: the P have analysis and the "Low Applicative projections" one.In section 5 we argue that (neo-)derivational analyses are well fitted to cover the main properties of the changes in dative constructions described so far in a natural way.

Alternate Underlying Configuration Hypothesis (AUCH!) approaches do not work
This section analyzes how the two main non-derivational analysis of dative alternations deal with the changes in Lapurdian Basque.9First we show in section 4.1 that a Harley ( 2002)-type P have analysis is incompatible with the facts, because the changes are not about semantically dedicated structures, but about syntactic configurations.A Pylkkänen (2008)-type Applicative Phrase analysis void of any semantic content, on the other hand, would not yield contradictory results, but even in that case it runs into serious problems and fails to even accommodate the central properties of the changes involved (section 4.2).

Dative alternations are not about semantically dedicated structures
A long tradition in the analyses of dative alternations in many languages (Green 1974;Oehrle 1976;Jackendoff 1989;Pinker 1989;Harley 2002Harley , 2004;;Krifka 2004  ), this proposal is highly implausible not only for Basque but in more general terms.
To begin with, give-, tell-or promise-type Vs do enter the alternation, despite the fact that they are never found in the caused motion event schema.The pair in (32) shows that the same transfer of possession may be realized in the dative agreementless and agreement structures: (32) a. Webster (1877: "Soldadu pobre bat eta aberatsa") gañerateko diruak ematen ditu pobree-ri.remaining money.plgive.habaux.3pA.3sEpoor.pl-dat'he gives the rest of the money to poor people.' b.Hiriart Urruti (1891-1914: "Narrazio bat") ni-k ez dautzut gehiago dirurik emanen ura-ri behera I-erg not aux.3sA.2pD.1pEmore money give.futwater-dat down arthiki-tzeko.throw-nom 'I won't give you any money again to throw it down the drain.' We may be sure that these predicates are in a caused transfer of possession frame, and not a caused of motion in some abstract sense, because in both agreement and agreementless dative constructions (examples in (21), repeated here in (33)) they show animacy effects, a diagnostic of cause of possession, according to AUCH! analyses: (33) a. *Ama-k merkatua-ri ogia eman dio.mother-erg market-dat bread.absgive aux.3sE.3sD.(3sA)'The mother gave the market (the) bread.' b. *Ama-k ogia eman du merkatua-ri.mother-erg bread.absgive aux.3sE.(3sA)market-dat 'The mother gave bread to the market.' In fact, the distribution of allative -ra and the dative-assigning P in Basque constitutes indirect support for Rappaport -Hovav & Levin (2008) and Ormazabal & Romero's (2010) argument that the English to-construction corresponds to two different semantic frames: the "caused transfer of possession" and the "caused motion" ones.In particular, all the verbs that enter the DOC/to-PP alternation in English (Gropen et al. 1989;Levin 1993), including give-, throw-and send-type verbs enter the agreementless dative construction in Basque, alternating with dative agreement constructions with the same meaning.However, only a subset of these verbs -those that are really compatible with a "caused motion" frame -allow complements headed by the allative postposition -ra.As we might expect, the complement of these verbs show the same animacy requirement when they are datives -in both agreement and agreementless contexts -but not when they are allative PPs.It is important to note that there is no doubt that we are dealing with a genuine dative alternation in these dialects.Remember that although the agreement and the ditransitive agreementless dative constructions structurally correspond to the DOC vs. to-construction -reversed c-command relations, etc. (see details in section 3.2, above) -they express exactly the same semantic content.In other words, the interpretation and historical evolution of agreement and agreementless dative constructions in Lapurdian Basque not only do not justify an approach in terms of semantically dedicated base structures but, in fact, they constitute a strong argument against that hypothesis.

"Applicative" analyses fail to capture the properties of the diachronic changes
The other main non-derivational family of analyses for dative alternations, the "Low Applicative" (LowAppl) vs. PP-structure proposal (Marantz 1993;Anagnostopoulou 2003;Cuervo 2003;Pylkkänen 2008;Bruening 2010b;Fernandez 2010Fernandez , 2014;;Oyharçabal 2010;Etxepare & Oyharçabal 2013; Etxepare 2014, among many others), is somehow in a better shape to deal with Basque data.However, the LowAppl head must be void of the transferof-possession relational semantics that most applicative analyses attribute to it; otherwise, this approach runs into the same problems discussed in the previous subsection.Let us thus assume that to be the case. 12Even in that case, the analysis fails to explain the properties of the changes altogether, as we show next.
11 For discussion of idioms and their relevance for the analysis of the ditransitive alternation, see Ormazabal & Romero 2012, 2017b; Larson 2017 and references there."Fixed theme idioms", where the empty element that may vary inside the idiom is the indirect object (e.g.give [somebody] a headache) have been traditionally associated to DOCs, as opposed to "fixed goal idioms" (e.g.send [somebody] to the wolfs), which have been said to be associated to the to-construction exclusively (Bruening 2010a and references); but see Bresnan & Nikitina 2007; Rappaport Hovav & Levin 2008; Ormazabal & Romero 2010, 2012.  1Incidentally, a LowApplP analysis where the two structures are made independent of the semantic interpretation gets very close to a "compatible frame" approach of the type proposed by Rappaport- Hovav & Levin (2008) in the framework of lexical semantics.The first innovation in Lapurdian, in consequence, would be the extension of this low applicative structure to cover new semantic fields -locative and aspectual constructions, semelfactives, etc. -that previously were covered only by adpositions -mostly allative, but also inessive, destinative, and instrumental.As observed, this extension does not affect syntactic relations or agreement morphology, although some ontology has to be provided to restrict the semantic scope of the construction to exactly those contexts previously specified only by means of adpositions.Thus, we may assume that such an analysis would maintain the same applicative structure in (38b) or similar for the newly created agreeing datives such as (10a), repeated in (39), and all the extensions discussed in ( 10)-( 15) in section 3.1, above, with no substantial modification.
(39) eta hiri-ko portalea-ri hurbildu zaion bezala... and city-gen door-dat approach aux.(3sA).3sDas 'and as he approached the door of the city...' [=(10a)] The second innovation, agreementless datives, is the result of adding to the lexical inventory of Lapurdian a new adposition modeled on the basis of the Romance dative preposition à ('to'), following Etxepare's (2014) insight. 14This new adposition exhibits a semi-functional nature to cover exactly the same semantic contexts as the extended applicative construction: unbounded path (≈'towards'), as proposed in Etxepare (2014), and all the uses of the dative in the language, including the goal in change of possession frames, etc .
Restricting our attention to the projection of arguments in the verbal complex, the new agreementless datives in (40a) would have the same PP-structure as the allative in (40b), represented in (41) (modified from Etxepare & Oyharçabal 2008).Remember that the only difference between allative and dative structures in ( 41) is the postpositional head: a phonologically null postposition P 1 (≈'towards') that assigns dative to its complement DP in (40a), and an overt allative one -ra ('to') in (40b), with a bounded path semantics: 14 The extension of the semantic fields covered by the dative in these dialects has been attributed to language contact with continental Romance, either French (Pikabea 1993; Etxepare 2014) or, most probably, Gascon (Ormazabal 2017).It is worth mentioning that if that influence is correct, in fact the new semantic fields are borrowed from a PP construction in the Romance language and projected as an agreeing LowApplP-type structure in Basque, by hypothesis a very different structure.That is so because when the semantic extension occurs the agreementless dative PP is not available as a possible structure in the language.See footnote 17 below for a possible alternative within the LowAppP analysis. 15Again, details of the internal structure of the PP are irrelevant for the discussion.That includes the issue of whether the direct object is projected as an argument of V, as represented in ( 40) and ( 42), or the two DP arguments constitute a small clause structure.Etxepare & Oyharçabal (2013) and Etxepare (2014) extensively argue that the dative is not the prepositional element but the case assigned by the presence of a directional, unpronounced postposition.They also argue that the two PPs also differ in their internal structure, which is more articulated in the case of the allative adposition.We follow their analysis of the internal structure of the PPs, but nothing important depends on this assumption.See references for extensive discussion of axial parts in PP structures.
As said before, the same parallel structure would have to hold between the allative (-ra) and the dative assigning null P 1 in transitive pairs like (42), this time with the addition of an external argument in the vP.In this case, the null P 1 would be interpreted differently: in ( 40)-( 41) it is interpreted as "unbounded path" ('towards'), while in ( 42)-( 43) it receives a transfer-of-possession interpretation: (42) a. Ama-k ogia igorri du semea-ri.mother-erg bread.abssend aux.(3sAIn this framework, the well-observed syntactic and morphological differences between agreement and agreementless dative constructions are structurally encoded in the two constructions proposed: the agreement asymmetry follows from the fact that an agreement relation is postulated in the applicative phrase in (38) but not PP-internally in ( 41) and ( 43).In addition, the two structures are also intended to reflect the differences in the categorial status of the two internal arguments (DP vs. PP) and in the hierarchical relation among them (IO > DO vs. DO > IO) discussed in section 3.2 and commonly observed in dative alternations in many languages since Barss & Lasnik's (1986) discussion of Double Object Constructions in English.What is not at all obvious is how the tight relation that the two constructions manifest both language internally and crosslinguistically may be captured in such a non-derivational approach, as we discuss below.
Together with these structures, ApplP analyses (as well as P have ones) must also account for the lack of alternation in two other types of constructions: (i) high datives (experiencer, possessor etc.), and (ii) dative assigning postpositions.Concerning the non-alternating high datives, most authors in the literature follow Pylkkänen's (2008) and Cuervo's (2003)  Regarding the case of complex postpositions (cfr.English in front of, up to, etc.), the minimal pair in (45) illustrates that in many such structures there is a slow process of substitution of the genitive, locative or instrumental postpositions by dative assigning ones.As argued by Etxepare & Oyharçabal (2013) and Etxepare (2014), all these constructions encode directional paths. 16Thus, (45b) and the like constitute natural extensions of the dative to unbounded path domains, but they are exceptions to the agreement/agreementless alternation: the dative has no possible way to show agreement internal to the complex PP-construction in Basque.The relevance of these structures is due to the fact that they appear when dative verbal agreement is still obligatory in the language, before the emergence of agreementless dative PPs of the type in ( 40)-( 45).Consequently, an ApplP analysis has two ways to incorporate these changes in the system, both of them far from adequate.(i) A LowApplP might be postulated within the complex PP, a rather implausible and stipulative move.(ii) Alternatively, what looks as a natural extension of the semantics of unbounded path datives already existing in the verbal system would be in fact the birth of a completely different PP headed by a dative assigning P within the complex PP structure (46).Notice, however, that this structure will not be generally available until a few generations later, when agreementless datives appear elsewhere in the dialect: 16 Complex postpositions are invariant postpositional complexes where the main element may be of different origin: "borrowings from Romance prepositions (Spanish contra 'against') or nouns (Spanish campo; Gascon land/lande 'open space'); most are derived from native locational nouns that have lost their autonomy as nouns, and mostly occur as a frozen part of the morphologically complex postposition.This is the case of gora 'up', behera 'down', barna 'into', and zehar 'across' in the sample.Gora and behera also function independently as adverbs."(Etxepare & Oyharçabal 2013) In addition to that, the system described in this subsection is not very suitable for theoretical reasons: even though the core three-way distinction described here could be taxonomically appropriate, it lacks explanatory power.Postponing the discussion of high applicatives until section 6, if we are dealing with different and independent structures, some obvious issues arise with LowApplP-type answers to the main questions posed by dative structures: (i) As mentioned above, the extension of the agreement dative construction far beyond caused-possession contexts to pure locative and aspectual ones diminishes Pylkkänen's semantic basis for the LowApplP and requires a very lax interpretation of the functional Appl head, in clear contrast to similar heads such as v.At the same time, some mechanism that restricts the meaning of the applicative is necessary (see Wood 2012;Wood & Marantz 2015 for a possible way out).Interestingly, the semantic interpretation of the applied argument would still depend completely on the selecting lexical verb.(ii) The sudden birth of a dative PP, which appears later, and, especially, its quick extension through the entire system is a complete mystery also: a) From a diachronic perspective, a big part of the dative's semantic extension to cover unbounded path contexts is chronologically previous to the optional loss of dative agreement, the opposite to what we would expect by combining a borrowing hypothesis perspective with a non-derivational view of dative alternations.That is, the use of datives in unbounded path contexts appears earlier in applicative (obligatory agreement) structures (37), than in dative selecting P 1 structures (40)/(42), in contradiction to the idea that the agreementless structure is motivated by the introduction of a new lexical item by influence of French preposition à ('to').17b) Moreover, what makes the move even more suspicious in this case is the fact that the agreementless dative encompasses exactly the same contexts where Low Applicative constructions were -and still are -an option.Both the agreement and the agreementless dative strategies, in principle two completely different structures, end up extending their domain to exactly the same semantic fields.Again, the non-derivational analysis would have to abandon important theoretical positions generally assumed in these frameworks to accommodate the facts with little explanatory gain, beyond a simple statement of the facts.
Note that there is nothing strange in either (i) or (ii) as diachronic processes: throughout redistribution of the semantic landscape among adpositions and between adpositions and dative is very common in both directions in the languages of the world, many of them similar or even almost identical to the ones occurred in Lapurdian.In fact, "emptying" of the dative in favor of existing or new adpositional elements or vice versa has been a well studied process in typological and diachronic studies. 18But ApplP/PP analyses have no saying on why and how that happens, other than stipulating their properties in the different structures.
The complementarity of the issues raised reinforce the position that there is a structural connection between the dative agreement and the agreementless PP strategy that the AUCH! and the LowApplP hypothesis are unable to capture.The hypothesis that the changes we are discussing are the primary consequence of a borrowing from a continental Romance language may be correct given, among other things, the geographical distribution of the phenomenon. 19But, given the facts, that is only possible in a context where the two datives are structurally connected.As far as we can see, that reduces the possible options to two basic ones: a derivational analysis based on P-incorporation and a "compatible frame" one based on an underlying PP-substructure common to the two alternating structures.In section 5 we argue that a derivational P-incorporation approach accounts most naturally for the distribution of facts if "Low Applicatives" are interpreted as derived from underlying PP structures.At the end of that section we briefly discuss a "compatible frames" alternative that also keeps the connection of the two structures but does not link them derivationally.Then, in section 6, we speculate on a different structural way to integrate so-called "High Applicatives" into the system that could explain why they do not alternate.

P-based analysis of dative alternations: Not birth of a new strategy, but recycling of existing ones
An important conclusion of our previous discussion is that we are dealing with two main changes very different in time and nature: (i) A steady process of semantic spreading of the agreement dative structure, which starting in the XVI century slowly extends to different locative and aspectual contexts throughout the entire historical record.(ii) A cluster of morphosyntactic changes -optional loss of dative agreement, inversion of c-command relations, and changes in the categorial status of datives and partially in word order -that occurred in the second half of XIX century.
We assume these changes are the manifestation of a single, more basic, change 18 Descriptively speaking, the process is the mirror image of the one attested in the change from Latin dative to Romance prepositional phrases headed by Lat.ad (Romance a/à); in that case, it is the directional preposition ad ('to') that expands to cover grammatical relations previously covered by the dative: goal, benefactive, etc.For a detailed analysis of the change from Latin to Romance see Fedriani & Prandi 2014, and references there.A similar process occurred also in other Indo-European languages (see, e.g.Eng.extension of the path preposition to).For a more general typological discussion, see Kahr (1975Kahr ( , 1976)), Luraghi (2003); Creissels (2008); Lambert (2010) and references. 19Given that part of the borrowing has already taken place in the XVII century, there are more plausible candidates than French as the source language.Until XVIII century, with the French Revolution, French does not penetrate in a systematic way in these areas, not even as a koiné language, since Béarnese, a dialect of Gascon, was the general language for administrative purposes and most probably for trade relations.According to Coyos (2012), most Basque speakers of Eastern dialects, at least in Soule, were bilinguals in this language as well.Moreover, the influence of Gascon extends further, since it may be tracked at least in one of the complex postpositions with dative in the construction -ri buruz, a calque of (de) cap a La casa 'towards the house', lit.'(of/with) head to the house', as already pointed out by Mitxelena/Sarasola (1987-2005) and Hualde (2002).See footnotes 14 and 16, and references there for discussion.
in the system yielding the sudden appearance of the agreementless dative construction.This new PP-construction creates a genuine, literal, dative alternation where both alternates cover virtually the same semantic fields, and compete with other lexical Ps in the distribution of the semantic space.
Our main point in this section is to show that the changes in (i) fit most naturally in a P-incorporation analysis of the type proposed in Ormazabal & Romero (2017a), and that such an analysis accounts for the cluster of properties associated to the changes in (ii) and for the resulting state of affairs.

Theoretical frame: Crosslinguistic variation in PP and applicative constructions
In Ormazabal & Romero (2017a) we argue that a derivational analysis of PP and applicative alternations involving P-incorporation provides a general framework to explain the rich superficial variation in dative constructions, and at the same time it constitutes a good model of how parametric variation may be formally constrained.We argue that PP and dative constructions share the same basic structure, a small clause-type structure headed by a P element, where the DO c-commands the IO, as represented in ( 48) for (47b, c): 20 (47) a. Ama-k semea-ri ogia igorri dio.mother-erg son-dat bread.abssend aux.(3sA).3sD.3sE'The mother sent bread to her son.' b.Ama-k ogia igorri du semea-ri.mother-erg bread.abssend aux.(3sA)-3sE son-dat 'The mother sent bread to her son.' c.Ama-k ogia igorri du merkatu-ra.mother-erg bread.abssend aux.(3sA).3sEmarket-all 'The mother sent bread to the market.'(48) g merkatu(a) (the market') -ra (allative) semea-ri ('the son' + dat) -P1 (dative-assigning postposition) The second claim in Ormazabal & Romero (2017a) is that applicative constructions 21 share the same basic derivation crosslinguistically, and involve the incorporation of the P 20 As in the previous sections, we will not consider issues concerning the internal articulation of a richer PP structure.As far as we can see, it does not affect our argumentation in any important way.See footnote 15 and references.head, an operation that triggers the derivation of the dative constructions and gives rise to the main surface differences in the agreement, word order and hierarchical structure. 2249) vP (=AGRoP) It is important to underlie that our position is not that examples in (47a-b) have a common derivation with the allative construction in (47c), but that they all share the same baseconfiguration (48).The PPs are headed by different postpositions, which have different lexico-functional specifications and different semantic properties, including their selectional features (see Ormazabal & Romero 2010, 2017a and references for discussion).In fact, from a derivational point of view there is no real dative alternation in Basque until the appearance of the agreementless dative.According to us, this is the new situation in which the incorporation of P becomes optional in these varieties, yielding (47a) and (47b) as we will see next. 23, 24 The general answer to the great crosslinguistic variability in dative alternations is that the only variation specific to the dative constructions resides in the properties of the incorporated P.More specifically, we argue that crosslinguistic variation reduces to i) whether the incorporated P 1 (the applicative head) is manifested morpho-phonologically in the verbal complex or not and, if so, how; 25 ii) the obligatoriness/optionality of the incorporation; iii) the Case/agreement feature-specifications of P 1 and v and, perhaps, iv) the set of incorporating Ps in each language.We argue that these specifications interact with the particular mapping between the conceptual space and the linguistically relevant lexical entries -most characteristically in the verbal and adpositional domains -and with regular Case and agreement resources in the system of the language.These combinations have the effect of making dative constructions look superficially very different across languages despite their common derivation and their well rooted similarities.
Considered from a synchronic point of view, Lapurdian and SW Basque are just two dialectal varieties that conform to the narrow possibilities of parametric variation described above and differ minimally in the lexical distribution of path postpositions and in the value of parameter (ii): P 1 obligatorily incorporates in Western dialects and optionally in North-Eastern ones .But the diachronic changes that yield this state of affairs are particularly helpful to clarify the syntactic articulation and the different locus of the parametric differences.

Changes in Lapurdian reconsidered
Leaving minor changes aside, the initial stage in the Lapurdian dialect corresponds to the current situation in more conservative Western ones, where the semantic extension of the dative is already quite broad, as described in section 2.1, but has not extended to the unbounded path and other contexts discussed in section 2. The latter is filled by lexical path postpositions such as the inessive -ra ('to'). 26The Basque abstract postposition that yields the dative construction incorporates obligatorily in Standard and Western Basque, and that was also the situation in Lapurdian.Consequently, dative agreement was obligatory (50a).In contrast, the allative postposition -ra ('to') never incorporates in any dialect and, consequently, never shows agreement with the auxiliary (50b).
(50) a. Ama-k semea-ri ogia igorri dio /*du.mother-erg son-dat bread.abssend aux.(3sA).3sD.3sE/aux.(3sA).3sE'The mother sent her son (the) bread.' b.Ama-k ogia igorri du merkatu-ra.mother-erg bread.abssend aux.(3sA).3sEmarket-all 'The mother sent (the) bread to the market.'Chronologically, the first innovation in Lapurdian is the redistribution of the semantic fields, in the direction proposed by Etxepare (2014): the dative extends to express 'unbounded path'. 27We assume that this change, induced by external factors in a context of language contact, basically reduces to a redistribution of semantic fields non-incorporating agreementless datives.To finish, Differential Object Marked arguments, which are dative agreeing DPs, would also remain unexplained, as the reviewer observes.
All these factors make the role of the dative flag opaque in synchronic terms, and it does not seem that it may be accomodated well in any of the many hypotheses of dative alternation in competition.That is not the desired situation, but the dative flag is not the only morpheme in the conjugated verb that may be explained diachronically but whose grammatical function or correlate nowadays is difficult to pinpoint.As for the morphological exponent of the applicative head in the verb, it is well known that languages show great differences, varying from cases where the different incorporated Ps show up transparently, languages that have a single applicative morpheme for all semantically different relations, and languages that do not mark the applicative head in the verb at all.See Peterson 2007;Ormazabal & Romero 2017a, and references there for discussion.Clearly, historical processes are responsible of a great deal of the superficial variability in this respect. 26Minor differences aside, the distribution is very much like clitic doubled dative and locative preposition à ('to') in nowadays standard Spanish (see Ormazabal & Romero 2013).
(54) This is just a natural extension of the same semantic spreading process that has expanded the dative assigning P 1 to unbounded path contexts, as argued by Etxepare & Oyharçabal (2013) and Etxepare (2014).It is worth remembering that these complex postpositions appear quite late in the chronology of unbounded path datives, but earlier than agreementless datives (see sections 3.1, 4.2 and references there).Note also that in these structures, and only here, the lack of overt agreement is independent of whether P 1 incorporates to the higher postposition, since the incorporation into the lexical postposition is internal to the complex PP, a context where Basque does not have functional heads that could host agreement morphology.The desired result follows.
In contrast, the cluster of changes that occurred in the second half of XIX century is of a very different nature.If our approach is on the right track, all these structural changes reduce to a minimal parametric difference in the morphological feature specifications of the PPs involved.In this case, the shift from a stage of the language where dative agreement is obligatory to one where it becomes optional reduces to a change in the morphological properties of the incorporating P, which becomes optional, yielding a real dative alternation.Consider, in that respect, example (36), repeated in (55), with an agreement and an agreementless instance of the same verb: (55) Otoi, ate horr-i hurbil zaite, ni hurbiltzen please door that-dat approach aux.2pA.imp,I.abs approach nitzaion bezala.aux.1sA.3sD.past as 'Please, approach that door as I approached it.'Assuming this change in the parameter value of P-incorporation, all the important properties of the phenomenon follow in a unified and natural way: i) the fact that the new construction extends to virtually the entire range of dative constructions that are possible with agreement datives in all dialects; ii) the fact that this spreading process takes place really fast, that is, that all the relevant dative agreement contexts become optional practically at the same time; iii) the cluster of radical morphological and syntactic changes that accompany the modification of the parameter value: c-command relations, categorial differences, (lack of) agreement.Let us consider them in more detail.
Once the incorporation of P 1 becomes optional, the fact that the totality of the contexts where dative agreement was obligatory begin to show optional agreement and quickly enter the dative alternation is a natural consequence of our proposal: the same P 1 , in all its semantic extension, may or may not incorporate.We also predict that agreementless datives are not allowed in what Etxepare & Oyharçabal (2013) and Etxepare (2014) consider "bounded path" contexts, where agreement datives were not allowed either (57).This remains the realms of the non-incorporating allative postposition, as predicted ( 56 Crucially, the alternation has no semantic basis, as expected: the shift from obligatory dative agreement to dative alternations extends across-the-board to all contexts, and only to them, because it is just the reflex of a morphological readjustment in P 1 , whose incorporation becomes optional.In consequence, this process does not discriminate between "old" and "new" datives, all of them are equally affected by the change. The surface morphological and syntactic differences between the agreement and the agreementless dative constructions also follow from the optionality of the incorporation.As mentioned in section 5.1, agreement and agreementless PPs both share the same basic structure in (58), where the direct object c-commands the indirect object: (58) Since P 1 incorporates optionally, when it does not incorporate the c-command and categorial properties are those in (58), similar to locative postpositions, where the DO c-commands the embedded dative PP and no agreement with the verb holds.If it incorporates, however, it triggers movement of the indirect object -generated as the DP complement of P 1 -to the object agreement position in (59), a position where it not only agrees with the verb, but also c-commands the direct object.( 59) In fact, the optionality of the incorporation and the appearance of a true alternation in Lapurdian may be accounted for by a conspiracy of factors in these dialects that are absent in Western ones.In particular, the extension of the unbounded path interpretation to complex postpositions such as mendiari gora ('up the mountain') discussed in ( 53)-( 54) indirectly creates the first context where dative agreement is not available.Morphological reanalysis of the lack of agreement in these constructions as not involving P 1 incorporation might be the trigger for further extension to the verbal domain, making the non-incorporation strategy available more generally.Furthermore, unlike Spanish, Continental Romance languages in contact with Lapurdian have not developed a dative clitic doubling strategy, a mechanism that may be analyzed as a dative agreement.This might have contributed to reinforce the obligatory vs. optional agreement split between Western and NE dialects, yielding the system described in this paper.
Finally, as in many other cases of true dative alternation crosslinguistically (see references in section 3.3), the dative agreement structure underwent a process of specialization and, associated to this, a recent tendency to prime the non-incorporated version -a phenomenon that occurs only when real dative alternations exist in the language.This factor is driving the progressive loss of locative applicative incorporation in some contexts, again most probably reinforced by the contact with the French prepositional structures, where à never incorporates (see Ormazabal 2017 for details).
It might be argued that our account stipulates the obligatoriness/optionality for each language.Indeed, the information about whether P 1 incorporates or not must be codified somewhere among its lexical properties.Notice that non-derivational analyses, in turn, must stipulate whether a language has dative construction, to-constructions or both.But in our case, this is encoded in a more abstract way, restricted to the feature properties of adpositional heads, and conforming to a very restricted set of possible parametric possibilities.Importantly, the cluster of morphological and syntactic properties associated to each construction is not stipulated, encoded in each syntactic construction, but results from the derivation itself.To finish, the changes conform to a series of i-language independent circumstances that may be pinpointed and evaluated with standard tools of historical linguistics, as we have shown.

A note on "compromise" approaches: Compatible frames
An anonymous reviewer suggests a possible alternative with a PP sub-structure common to both dative alternates -be it an applicative or an adpositional head, or a mixture of both -that merges in different positions: higher than the DO in the agreement construction and lower in the agreementless one.In both cases, this PP would have the appropriate semantic interpretation, capturing the non-accidental extensional relationship between the two structures, but without postulating a direct derivational connection between them.The reviewer mentions Rezac's (2011) proposal to deal with repair strategies.
It is worth noting that, as far as we can see, Rezac's proposal is not a general proposal meant to account for dative alternations in full.But even if extended, we do not see the advantage of such an approach over the derivational one.Instead, a reason to prefer a derivational approach is that the specific interpretation of the IO heavily depends on the combined contribution of the "bleached" P head together with the lexical specifications of the verb.Thus, the dative argument is interpreted as a goal/recipient under igorri ('send') in (60a-b) but as an unbounded path under hurbildu ('approach') in (61a-b) [repeated from the introduction], strongly suggesting that selection is playing a crucial role: (60) a. Ama-k semea-ri ogia igorri dio.mother-erg son-dat bread.abssent aux.(3sA).3sD.3sE'The mother sent (the) bread to the son.' b.Ama-k ogia igorri du semea-ri.mother-erg bread.abssent aux.(3sA).3sEson-dat 'The mother sent (the) bread to the son.' (61) a. Ate horr-i hurbil-tzen nitzaion.door that-dat approach-hab aux.1sA.3sD.pastI approached that door.' b Ate horr-i hurbil-tzen nintzen.door that-dat approach-hab aux.1sA.past'I approached that door.'If we assume a syntactically compositional approach to argument structure (Hornstein 2001;Ramchand 2008, among others), this complex relation might be codified if created via head movement.But, leaving aside unimportant structural details,29 we find it difficult to see how the two hypotheses could be efficiently distinguished empirically once V-to-I raising of the verbal heads occurs, and theoretically it seems to us that we would be dealing with notational variants of the same derivation after all.
Under the alternative assumption that thematic relations are obtained uniformly under the application of external merge (e.g.Chomsky 2014), a derivational analysis with a common basic structure where the selectional properties of V and P are uniformly represented would be preferred.Given all that, we do not see any advantage in postulating two different but compatible structures, not even circumventing the technical problems the incorporation analysis might have, to end up with the same semantic representation/interpretation.
Be it as it may, given all the above, the only case that remains to be considered is the lack of alternation in so-called "high applicatives", that is, experiencer, possessor, etc., which always show obligatory agreement.Extending a proposal by Odria (2014Odria ( , 2016)), in the next section we sketch a possible way to go.

Some observations on the High/Low Applicative distinction
Strictly speaking, no extra assumption is needed to deal with high datives, other than a minimal adaptation of the "High/Low applicative" analysis to our proposal.Unifying agreementless PPs and low applicatives does not necessarily mean that high applicatives could not be a different structural relation altogether as proposed by Pylkkänen (2008), etc.However, the same conceptual arguments that motivated eliminating the ontological distinction between dative PPs and "low applicatives" in the first place also suggest that we might try to apply the same reductionist view to the special status of high datives, and analyze them in terms more in accordance with our derivational approach.It is not our goal in this paper to present a detailed syntactic analysis, but just to point at some directions that we think could help us reduce their special properties, sketching the general lines of an hypothesis in that direction.
As in the case of low datives, a distinctive "High Applicative" projection per se lacks explanatory power; moreover, even as a descriptive tool, the properties attributed to it, as opposed to the low applicative projection, do not seem to make the right cut once additional evidence is considered in detail (see Ormazabal & Romero 2010;Odria 2014 for discussion).
Paul & Whitman (2010) and Odria (2016) already open a way for unification.They propose that the ApplP locus for high and low datives is the same projection, but while low datives reach that position by internal merge, moving from lower structural positions, high datives are base-generated directly in the specifier of the applicative projection.Odria presents extensive empirical evidence related to secondary predication, agreement restrictions of the Person-Case Constraint type, repair strategies in contexts of competing dative arguments, and the distribution of dative Differential Object Marking in Western-Central dialects showing that agreeing dative arguments do not all share the same category in the base, and that the standard criteria do not cut across the high/low line, but across categorial properties of the position of origin. 30 Our proposal is to go one step further.Theoretically speaking, the situation pictured by these authors is very similar to the analysis of subjects in the early and middle eighties, before the first VP-internal Subject Hypotheses were articulated (Kuroda 1988;Koopman & Sportiche 1991;etc.).At that time, the standard view was to assume that the subject of transitive predicates was base-generated in the Specifier of IP (Spec, TP), but the subject of unaccusatives, raising predicates, passives, etc. moved from a lower position to end up occupying the same specifier position.One of the innovative forces of the VP-internal hypothesis was that it generalized the derived nature of all subjects and associated them structurally to a functional projection were they all would end up at some point of the derivation, contributing to further dissociate the Case and agreement properties and the functional relations from properties related to argument structure and selection.A consequence of that was that the different structural properties the various subject types show could be derived, not from their subjecthood, but from their diverse origin, while the properties associated to subjecthood might be derived from the fact that they share the structural property of being in the same (Spec, IP) position.
In the same vein, our proposal is that all applied/dative arguments might be derived, and that the distinction points to different base-positions from which dative shift occurs (see Michelioudakis 2012 for a fully developed analysis in the same direction based on Greek).In fact, the elements standardly classified as belonging to the class of high datives, (i) possessor raising, (ii) experiencers and (iii) ethical datives of all sorts, constitute a very heterogeneous group, each of them presenting different and specific syntactic and semantic properties, what makes them hard to unify under a single label, which would become a bric-a-brac. 31 This is already standard for causee datives in causative constructions, which are generally assumed to be generated in the external argument position of the verb embedded under the causative (Baker 1988 and much subsequent literature including most AUCH! approaches).This accounts, for instance, for the fact that causee datives, unlike PP-alternating ones, allow secondary predication.The same general strategy may be extended to high applicatives as well.Let us suppose, for instance, that some version of 30 It is important to emphasize that we are referring to differences among dative agreeing DPs, not to the classical, and much more analyzed DP/PP distinction associated to the agreement/agreementless datives (section 3 and references).Odria's innovation is that she presents clear criteria to make a much more sophisticated distinction among dative agreeing DPs and, as said in the text, that this division does not correspond to the standard high/low dative cut, but to their categorial and configurational properties in the base position.Her important contribution not only provides objective criteria for classification but, more importantly, it presents an articulated proposal that captures these differences at the base level, subsumes the high/low dative distinction, and achieves a structural unification in terms of their "landing site".That analysis -not only the first articulated one Basque-internally but extremely relevant in more general terms (also see Paul & Withman 2010) -goes far beyond the intricacies of the particularly rich Basque data on dative constructions. 31Apart from the fact that they do not alternate with the agreementless PP construction, there are no clear criteria to group them together.Thus, for instance, Pylkkänen extensively argues that possessor datives are "reversed low applicatives".If that is the case, the fact that they do not alternate while all the other low applicative do is even more mysterious.Moreover, as an anonymous reviewer observes, the label "ethical datives" has been used to cover a very broad cluster of phenomena that look different pragmatically and in their structural properties (see Fernández 2010).
the raising analysis of possessor datives is on the right track.In that case, if the dative originates within some position internal to the object DP (see, e.g.Landau 1999;Arregi 2003b; and references), we would not expect it to ever show up in an agreementless dative frame, because there is no possible PP source in VP from which the non-incorporating P and the dative argument would originate.If the dative alternates at all, it would be with some genitive argument in the DP-internal position.On the other hand, experiencers are arguments selected by a specific set of verbs, they are universally associated to specific syntactic structure alternations and, in many languages, they are the only dative arguments that may act as subjects.However we account for their complex syntax, it is necessary to take into account the fact that they are subject to lexical selectional conditions (Belletti & Rizzi 1988;Landau 2010; among many others).In a radically different situation, ethical datives are non-argumental items restricted to pronominal forms.They differ from the rest of the datives in important respects such as the fact that they do not trigger PCC effects, or that they cannot be doubled, what makes them clitic-like elements instead of agreement relations (see Fernández 2010; Romero 2014, among others).

Summary and final remarks
Summarizing, the historical changes occurred in Lapurdian Basque constitute an excellent showcase in which to look for the answer to important theoretical issues concerning the syntax and semantics of dative alternations.They show that the semantic extension of dative constructions is independent of the dative alternants themselves.On the one hand, we may see that the applied arguments are not semantically restricted to changeof-possession, but extend to other domains.That is clear in many languages of the world that show a dative construction richer than English, but the extension process is neatly documented in the case of Lapurdian Basque, where we can follow how the dative is gradually occupying other semantic fields as it extends to the conceptual domain of unbounded paths.It is a slow and progressive process that extends the domains of dative agreement to new contexts previously occupied by allative and locative postpositions, far beyond change of possession contexts.The mere extension to non-possession domains already poses serious problems for theories that associate the applicative construction to the semantics of change-of-possession.The fact that this semantic extension is completely dissociated from any syntactic change, as we have shown, makes theories that postulate a one-to-one relation between a specific syntactic construction and a corresponding interpretation untenable.
At the opposite end of the scale, Lapurdian also shows that PP-constructions are not restricted to path contexts either.We have shown that the dialect accommodates a nonapplied PP strategy, from the second half of XIX century on.This strategy extends virtually through the entire semantic domain of dative DPs, independently confirming Rappaport Hovav & Levin's (2008) important results that the semantic extension of the PP-construction depends on the selecting predicate, and not on the construction itself.Moreover, the properties of this diachronic change, the way it generalizes, and the fast pace support a connection between the agreeing dative construction and the newly created agreementless PP.As we have shown, both structures differ radically in the same well known syntactic terms as Double Object Constructions and to-constructions do.However, the conceptual spaces they occupy mostly coincide -except for the case of high dativesstrongly supporting a derivational relation between them.
In a more speculative vein, we have also suggested in the last section of the paper a possible analysis to deal with high datives -most clearly possessor and psychological datives -as well as causees in causative constructions, which do not have a PP-origin and, consequently, do not enter the dative alternation.We propose to apply what we erg Kutxi street-dat lights.abschange aux.(3pl.A).3sD.3sE'The city hall changed the lights in Kutxi street.' b.Jon-ek aulkia-ri hanka konpondu dio.
Concerning their structural hierarchy, non-agreeing dative construction exhibit the opposite hierarchical relation to the agreement dative one analyzed in section 3.1.The minimal pair in(19)and the examples in(20)show that in this construction the dative is c-commanded by the absolutive DO.
7 (19) a. *Jon eta Miren i ez ziren ohartu pro i elkarr-i i lotu Jon and Mary.abs not aux realize recip-dat tie nizkio-la.aux.(3pA).3sD.1sE-comp'Jon and Mary i didn't realize that I was tying them together.'(Lit.: 'to each other') b.Jon eta Miren i ez ziren ohartu pro i elkarr-i ; Fernández; Ortiz de Urbina & Landa 2009; Etxepare & Oyharçabal 2013; Odria 2017, and references there): Etxepare & Oyharçabal (2013)al (2013)present various fixed theme-type idiomatic expressions.That includes the minimal pair in (35) -both examples corresponding to10 Ormazabal (2017)has conducted a systematic analysis of all the occurrences of the predicate igorri in the As for the innovative unbounded path locative and aspectual contexts, no semantic difference has been described between the agreementless dative structure and the agreement one, as is clearly shown in examples like (24b), repeated in (36), where both agreement and agreementless constructions are paired together in the same sentence: Ormazabal & Romero (2010)2008)2656 occurrences of igorri ('send') in the corpus, not a single one of them has an inanimate dative -neither agreement dative nor agreementless one -and all the cases of inanimates appear with the allative -ra.Similarly, our informants absolutely agree with that distribution.This is a particularly important result, given that inanimates do show up dative-marked with other predicates, as discussed throughoutly in this paper.This strongly suggests that animacy must be associated to the selectional properties of the predicate types, not to one of the two alternating structures in dative configurations, as argued byRappaport-Hovav & Levin (2008);Ormazabal & Romero (2010).
Such an analysis postulates a general LowAppl strategy, common to all dialects of Basque, where general dative agreement takes place:13 Ormazabal & Romero 2010;008;rd dative structure, the so-c2016d Oyharçabal 2010; constructions.In addition to their obligatory agreement properties, high applicatives are said to differ from low applicatives in several respects.For instance, they are said not to contribute to the event schema of the sentence(Cuervo 2003;Pylkkänen 2008; Fernandez 2010 Fernandez  , 2014;;Oyharçabal 2010; Etxepare 2014, among others; but see Larson 2010).Furthermore, it has been argued that High Applicatives allow secondary predication (idem; but seeOrmazabal & Romero 2010; and, very specially, Odria 2014;2016; also see section 6, below).
. See this work for discussion of internal structure and properties.