A direct analysis of Lithuanian phrasal comparatives

Phrasal comparatives like John runs faster than Tom can be derived via two routes: i) Direct Analyses assume that the argument of than is a DP; ii) Reduced Clause Analyses assume that the than -phrase is a clause that is subject to some reduction operation ( John runs faster than Tom runs ). Based on standard diagnostics used to adjudicate between the two analyses, I argue that Lithuanian phrasal comparatives are best analyzed as directly licensed. I develop a direct analysis of Lithuanian phrasal comparatives that captures the empirical generalizations previously discussed in the literature and a new empirical generalization identified in this paper. This proposal is contra Grinsell (2012), who argued that Lithuanian phrasal comparatives are underlyingly clausal based on data that are suggested to show island sensitivity. I argue against Grinsell’s (2012) reduced clause analysis by showing that the island sensitivity of Lithuanian phrasal comparatives can be successfully captured in a direct analysis.


Introduction
Phrasal comparatives are comparatives in which the standard of comparison superficially is a DP. 1 In the Lithuanian example in (1), the standard of comparison is the accusative case-marked DP Tomą 'Tom', which contrasts with the nominative case-marked associate Jonas 'John'. In phrasal comparatives, the standard marker is už 'than'.
(1) Jonas aušt-esn-is už Tomą. John.nom tall-er-nom than.phrasal Tom.acc 'John is taller than Tom.' Lithuanian also has a clausal comparative, a comparative in which the standard of comparison is a clause. In (2), the standard of comparison is a clause that can optionally contain the complementizer kad 'that'. In clausal comparatives, the standard marker is negu 'than'.
(2) Jonas auštesnis negu (kad) Tomas. John.nom taller.nom than.clausal that Tom.nom 'John is taller than Tom.' Clausal comparatives are traditionally analyzed as derived from clausal sources, i.e. the standard of comparison is a clause that is subject to a reduction operation (e.g. Lechner 2001;Merchant 2009, among others).
Regarding the syntax of phrasal comparatives, there are two classes of analyses, which differ in the assumed syntactic size of the standard of comparison: reduced clause analyses (RCAs) and direct analyses (DAs). Under RCAs, phrasal comparatives are bi-clausal. The standard of comparison is a clause that is subject to some reduction operation(s) (e.g. Lechner 2001;2004;Pancheva 2006;Merchant 2009, among others). In (3), the standard of comparison undergoes ellipsis (the elided material is indicated by the angled brackets) and the only phrase that survives ellipsis is the remnant, Tomą 'Tom'. The remnant and Jonas 'John' occupy parallel positions in separate clauses, i.e. they are subjects in their respective clauses.
It has been accepted in the literature that both RCAs and DAs are attested cross-linguistically. The big question is how to decide which strategy is employed in a particular language.
Given that RCAs and DAs make different assumptions regarding the size of the standard of comparison, a number of diagnostics that are sensitive to clausal boundaries have been proposed to adjudicate between the two analyses. RCAs and DAs make different predictions with respect to i) case-marking on the standard of comparison; ii) binding of pronouns and anaphors; iii) whether only a single phrase can follow the standard marker (e.g. Hankamer 1973;Merchant 2009;Bhatt & Takahashi 2011, among others). I will now discuss how Lithuanian phrasal comparatives fare with respect to the standard diagnostics.
Anaphors (reflexive pronouns) must be locally bound, i.e. they need to have an antecedent in the same clause. Free pronouns, on the other hand, must be locally free, i.e. their antecedent cannot be in the same clause. Consequently, whether a phrase can serve as an antecedent to a (reflexive) pronoun in the standard of comparison can inform us as to whether phrasal comparatives are mono-clausal (as predicted by DAs) or bi-clausal (as predicted by RCAs).
Lithuanian has a strictly clause-bound reflexive anaphor save 'self', which can be inflected for case. The anaphor shows subject orientation as shown in (4), i.e. it can only be bound by a subject in the same clause. (4) Jonas i pristatė Tomui j savo i/*j brolį. John.nom introduce.pst.3 Tom.dat self.gen brother.acc 'John introduced his brother to Tom.' In phrasal comparatives, the anaphor savo 'self' can be bound by the subject associate, as shown in (5a). In the Lithuanian clausal comparatives, the anaphor savo 'self' cannot be used as shown in (5b). The data in (5) thus shows that the associate Jonas 'John' and the possessive anaphor savo 'self' are in the same clause in phrasal comparatives. This binding pattern is predicted by DAs but not RCAs.
(5) a. Jonas i bėga greičiau už savo i brolį. John.nom run.prs.3 faster than.phrasal self.gen brother.acc 'John runs faster than his brother.' b. *Jonas i bėga greičiau negu savo i brolis. John.nom run.prs.3 faster than.clausal self.gen brother.nom 'John runs faster than his brother.' Free possessive pronouns in Lithuanian show anti-subject orientation within the same clause. In the Lithuanian example in (6), the possessive pronoun jo 'his' can be co-referential with the indirect object Tomas 'Tom' but not with the subject Jonas 'John'.

(6)
Jonas i pristatė Tomui j jo *i/j brolį. John.nom introduce.pst.3 Tom.dat he.gen brother.acc 'John introduced Tom to his brother.' In the phrasal comparative in (7a), the pronominal possessor jo 'his' cannot be coreferential with the subject Jonas 'John'. Therefore it follows that that the associate Jonas 'John' and jo 'his' are in the same clause as predicted by DAs but not RCAs. In the clausal comparative in (7b), on the other hand, jo 'his' is ambiguous between referring to the subject of the clause and some contextually salient individual, which means that the associate Jonas 'John' and jo 'his' are in separate clauses.
Jonas i bėga greičiau negu jo i/j brolis. John.nom run.prs.3 faster than.clausal self.gen brother.nom 'John runs faster than his brother.' The second diagnostic is case-marking on the standard of comparison. Under RCAs, the standard of comparison occupies a position that is parallel to the one occupied by the associate. The case-marking on the standard thus depends on its structural position. Under DAs, on the other hand, case is assigned locally by the preposition. In Lithuanian, the standard of comparison is invariably accusative case-marked, even when the standard contrasts with the nominative casemarked subject as shown in (8).
(8) Jonas auštesnis už Tomą / *Tomas. John.nom taller.nom than.phrasal Tom.acc / Tom.nom 'John is taller than Tom.' This case-marking pattern contrasts with German, a language in which phrasal comparatives have been analyzed as underlyingly clausal (Lechner 2001;2004). In the phrasal comparative in (9), the first person singular pronoun bears either nominative case (ich) in (9a) where it is the subject of the elided clause or accusative case (mich) in (9b) where it is the object of the elided clause.
(9) German a. Er mag dich mehr als ich. he.nom like.prs.3.sg you.acc more than I.nom 'He likes you more than I (like you).' Er mag dich mehr als mich. he.nom like.prs.3.sg you.acc more than I.acc 'He likes you more than (he likes) me.' Case invariability in Lithuanian is predicted under a direct analysis: the standard marker už 'than' is a preposition and assigns case to its argument. In Lithuanian, prepositions select for a DP argument and assign case. In (3), už 'for' is a preposition (homophonous with the standard marker už 'than') that assigns accusative case to its DP argument.
(10) Jonas balsavo už Grybauskaitę. John.nom vote.pst.3 for Grybauskaite.acc 'John voted for Grybauskaite.' Lastly, RCAs and DAs make different predictions with respect to what can follow the standard marker. Under DAs, the standard of comparison can only be a single phrase, since prepositions take only one argument. Under RCAs, more than one phrase can survive ellipsis, i.e. there can be multiple remnants.
Japanese yori-comparatives allow multiple remnants as shown in (11) and they have received a reduced clause analysis (Bhatt & Takahashi 2011). 2 (11) Japanese (Bhatt & Takahashi 2011: 608) Taroo-ga Tokyo-de-yori Jiroo-ga Kyoto-de ooku-no hito-ni atta. Taro-nom Tokyo-in-than.phrasal Jiro-nom Kyoto-in many-gen people-dat met. 'Jiro met more people in Kyoto than Taro in Tokyo.' In English, syntactic categories other than DPs can function as the standard of comparison as shown in (12), where the standard is an adverb.

(12)
John read more books today than yesterday.
Lithuanian phrasal comparatives allow only a single argument to follow the standard marker už 'than' as shown in (13a). Furthermore, the standard of comparison must be a DP as shown in (13b), where the standard is the adverb vakar 'yesterday'. 3 John.nom read.pst.3 more book.gen than.phrasal yesterday 'John read more books today than yesterday.' Since prepositions take only a single DP argument in Lithuanian, this empirical generalization is predicted by DAs but not RCAs.
In summary, the following generalizations about Lithuanian phrasal comparatives emerged from applying diagnostics used to adjudicate between RCAs and DAs: 2 Yori-comparatives with multiple remnants are derived from clausal sources via RCA. When the remnant is a single DP remnant, Bhatt & Takahashi (2011) propose that the yori-comparative is derived via DA based on binding properties of the standard of comparison. RCA in constructions with a single DP is blocked by an economy constraint. 3 Both of these sentences can be rendered grammatical with a negu-comparative as shown in (i). i. the standard of comparison and the associate are in the clause, ii. the standard of comparison is invariably accusative case marked, iii. the standard of comparison must be a DP.
All these empirical generalizations support a direct analysis of Lithuanian phrasal comparatives and not a reduced clause analysis. 4 Despite the fact that the diagnostics overwhelmingly support a direct analysis, Grinsell (2012) develops a reduced clause analysis. His motivation for a clausal analysis is based on the fact that phrasal comparatives seem to exhibit island effects.
Islands are assumed to be syntactic constraints that prevent extraction from certain syntactic configurations (e.g. Ross 1967; Chomsky 1995). Since under DAs, phrasal comparatives are base-generated mono-clausal structures, the presence of island effects would be unexpected. Since Merchant (2009), island effects have been used as an argument in favor of RCAs.
In this paper, I propose a direct analysis of Lithuanian phrasal comparatives under which the island violations emerge due to the movement of an associate at the LF. One consequence of this empirical discussion, which extends beyond the proper analysis of Lithuanian phrasal comparatives, concerns the use of island effects as a diagnostic used to adjudicate between RCAs and DAs originating in Merchant (2009). The use of this diagnostic has been extended into the description and analysis of phrasal comparatives of other languages (Grinsell 2012;Lindenbergh 2016). I suggest that the apparent island effects should not be considered as a decisive diagnostic, since they are compatible with RCAs and DAs contra claims in Merchant (2009).

Empirical characterization of Lithuanian phrasal comparatives
In this section, I discuss empirical generalizations about Lithuanian phrasal comparatives that are known in the literature. In subsection 2.2, I introduce a novel empirical generalization, which states that the associate must be the subject of the clause.

Previously discussed empirical generalizations
Three empirical generalizations about Lithuanian phrasal comparatives have been identified in Grinsell (2012).
First, phrasal comparatives are incompatible with measure phrases like du metrai 'two meters' as exemplified in (14a) (Grinsell 2012). The clausal comparative in (14b), on the other hand, is grammatical with the same measure phrase.

b.
Jonas aukštesnis negu du metrai. John.nom taller.nom than.clausal two.nom meter.nom 'John is taller than two meters.' Second, in Lithuanian nominal comparatives are formed with a comparative word daugiau 'more', which consists of two morphemes: daug 'a lot' and -iau '-er'. Phrasal comparatives in Lithuanian are ungrammatical if the subject of the sentence is preceded by daugiau 'more' as shown in (15a)  While all these properties are predicted by a direct analysis, Merchant (2009) develops a reduced clause analysis that aims to capture the local relation between the standard of comparison and the associate in terms of binding and case-assignment. The assumptions that Merchant (2009) has to make to capture this local relationship will be discussed in more detail in Section 4. Daugiau student lanko Čikagos universitetą negu more student.gen attend.pst.3 Chicago.gen univeristy.acc than.clausal Northwesterno. Northwestern.gen 'More students attend the University of Chicago than Northwestern.' Third, Grinsell (2012) argues that phrasal comparatives exhibit island effects. Grinsell provides only one example of a phrasal comparative exhibiting the purported island effects, namely the example in (16) (Grinsell 2012: 41) The ungrammaticality of the example cannot be attributed conclusively to island sensitivity. The ungrammaticality is expected because, as discussed in Section 2, 'more DP subjects' are ungrammatical with phrasal comparatives. 5 Consequently, the data provided in Grinsell (2012) does not constitute evidence for island effects.
To test a purported island sensitivity of phrasal comparatives, I constructed the minimal pair in (17) The clausal comparative in (17a) is ambiguous between two readings: Tomas 'Tom' can either be interpreted as the subject of the matrix verb suvalgė 'ate' (leading to Reading 1) or the subject of the relative clause internal verb iškepė 'baked' (leading to Reading 2). 7 The two readings and the syntactic structure associated with them is presented in (18).
Reading 1: John ate more cookies that Agne baked than Tom ate the cookies that Agne baked. b.
Reading 2: *John ate more cookies that Agne baked than he ate the cookies that Tom baked.

5
The sentence also contains some punctuation and vocabulary errors that could have influenced the acceptability of the sentence.

6
There is variability of acceptability of the comparatives in (17). Two out of three native speakers consulted found both sentences acceptable.

7
The most natural word order for these sentences as judged by three native speakers of Lithuanian is given in (i). In these examples, the DegP už Tomą and negu Tomas 'than Tom' is not sentence final. With this word order, the clausal comparative in (ia)  The phrasal comparative in (17b) is grammatical, but in contrast to (17a), it only allows Reading 1, i.e. it is not ambiguous. The question is why Reading 2 is unavailable in (17b). Grinsell (2012) suggests that phrasal comparatives exhibit island effects. Based on the apparent island effects, Grinsell's (2012) concludes that since phrasal comparatives warrant a reduced clause analysis. In Section 3, I will show that island sensitivities are successfully captured in a direct analysis. Consequently, the unavailability of the relative clause internal reading cannot be viewed as evidence in favor of RCAs.

Novel empirical generalization
The novel empirical generalization about Lithuanian phrasal comparatives is that the associate must be the subject of the clause as shown in (19). 8 In (19a), the associate is the subject of the clause Jonas 'John'. The comparison is between the number of doughnuts that were eaten by John and the number of doughnuts eaten by Tom. In the ungrammatical example in (19b), the associate is spurgų 'doughnuts', the direct object of the verb valgyti 'to eat'. The comparison is between the number of people eating doughnuts and the number of of people eating cookies. One might wonder whether the correct empirical generalization for Lithuanian is that the associate cannot be a nominal that is part of 'more NP'. The sentence in (20) shows the privileged status of the subjects in phrasal comparatives more clearly. The phrasal comparative in (20) contains a ditransitive verb. Consequently, there are two nominals that are not part of 'more NP' and therefore could potentially serve as the associate. However, the sentence is not ambiguous in Lithuanian. Only the subject of the sentence Jonas 'John' can be interpreted to be the associate. The reading in which Marijai 'Maria', the indirect object of padovanojo 'gifted', is the associate does not arise in Lithuanian. (20) Jonas padovanojo Marijai daugiau dovanų už Tomą. John.nom gift.pst.3 Maria.dat more present.gen than.phrasal Tom.acc 'John gave more presents to Maria than Tom (did).' #'John gave more presents to Maria than (he did to) Tom.' The example in (20) shows that the associate must be the subject of the sentence in Lithuanian.
In summary, this section discussed empirical generalizations that hold about Lithuanian phrasal comparatives as well as introduced a novel empirical generalization. An empirically adequate analysis needs to capture the following empirical generalizations about Lithuanian phrasal comparatives: 8 Lithuanian has dative experiencers, which in some languages behave as quirky subjects. Although dative experiencers are found in sentence-initial position, a position canonically reserved for subjects in Lithuanian, they are not syntactic subjects. Dative experiencers do not bind reflexives nor control agreement on the predicate with respect to person as shown in (i). In Lithuanian phrasal comparatives, the associate is the nominative case-marked DP and not the dative experiencer as shown in (ii), which is consistent with the novel empirical generalization.
(ii) Man spurgos patinka labiau už Joną. I.dat doughnut.nom like.prs.3 more than John.acc 'I like doughnuts more than I like John.' #'I like doughnuts more than John likes doughnuts.' The binding pattern in experiencer constructions is changing according to Holvoet (2013), who suggests that in colloquial Lithuanian binding by dative experiencers is accepted by some speakers. i. they are ungrammatical with measure phrases (Grinsell (2012)); ii. they are ungrammatical if the subject of the sentence is preceded by daugiau 'more' (Grinsell (2012)); iii. they exhibit island effects (Grinsell (2012)); iv. the associate must be the subject of the sentence.

A direct analysis of Lithuanian phrasal comparatives
In this section, I develop a direct analysis of Lithuanian phrasal comparatives (Hankamer 1973;Bhatt & Takahashi 2007;2011;Merchant 2012, among others). To illustrate basic assumptions of the analysis, consider the example in (21). The standard marker už 'than' is a preposition and the standard of comparison Tomą 'Tom' is a DP argument of už 'than'. A partial syntactic derivation of the comparative is shown in (22)  Following standard assumptions, I assume that gradable adjectives denote a function of type ⟨d,⟨e,t⟩⟩, as in (23a). The morpheme -esn-'er' marks the adjective in the comparative constructions. I assume that the comparative morpheme has the semantics of the comparative operator in (23b) (e.g. Heim (1985); Beck (2011); Bhatt & Takahashi (2011)). The operator combines three arguments: an individual-denoting standard of comparison, a gradable predicate and an individual-denoting associate.
Two points need to be explicated further about the proposed analysis.
First, in Lithuanian phrasal comparatives both the associate and the standard of comparison can be quantified phrases as shown in (25) Since the comparative operator in (23b) expects two individual-denoting expressions but instead finds a quantified expression, a type mismatch arises. The type mismatch is resolved by Quantifier-Raising the quantified expression. A schematic LF of the sentence in (25a) is given in (26). Second, the analysis of nominal comparatives relies on quantifier raising of the degree phrase. The movement of DegP is motivated by a type clash. The degree head expects the second argument to be a degree predicate. There is however, no degree predicate on the surface. I assume that daugiau 'more' in nominal comparatives is composed of determiner daug 'a lot' and the comparative morpheme -iau (cf. Bresnan (1973); Hackl (2000), a.o. for the view that English more is composed out of many and the comparative morpheme -er). I assume Hackl's semantics for the generalized quantifier daug 'many'. Following Bhatt & Takahashi (2011), I assume that nominal comparatives are derived via the mechanism of parasitic scope. 9 Parasitic scope involves two subsequent applications of Quantifier Raising: the first one targets the associate and the second one targets the DegP. In the Lithuanian example in (27a), first the associate Jonas 'John' raises to a higher scope-taking position. The first movement of creates a predicate via lambda abstraction. The movement of DegP to a position below the raised associate (also known as tucking-in) turns the existing predicate into a degree predicate. Due to space considerations the QR of DP obuolių 'apples' is not shown in the LF. The mechanism whereby a movement targets a position created by a preceding operation has been proposed and motivated in the literature independently (e.g. Nissenbaum (2000), Barker (2007) Having introduced the basic tenets of the direct analysis argued for in this paper, I turn to the empirical generalizations. First, as already discussed in Section 1, some of the empirical generalizations about Lithuanian phrasal comparatives straightforwardly fall out from the direct analysis. The fact that the standard of comparison must be a single DP is captured by the assumption that už 'than' is a preposition which selects for a DP argument. Invariant casemarking on the standard of comparison is captured by the assumption that the preposition už 'than' assigns accusative case to its argument. Lastly, the facts that the reflexives in the standard of comparison can be bound by the subject associate and that free pronouns cannot be bound by the subject associate are a direct consequence of the assumption that phrasal comparatives are mono-clausal. In the remainder of this section, I show how the direct analysis can also account for the four empirical generalizations laid out in Section 2.
First, measure phrases are incompatible with phrasal comparatives in Lithuanian. Measure phrases like 2 meters are ambiguous between two interpretations: a point on a scale (⟨d⟩, comparable to a temporal 2 o'clock) or a predicate over scale intervals (⟨d, t⟩, comparable to 2 hours, see Schwarzschild (2005) for a detailed discussion). Since the comparative operator in (23b) expects the standard to denote an individual, measure phrases are predicted to be ungrammatical with phrasal comparatives.
Second, only the subject DP can be interpreted as the associate in Lithuanian phrasal comparatives. This restriction has not been observed in phrasal comparatives in other languages. For instance, the Japanese phrasal yori-comparative in (28a) is ambiguous between two readings, as both watashiwa 'I' or nekoo 'cats' can be interpreted as the associate (Matsui & Kubota 2010: 126). This ambiguity is readily predicted by the parasitic scope mechanism: both watashiwa 'I' and nekoo 'cats' can be targeted by the first application of Quantifier Raising, resulting in two different LFs, (28b) and (28c) respectively.

(28)
Japanese ( Since Lithuanian phrasal comparatives are not ambiguous, contrary to the Japanese example in (28a), I argue that Quantifier Raising is subject to an economy constraint in Lithuanian in the sense of Richards (1997) Shortest principle. 10 The idea that Quantifier Raising is subject to an economy constraint is instantiated in Fox (1995) and Bruening (2001).
(29) Shortest (Richards (1997): 113) A pair P of elements [α, β] obeys Shortest iff there is no well-formed pair P' which can be created by substituting ɣ for either α or β, and the set of nodes c-commanded by one element of P' and dominating the other is smaller than the set of nodes c-commanded by one element of P and dominating the other.
Following Heim & Kratzer (1998: 210), I assume that the QR operation is available for all DPs, quantificational or not. Consequently, the subject DP is always in competition with other DPs.
In ( As pointed out by an anonymous reviewer, adopting Shortest makes a more general prediction about quantifier scope in Lithuanian, namely it predicts that the language is scopally rigid. This prediction is borne out. In the Lithuanian example in (31) that contains two quantified expressions, only the surface scope reading is available as judged by three native speakers consulted. The fact that the surface scope reading is available, whereas the inverse scope reading is not available indicates that quantifier raising is not freely available in Lithuanian. 12 (31) Kažkoks vyras pabučiavo kiekvieną merginą. some.nom man.nom kiss.pst.3 every.acc woman.acc 'Some man kissed every woman.' (some>every): One (specific) man kissed every woman. *(every>some): For every woman, she was kissed by some (potentially different) man.
Shortest predicts that in a phrasal comparative that contains a relative clause, only the subject of the matrix clause can function as the associate. This prediction is borne out as shown in (32a). Only Jonas 'John' can undergo Quantifier Raising, as shown in (32b). By contrast, the structure in (32c), in which Agnė 'Agne' undergoes Quantifier Raising is ruled out as it violates Shortest. The relative clause internal reading in (32c) is also be ruled out independently provided that Quantifier Raising is subject to island constraints (see e.g. Tanaka 2015 for experimental evidence). The associate Agnė would have to QR out of a relative clause resulting in an islandviolation. The apparent island-effects thus can be explained in a direct analysis. The movement of the associate cannot be island-violating. This analysis differs from the analysis presented in Grinsell (2012). The latter assumes that the island-violation arises due to the island-violating movement of the remnant, in this example Tomą. Grinsell's analysis is discussed in more detail in Section 4.
Lastly, Lithuanian phrasal comparatives are ungrammatical with 'more DP subjects'. In Lithuanian, 'more DP subjects' are ruled out by Shortest since the subject DP žmonių 'people' intervenes and thus rules out the QR of vyną 'wine' as shown in (33b).

wine [more than beer [λdλx [d-many people drank x]]]]
Lithuanian shares the 'more DP subjects' restriction with some other Balto-Slavic languages, for instance, Bulgarian as shown in the minimal pair in (34) (Pancheva 2009: 2). As in Lithuanian, the subject of a phrasal comparative cannot be preceded by a comparative word poveče 'more' in Bulgarian, as shown in (34a). The sentence can be rendered grammatical with a clausal otkolkoto-comparative as shown in (34b).

(35) Parasitic Scope Generalization (PSG)
In environments where movement of α provides the semantic context for type driven movement of β, the base position of α c-commands the base position of β.
A DegP can only move if the associate in its base position c-commands the DegP. The Bulgarian phrasal comparative in (34a) is predicted to be ungrammatical, since the associate Sofia 'Sofia' does not c-command the DegP poveče ot Varna 'more than Varna' as shown in (36).

(36) [Sofia [more than Varna [λdλx [d-many tourists visited x]]]]
It is possible that PSG is operational in Lithuanian phrasal comparatives just as it is in many Slavic languages, thus restricting the parasitic scope mechanism even further. The sentence in (33a) could also potentially be ruled out by PSG. Since the associate DP vynų 'wine' in its base position does not c-command the DegP daugiau už alų 'more than beer' as shown in (33b). However, it is not possible to test whether PSG is operational in Lithuanian phrasal comparatives. For instance, PSG would predict that the sentence in (37) is ambiguous between subject associate and indirect object associate reading. However, Shortest rules out the indirect object associate reading. Shortest thus only allows a subset of sentences that do not violate PSG. Further research is needed to determine whether there is independent evidence to assume PSG is operational in Lithuanian.
By virtue of adopting the direct analysis of phrasal comparatives and enriching it with one constraint on Quantifier Raising (Shortest), we can account for all empirical generalizations about Lithuanian phrasal comparatives that we set out to capture in Section 2.

Grinsell's (2012) reduced clause analysis
Standard diagnostics applied to Lithuanian phrasal comparatives motivates a direct analysis, yet the first analysis of Lithuanian phrasal comparatives in Grinsell (2012) argues them to be underlyingly clausal. The aim of this section is two-fold: i) to introduce Grinsell's reduced clause analysis of Lithuanian phrasal comparatives; ii) to critically assess the analysis in terms of its empirical coverage. Grinsell (2012) adopts the reduced clause analysis originally proposed in Merchant (2009) for phrasal comparatives in Modern Greek. Under this analysis, the standard of comparison is a clause, as shown in (38b) for the phrasal comparative in (38a). The remnant Tomą 'Tom' moves to a clause external position, while the clause itself undergoes TP-ellipsis (ellipsis site indicated by the angled brackets). The remnant Tomą 'Tom' moves cyclically to specPP via specFP, while the standard marker už 'than' moves from P to p. Grinsell's reduced clause analysis thus captures the ungrammaticality of (39a), attributing the ungrammaticality to island effects (though recall that this particular sentence is also ruled out by 'more DP subject' constraint).

Reduced clause analysis of Lithuanian phrasal comparatives
Under this analysis, both phrasal and clausal comparatives are bi-clausal, i.e. the standard of comparison is a clause. As shown in (40), a clausal comparative is grammatical. It is thus in need of explanation as to why clausal comparatives do not exhibit island effects.  (Grinsell 2012: 41) Following Merchant (2009), Grinsell (2012 suggests that the final landing site of the remnant is different in phrasal and clausal comparatives. In clausal comparatives, the final landing site is specFP. Grinsell (2012) argues that the clausal standard marker negu 'than' is morphologically complex as opposed to the morphologically simplex phrasal standard marker už 'than'. Historically, negu 'than' consists of two morphemes: ne 'neg' and gu a clitic (Bender (1921) as cited in Grinsell (2012)). Grinsell suggests that the two morphemes occupy distinct positions in the syntax. Crucially, the negation morpheme occupies specPP. The remnant Medvedevą 'Medvedev' therefore cannot move higher than specFP in the clausal comparative in (40), which is analyzed in (41). Since the remnant occupies specFP, there is no island-violating trace as shown in (41), which captures the absence of island effects in clausal comparatives.

(41)
Clausal comparative adopted from Grinsell (2012:  Grinsell's analysis thus is equipped to capture the purported island effects in phrasal comparatives as well as explain the lack of island effects in clausal comparatives. Grinsell's analysis, however, faces a number of challenges that are discussed in subsection 4.2.

Critique of Grinsell (2012)
In this subsection, I will discuss the stipulations Grinsell makes to account for the seeming local relationship between the associate and the remnant as revealed by the diagnostic tests in Section 1. I will then show that the reduced clause analysis does not account for the empirical generalizations discussed in Section 2 with the exception of the purported island effects.
First, to account for case-invariability on the remnant, Grinsell following Merchant (2009), posits an accusative case feature on p. To account for the fact that only a DP can serve as a remnant, Grinsell suggests that this fact can be captured either by assuming that only a DP can enter into an Agree relation with the case feature on p, or DP raising is triggered by a strong category feature (both Merchant (2009) and Grinsell (2012) remain agnostic with regard to the two options).
Grinsell's reduced clause analysis thus can successfully derive the local effects between the preposition už 'than' and the remnant as well as between the matrix clause and the remnant. The analysis is, however, less successful in accounting for the empirical generalizations discussed in Section 2.
First, Grinsell (2012)'s analysis does not capture the fact that Lithuanian phrasal comparatives are ungrammatical with measure phrases. The phrasal comparative with a measure phrase du metrai 'two meters' in (42) is incorrectly predicted to be grammatical in Lithuanian. The remnant is an accusative case-marked DP which is not contained within an island, i.e. it does not leave an illicit island violating trace, consequently it should be grammatical.
(42) *Jonas auštesnis už du metrus <aukštas>. John.nom taller.nom than.phrasal two.acc meter.acc tall 'John is taller than two meters.' Second, Grinsell (2012)'s analysis cannot capture the 'more DP subject' restriction. Grinsell (2012: 46) suggests in passing that the restriction can be analyzed 'as the result of an unelided island-violating trace'. However, the 'more DP subject' restriction cannot be reduced to a prohibition against unelided island-violating traces. In (43), the remnant Baltą Drobulę 'White Cloth' does not originate in an island. Under his reduced clause analysis, the phrasal comparative in (43) would be incorrectly predicted to be grammatical in Lithuanian. phrasal comparatives. The same line of reasoning has been applied for phrasal comparatives in other languages (e.g. Dutch Lindenbergh (2016) and Lithuanian Grinsell (2012)).
This paper has shown that the data that has been interpreted as evidence for island effects can be captured without making reference to underlying syntactic structures. Island violations arise due to the movement of the associate as LF movement cannot be island-violating. Our conclusion is that such data can be consistent with both direct and reduced clause analyses and should not be a priori interpreted as irrefutable evidence for RCAs. This raises the question about a proper analysis of Modern Greek apo-comparatives, especially since the Modern Greek phrasal comparatives behave similarly to Lithuanian phrasal comparatives in terms of casemarking on the associate and binding. That being said, it is beyond the scope of this paper to explore an alternative direct analysis of Modern Greek phrasal comparatives and its potential advantages.
Finally, this work is of typological interest as it adds a new empirical generalization into the discussion of variation in phrasal comparatives. Namely in Lithuanian only the subject of the clause can function as the associate. Bulgarian, a Balto-Slavic language, in which the 'more DP subject' constraint is active as it is in Lithuanian, allows non-subjects to be interpreted as associates as shown in (46).

(47)
Modern Greek (Merchant 2009: 136) *Perisoteri anθropi θrelun na maθrun anglika apo germanika. more people want.3p subj learn.3p English than.phrasal German 'More people want to learn English than German.' Lithuanian thus seems to be the only language known to exhibit the requirement that the associate is the subject of the sentence. Whether any other language exhibits this empirical generalization requires a thorough description of phrasal comparatives cross-linguistically and therefore is left for future work.