engleski

Croatian Conjunct Agreement Preferences Disambiguate Relative Clause Antecedent

The experimental advances into the attachment site of relative clauses (RC) headed by a complex NP have attested both HA and LA preference in typologically diverse languages. While it has been shown that structural properties within a language and experimental design itself have an effect on attachment preferences (e.g. Frazier & Clifton 1996, Hemforth et al. 1998, Hemforth et al. 2015), the investigations into the HA+LA languages have further fortified the efforts to relate the (residue of) cross-linguistic variation to the availability of certain language inherent properties such as the optionality in prosodic phrasing posited for Croatian (Lovrić and Fodor 2000, Lovrić 2003) or the availability of pseudo-relatives for Italian (Grillo and Costa, 2014). To expand the list of language-specific attachment height predictors, we ran a questionnaire and an eye-tracking study exploiting the syncretic form of the relative pronoun and the three-way conjunct agreement potential in Croatian. Similar within-language variation in preferences with respect to agreement height has been observed for conjunct agreement (CA) in South Slavic languages (e.g. for Slovenian Marušič et al. (2015), for BCS and Slovenian Willer-Gold et al. (2016, 2018)), where predicate has been shown to agree in Gender+Number (G+N) with the hierarchically higher conjunct, the linearly closer conjunct and the entire conjunction phrase. Even with the same structural configuration and the same set of features undergoing valuation, CA and RC agreement are distinguished by the coindexation. In RC, the morphological marking on the relative pronoun points to its antecedent, hence the two have to be co- indexed. In CA, irrespective of the morphological marking on the predicate, the semantically valid subject can only be the entire conjunction phrase (c.f. Link-Agree in Marušič et al. 2015). Consequently, while in, for example, CA linear agreement exponent is purely morphological and carries no interpretative component, the opposite is predicted to hold for LA RC. Current study is designed to test two hypotheses: (i) the three- way potential attested in the experimental literature for conjunct agreement can serve as a predictor of attachment preferences in RC ; and, if confirmed (ii) the independently available underlying mechanics of agreement (a) constrains the attachment potential and, hence, (b) facilitates the RC interpretation. To achieve the stated research objectives we focus on RC headed by coordinated phrase in Croatian and capitalise on language-inherent property to morphologically mark the G+N values on the complementiser, spelled-out as the relative pronoun. In the experimental study, the RC was headed by a conjunction of two human masculine singular nouns in subject position. A violation paradigm manipulated the RC internal properties crossing pronoun case [nominative, accusative] and predicate number [singular, plural] to design a single set of materials. These were tested by two complementary methodologies: an offline questionnaire [exp1] to attest attachment preferences, (i), and an online eye- tracking study [exp2] to detect processing difficulties, (ii). Participants were Croatian native speakers, university students of comparable demographics. The questionnaire results (n=30) replicate three-way attachment potential observed for CA in Ambiguous condition. In the remaining three conditions favouring single antecedent, the results show the overall preference for LA and the lack of HA in the Case mismatch condition (Figure 1). The results of the eye-tracking experiment (n=30) consistently show processing difficulty arising in late measures at the relative pronoun region |rel.pro+∅/cl+aux|, taking longer to read sentences with the Ambiguous and the Ungrammatical condition [tfd: t: 1.919, t: 1.970 ; N of fix: t: 1.729, t: 1.946 (Figure 2 and Figure 3) ; and, N of reg: t: 1.795 with the Ungrammatical condition]. In addition, the main effect of the pronoun case and the predicate number has been observed for the late measures [tfd: t: 1.919, t: 1.970 ; N of fix: t: 1.729, t: 1.942 ; for predicate number N of reg: t: 1.795]. We contend that while processing difficulties in reading observed with the Ungrammatical condition are expected under standard violation paradigm, processing difficulties that arose with the Ambiguous condition can be taken as indications of disambiguation under (parallel) processing of multiple available agreement strategies (Hwang and Kaiser 2014) thereby creating the environment for the three-way attachment potential. Consequently, our research provides further support for language-independent investigations into RC (Betancort et al. 2009), contributes to the expanding experimental work on South Slavic conjunct agreement and challenges theories of agreement to be further refined in terms of the unvalued referential index feature (c.f. [ID] - Arregi and Hanink to appear).

Agreement ; Relative Clause ; Attachment ; Croatian Language

nije evidentirano