Abstrakt
Background: There is a controversy about the structure and interpretation of correlative and the corresponding relative clauses. Some analyses treat correlatives essentially as a subtype of a relative clause (particularly a headless or head-internal relative; Srivastav 1991; Dayal 1996), while others treat correlatives more on a par with conditional antecedents, ascribing them a simpler, propositional semantics (e.g. Bittner 2001; Brasoveanu 2008). L1-acquisition can provide a valuable window on this issue: conditional if-clauses (and when-clauses, arguably a type of correlative) appear before the age of 3 (Reilly 1986), well before wh-based relative clauses (cf.
Guasti & Shlonsky 1995; Diessel 2004; Clauss 2017).
Contribution: If correlatives are a type of conditionals, they are predicted to be mastered earlier than the corresponding relative clauses. If, on the other hand, they are a subtype of relatives, and the major difference from genuine relatives is their location with respect to their host clause
(left-peripheral for correlatives, right-peripheral for relatives), relatives are expected to be mastered earlier, as they occur in the position default/more frequent for subordinate clauses (Diessel 2004). We provide experimental evidence from Czech, a language which productively uses both correlatives and relatives, for the former position.
Design and materials: We designed a sentence repetition experiment manipulating (within items and subjects) the construction type (correlative vs. relative). The type of relative used was the so-called light-headed relative (LHR; Citko 2004), which facilitates direct comparison with the kind of correlative productive in Czech; an item in both conditions is illustrated in (1); note that superficially, the conditions differ only in word order. We constructed 16 items (varying between items the wh-words for 'where', 'who', and 'what', the latter two in both subject and object positions) and complemented them with 10 fillers (interrogatives, conditionals, nominally headed relatives); the items were prerecorded, distributed on two lists (Latin square), and pseudorandomized. The participants were also given picture vocabulary and grammar comprehension tests (correlation: 0.49; grammar scores was used for analyses).
Participants and procedure: 30 children (2;10-4;3) took part, of whom 29 completed the grammar test; see Tab. 2. The data were collected in kindergardens in and around Prague. Repetitions were recorded, transcribed, and annotated using a 2-0 scale (2 = accurate repetition; 1 = grammatical with 1-2 errors; 0 = ungrammatical or > 2 errors).
Results: We modeled (cumulative link mixed models; clmm of R/ordinal) the effect of construction and its interaction with the grammar score on the repetition score. The model (Fig./Tab. 1) revealed that relatives (LHR) are significantly more difficult to repeat than correlatives (Cor)
(𝑧 = -2.217, 𝑝 = .027) and the difference decreases with grammar competence (marginally significant interaction; 𝑧 = 1.871, 𝑝 = .061). An additional model that includes the filler data (to be reported) shows that relatives are more difficult to repeat than conditionals (treated as baseline)
(nominally headed relatives/HRs more so [𝑧 = -3.612] than LHRs [𝑧 = -2.045]), but correlatives show no difference to conditionals (𝑧 = -.753). Overall, the results are consistent with the view that correlatives are on a par with conditionals, despite being rare in both adult input and child output (LHRs are twice as frequent in the Czech CHILDES; Chromá et al. 2020).