Authors:
Pavlinušić Vilus, Eva, eva.pavlinusic.vilus@erf.unizg.hr, University of Zagreb Faculty of Education and Rehabilitation Sciences
Matić Škorić, Ana, ana.matic@erf.unizg.hr, University of Zagreb Faculty of Education and Rehabilitation Sciences
Cergol, Kristina, kristina.cergol@ufzg.hr, University of Zagreb Faculty of Teacher Education
Palmović, Marijan, marijan.palmovic@erf.unizg.hr, University of Zagreb Faculty of Education and Rehabilitation Sciences
Keywords: vocabulary test; lexical decision task; L1 speakers; reading comprehension
Abstract:
Numerous studies have shown that vocabulary knowledge is a strong predictor of reading comprehension. Unfortunately, standardised vocabulary tests are typically expensive, time-consuming and require one-to-one communication between experimenter and participant. The first quick, freely available, digitalised vocabulary test based on lexical decision task (LDT) was developed by Lemhöfer and Broersma (2012) for English as L2. Similar L2 vocabulary tests have been developed for several other languages, while no attempt has been made to develop such tests for L1. The aim of the present study was to develop a LexTALE-like test for Croatian L1 speakers. In previous studies, L1 speakers’ results on LDT were at ceiling level (e.g., Amenta et al., 2020). To overcome this, in the present study words were selected from a lower range of relative frequencies. Ninety Croatian words were selected from the Croatian web corpus Classla-web.hr (Ljubešić et al., 2024). Additional 90 Croatian words were used as input for the pseudoword generator Wuggy (Keuleers & Brysbaert, 2010). 233 native Croatian speakers (university students) performed LDT for the 180 items and filled in a socio-demographic and linguistic background questionnaire. Following the analysis described for other LexTALE-like tests, the quality of the items was assessed using point-biserial correlation and item-response theory (IRT) analysis. In general, the participants’ scores were high, but no ceiling effect was observed (% correct: M = 84.43 for words, M = 82.38 for pseudowords). The point-biserial correlation analysis returned no items with negative correlation coefficients. The coefficients ranged from 0.01 to 0.53 for words and from 0.14 to 0.57 for pseudowords. The IRT analysis revealed that the item difficulty and discrimination parameters were low across all items, due to low variance in the results. Since the selection of the final set of 60 words and 45 pseudowords is not possible based on the current IRT results, another round of item quality testing is currently being performed with Croatian speakers of more varying education levels as participants.