| Vocabulary knowledge is one of the crucial factors for students' general academic comprehension (Anderson, 2008; Grabe, 2004 & 2009; Laufer, 1992; Nation, 2001; Qian, 2002). Thus, researchers (Coxhead, 2000; Garnder & Davis, 2013) have created academic vocabulary lists which other researchers and material developers have implemented into their studies or materials for instruction in English as a second language (ESL) and English for specific purpose (ESP). Although the academic vocabulary lists have been widely used for research and language teaching, few studies have probed into the representativeness of the lists. Miller and Biber (2015) pointed out this gap and suggested that the reliability of word lists would be investigated by simply dividing a corpus into random smaller sub-corpora.;This study investigated the effect of word selection criteria on the reliability of an academic vocabulary list. Four questions are addressed. First, how many words would be academic words in research articles? Second, to what extent did the criterion of register frequency contribute to a reliable vocabulary list? Third, if different sets of criteria including register frequency, range, discipline measurement and dispersion were implemented to select words, how many lemmas would remain in vocabulary lists of 1,000 lemmas? Fourth, which set of criteria would extract the most reliable academic vocabulary list?;A thirty-million-word corpus was created with 1,800 academic articles across nine disciplines. Then the corpus was divided into two sub-corpora and thirty cases of criteria were applied to each sub-corpus for vocabulary list creation. The replicability of the vocabulary lists was examined to find the case of criteria extracting the most reliable list.;The results showed that the reliability of a vocabulary list was differentiated based on the thirty cases. The highest reliability was drawn with the combination of 1.5 Register Frequency, 3.0 Discipline Measurement and .8 Dispersion. The range did not influence list reliability. It was concluded that the most desirable condition to create a reliable academic vocabulary list resulted from the combination of the lenient register frequency and discipline measurement, and a strict dispersion. The implications for researchers, material developers, teachers and learners, and limitations were also addressed. |
