Towards Understanding Text Factors in Oral Reading

Proceedings of NAACL-HLT 2018, pages 2143–2154 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Towards understanding text factors in oral reading Anastassia Loukina, Van Rynald T. Liceralde, Beata Beigman Klebanov Educational Testing Service Princeton, NJ, USA {aloukina,vliceralde,bbeigmanklebanov}@ets.org Abstract Using a case study, we show that variation in oral reading rate across passages for pro- fessional narrators is consistent across readers and much of it can be explained using features of the texts being read. While text complexity is a poor predictor of the reading rate, a sub- stantial share of variability can be explained by timing and story-based factors with perfor- mance reaching r=0.75 for unseen passages and narrator. 1 Introduction Listening to and performing oral reading are activ- ities that permeate daily life, from parents reading aloud to young children, through reading instruc- tion in elementary school, to audiobook narrations increasingly chosen by adults as the form of book- reading that fits in a busy schedule. Oral reading is also used in assessment of language skills for children and language learners, and in professions such as teaching and news broadcasting. Reading rate is a common metric used to con- trol or evaluate oral reading. It is usually com- puted as a number of words read per minute, and is used in many applications. For example, re- search in second language acquisition has consid- ered both optimal reading rates for listening mate- rials aimed at English language learners and read- ing rates that ensure the highest comprehensibility of accented speech (Munro and Derwing, 1998). Speech rate is a standard feature in systems for automated scoring of second language proficiency (Higgins et al., 2011) including read aloud tasks (Zechner et al., 2012; Evanini et al., 2015). Read- ing rate is also one of the main measures used to assess the fluency of oral reading (Hasbrouck and Tindal, 2006). The assumption underlying these uses is that reading rate is a property of the reader (or con- trolled by the reader). However, variation in read- ing rate across different passages for the same readers has also been reported (Foulke, 1968; Tau- roza and Allison, 1990; Ardoin et al., 2005; Comp- ton et al., 2004; Beigman Klebanov et al., 2017). Improving the understanding of the properties of oral reading, such as reading rate, is thus an important theoretical goal. We also have a spe- cific practical reason to study text-based variation in reading rate. We are developing an interven- tion for improving literacy that would encourage sustained reading by having the student read aloud multiple passages from an engaging novel-length book, taking turns with others. While it is techni- cally easy to compute reading rate by timing the readers, if a reader’s rate across different texts is not stable given his current reading skill, it is not clear that tracking the rate over time would yield a valid measurement of improvement in skill. How- ever, if such variation is systematically dependent on the text being read, rather than a …

Content truncated to 500 words