From Strings to Other Things: Linking the Neighborhood and Transposition Effects in Word Reading

Proceedings of the 22nd Conference on Computational Natural Language Learning (CoNLL 2018), pages 75–85 Brussels, Belgium, October 31 - November 1, 2018. c⃝2018 Association for Computational Linguistics 75 From Strings to Other Things: Linking the Neighborhood and Transposition effects in Word Reading St´ephan Tulkens and Dominiek Sandra and Walter Daelemans CLiPS - Computational Linguistics Group Department of Linguistics University of Antwerp {stephan.tulkens,dominiek.sandra,walter.daelemans}@uantwerpen.be Abstract We investigate the relation between the trans- position and deletion effects in word read- ing, i.e., the finding that readers can success- fully read “SLAT” as “SALT”, or “WRK” as “WORK”, and the neighborhood effect. In particular, we investigate whether lexi- cal orthographic neighborhoods take into ac- count transposition and deletion in determin- ing neighbors. If this is the case, it is more likely that the neighborhood effect takes place early during processing, and does not solely rely on similarity of internal representations. We introduce a new neighborhood measure, rd20, which can be used to quantify neigh- borhood effects over arbitrary feature spaces. We calculate the rd20 over large sets of words in three languages using various feature sets and show that feature sets that do not allow for transposition or deletion explain more vari- ance in Reaction Time (RT) measurements. We also show that the rd20 can be calculated using the hidden state representations of an Multi-Layer Perceptron, and show that these explain less variance than the raw features. We conclude that the neighborhood effect is un- likely to have a perceptual basis, but is more likely to be the result of items co-activating after recognition. All code is available at: www.github.com/clips/conll2018 1 Introduction Despite their many disagreements and differences, a common thread among many models of word reading is that they attempt to explain differ- ences in reading speeds by assuming that sim- ilarity between words modulate reading speed. There is good reason for this assumption; many experiments have shown that responses on tri- als are modulated by a word’s similarity to other words, be it semantic (Rodd et al., 2002, 2004), orthographic (Andrews, 1997; Perea and Pollat- Time U N C E R T A I N T Y Span Span Snap Spin Span Span Snap Spin Spin Neighborhood Neighborhood Figure 1: This diagram shows the two positions con- trasted in this paper. The left model is the early model, in which the neighborhood effect arises before percep- tual uncertainty is resolved; this causes transposition and substitution neighbors to count as neighbors. In the late model, the neighborhood effect only arises af- ter perceptual uncertainty is resolved, and transposi- tion and substitution neighbors do not count towards the neighborhood. sek, 1998), or phonological similarity (Van Orden, 1987; Rastle and Brysbaert, 2006). In psycholinguistic research on word reading, this has led to the common practice of including a measure of orthographic neighborhood similar- ity as a control variable, as these neighborhood measures explain variance in word reading even when controlling for frequency and length (Colt- heart, 1977; Yarkoni et al., 2008). Orthographic neighborhood measures are …

Content truncated to 500 words