We used this model to explain human performance on visual search as well as word recognition tasks. It does not capture any information about specialized detectors for longer strings, or about other lexical or linguistic factors. In contrast to an influential proposal that requires neurons tuned to letter combinations ( Dehaene et al., 2005, 2010), our model only assumes neurons tuned for letter shape and retinal position, as observed in high-level visual cortex ( Lehky and Tanaka, 2016). Accordingly, we created neural responses to letter strings as a linear sum of single letter responses. Second, the neural response to multiple objects is an average of the response to the individual objects, a phenomenon known as divisive normalization ( Zoccolan et al., 2005 Ghose and Maunsell, 2008 Zhivago and Arun, 2014). Accordingly we used visual search for single letters to create artificial neurons tuned for single letters. First, images that are perceptually similar elicit similar activity in single neurons ( Op de Beeck et al., 2001 Sripati and Olson, 2010a Zhivago and Arun, 2014). We drew upon two well-established principles of object representations in high-level visual cortex. To overcome this confound, we asked whether visual search involving letter strings can be explained using a neurally plausible model containing only visual factors. However, subjects may have been reading during visual search, thereby activating non-visual lexical or linguistic factors. This example suggests that word reading could be explained by purely visual processing as indexed by visual search. This difference in visual similarity ( Figure 1C) explains why transposing the middle letters renders a word easier to read than transposing its edge letters. It can be seen that finding OFRGET is easy among FORGET whereas finding FOGRET is hard ( Figure 1B). An example visual search array containing two oddball targets is shown in Figure 1B. This task does not require any explicit reading and is driven by shape representations in visual cortex ( Sripati and Olson, 2010a Zhivago and Arun, 2014). To probe purely visual processing, we devised a visual search task in which subjects had to find an oddball target among distractors. Here, we hypothesized that word reading is enabled by a purely visual representation. Despite these insights, it is not clear how these factors combine, what their distinct contributions are, and more generally, how word representations relate to letter representations. Word reading is also easier for words with frequent bigrams or trigrams, for frequent words and for shuffled words that preserve intermediate units such as consonant clusters or morphemes ( Norris, 2013 Grainger, 2018). Word reading is easy when similar shapes are substituted ( Perea et al., 2008 Perea and Panadero, 2014), when the first and last letters are preserved ( Rayner et al., 2006), when there are fewer transpositions ( Gomez et al., 2008) and when word shape is preserved ( Norris, 2013 Grainger, 2018). Reading a word or a jumbled word can be influenced by a variety of factors ( Norris, 2013 Grainger, 2018). What makes a jumbled word easy or hard to read? This question has captured the popular imagination through demonstrations such as the Cambridge University effect ( Rawlinson, 1976 Grainger and Whitney, 2004), depicted in Figure 1A. Reading is a recent cultural invention, yet we are remarkably efficient at reading words and even jmulbed wrods ( Figure 1A).
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |