Word Length effects in visual word recognition

Word Length effects in visual word recognition The way letters in words are processed in our orthographic system is subject to considerable controversy. Theorists argue that letters in words are processed either in series or in parallel. This study attempts to resolve this debate by examining the effect of Word Length on decision speeds for words in a lexical decision task. A multiple regression analysis which included other linguistic descriptors such as Familiarity and Age of Acquisition was performed. The analysis indicated that Word Length is not a determinant of decision speed which implies that letters in words are processed via a parallel processing system. Further research needs to be conducted in this area in order to answer the research question. Implications of these findings in relation to word recognition models are discussed. Introduction Are letters within a word processed one at a time, in series, or does the skilled reader process all the letters in a word in parallel? Despite the deceptively simple nature of this question, researchers are still in disagreement as to how exactly we execute such a complex process. Early models of word recognition argued that words are read letter-by-letter serially from left to right (Gough, 1972). This model is consistent with Herons (1957) findings that when English speaking participants are briefly presented with a sequence of letters, they are more accurate at recalling left-hand letters compared to right-hand letters. This observed effect is reversed in readers of Yiddish (who read from right-to-left) which supports the notion that letters are processed in series (Mishkin Forgays, 1952; Sperling, 1963). The serial letter recognition model has been criticised for the fact that it fails to explain the Word Superiority Effect, which demonstrates how readers are better able to identify letters in the context of a word than in isolation (Paap et al., 1982). Over the past few decades, psychologists have therefore developed parallel distributed models (PDP) of reading which assume that letters in words are encoded simultaneously and draw heavily on what we know about the human neural system (Seidenberg Harm, 1991). http://www.microsoft.com/typography/ctfonts/science_of_word_recognition_files/Figure4.gif Figure ; Rumelhart McClellands Interactive-Activation Model (1981) An example of a PDP model is shown in Figure 1. According to this model, when a reader is presented with a word such as WORK, each letter either stimulates or inhibits different feature detectors (e.g. a circular shape for O, or horizontal and vertical bars for K,). These feature detectors then stimulate or inhibit different letter detectors, which finally stimulate or inhibit different word detectors. Each activated connection carries different weights, and therefore the word WORK in Figure 1 is activated more than any other word and is the one recognized by a reader. Although most recent findings are generally assumed to support the parallel processing hypothesis; a model which has recently gained attention is the Self-Organizing Acquisition and Recognition (SOLAR) model which implies an element of serial processing (Davis, 2010). Davis argues that visual word recognition relies on the formation of a spatial code to understand the orthographic input which is formed by a rapid left-to-right scan across letter representations that combines letter identity information with letter position information. A common way of studying how letters in words are processed is by measuring Word Length (WL) effects in Lexical Decision Tasks (LDT). In these tasks participants are presented with a word/non-word and their correct Decision Speeds (DS) are measured. The SOLAR model strongly predicts a WL effect in such tasks as the length of vector across the input layer should impact processing time. In contrast, parallel models such as the Interactive-Activation model predict no WL effects as it shouldnt matter how many letters there are in a word because they are all processed simultaneously. Several studies have examined WL effects however the findings are inconclusive with some observing significant WL effects; Forster Chambers ,1973; Whaley ,1978; ORegan Jacobs ,1992; Gilhooly Logie, 1982, whilst others failed to find a significant effect of WL; Frederiksen Kroll , 1976; Richardson ,1976. It is therefore clear that more research needs to be conducted in this area in order to resolve these conflicting findings. One of the main problems faced in investigating WL effects is the observed collinearity between over properties of words such as Familiarity, Age of Acquisition and Frequency which have been observed as predictors of DS in previous word recognition tasks (Gernsbacher, 1984; Gerhand Barry, 1998). Therefore when conducting an experiment which seeks to examine WL effects it is also important to account for other linguistic properties. A multiple regression analysis allows the experimenter to calculate the unique contribution of each linguistic predictor to the variation in DS and so is a popular method employed by many researchers in this area of research where collinearity amongst other variables is so high. The aim of this experiment therefore is to examine the effect of WL on DS in a LDT in which the linguistic stimuli has been generated taking into consideration; WL, familiarity, frequency, Age of Acquisition (AoA), Number of Neighbours (NoN) and Imageability, which have all been found to predict DS in previous research. If WL emerges as a significant predictor of DS then this is evidence for serial word processing and thus provides experimental support for the SOLAR model (Davis, 2010), whereas if no WL effects are found then this supports the assumptions underlying PDP models. Method Participants 12 second year undergraduate Psychology students (M= 19.75 years, m=2, f=10) from the University of Bristol with normal/corrected vision were recruited via an opportunity sample. Materials A computer based LDT was created using DMDX software. The stimuli in the task comprised of 150 words (all of which were nouns) and 150 non-words which were manipulated by the experimenter by changing a vowel in a word to another vowel or a consonant to a different consonant (Appendix 1). Linguistic stimuli were generated using the MRC psycholinguistic database (Coltheart, 1981) which selects lists of words, together with linguistic descriptors based on a number of selection criteria. Table 1 shows the linguistic descriptors and selection criterion of stimuli used in this experiment. Design This experiment used a within subjects repeated measures design. The LDT comprised 300 trials split into 6 test blocks; each block contained 25 words and 25 non-words which were randomly intermixed. Linguistic stimuli were displayed on the screen for 100ms. There was a 50ms break in between trials and participants had 2000ms to make their decision before the next trial commenced. Participants were given a break in between blocks to reduce fatigue effects. The experiment lasted approximately 30 minutes. Procedure In the computer based LDT Words or Non-words were presented individually to participants in the centre of a 1414 computer monitor in size 14 bold black Arial font with white background. Participants were instructed to press the Right Hand Shift button if they saw a Word or the Left Hand Shift button if they saw a Non-Word. Participants were informed of their confidentiality of results. Data analysis Participants correct Decision Speeds (ms) were recorded. Non-words were used as fillers and were not statistically analysed. Words with error rates of over 25% were excluded from the data set. Participants who responded incorrectly to over 50% of stimuli were excluded. Table 2 shows the 16 words that were removed from the data set as they failed to meet the criteria. The final data set is based on N=134 words and 12 participants. Results In the LDT, speeds of correct responses to words were recorded (M=669.62ms, SD=81.54, Error rate=8.28%). The experimental hypothesis aimed to examine the relationship between DS and WL; no significant correlation was found between WL and DS (r=.04), which is visually presented in Figure 1. Figure 1; Scatter graph demonstrating the non-significant correlation between Decision Speed (ms) and Word Length. A significant negative correlation was found between DS and Familiarity, which suggests that participants respond faster to words that are highly familiar (r= -.34, p Word length was found to be intercorrelated with AoA, Imageability and NoN (Table 3) which highlights a potential problem of collinearity. Correlations among measures of word attributes and decision speed, together with summary statistics. N=134 words. A simultaneous and a stepwise multiple regression analysis was conducted on the correlation matrix to assess the contribution of the independent variables in predicting DS. In the simultaneous analysis all variables were entered into the regression at the same time and the effects of all other variables were partialled out from each other (Table 4). The overall fit of the model was R2=.121, F(6,132)= 3.24, p Simultaneous multiple regression analysis on correct response speeds in the LDT. Next, a stepwise multiple regression analysis was conducted (Table 5). This analysis differs from the simultaneous analysis in the way that it enters variables into the regression one at a time on the basis of their contribution to improving the predictive power of the equation. The findings of the stepwise regression analysis was essentially the same as the simultaneous regression with Familiarity emerging as the only variable in predicting DS (F1,132= 18.693, p Stepwise multiple regression analysis on correct response times in the LDT. Discussion This experiment failed to demonstrate that Word Length (WL) is an effective factor in visual word recognition. There was no correlation between WL and decision speeds (DS) on the lexical decision task (LDT) which refutes the notion that letters in words are processed in series and therefore questions the theory underlying the SOLAR model (Davis, 2010) which postulates that we recognize words by performing a left-to-right scan across letters and so would have strongly predicted WL effects in the performed study. The results from this experiment therefore imply that letters in words are processed in parallel and thus support the assumptions embedded in parallel and connectionist models of reading (Seidenbery Harm, 1999; Rumelhart McClelland, 1981). The only linguistic predictor to emerge as an effective factor in decision speeds was familiarity, which is consistent with previous studies that have reported its contribution to the efficiency of word processing (Gilhooly Logie, 1982; Connine et al., 1990). The fact that familiarity emerged as a predictor of DS may in fact provide support for parallel processing. It can be argued that if a participant is more familiar with a word, this suggests that they may have formed some internal representation of the word within their mental lexicon and therefore recognize it faster by processing letters in parallel. In this study WL was significantly correlated with several of the other linguistic descriptors such as AoA and NoN which highlights the problem of collinearity and so it is difficult to draw any clear conclusions from our data. Another factor which may have affected our results is the fact that WL was restricted to 9 letters. In a study conducted by Gilhooly Logie (1982) they used words of various lengths and found WL to be the strongest predictor of DS. This raises an interesting research question as to whether there is a maximum capacity of letters that we can process in parallel and then perhaps after a certain length, the demands on our processing system are too large and so we have to rely on other processes. It is therefore plausible that visual word recognition may involve a combination of both serial and parallel processing depending on WL, which challenges the assumptions of current models of word recognition. It would be interesting to expand upon this study by making several alterations; firstly by abolishing the word letter restriction to see if we observe WL effects, and also by increasing the number of linguistic stimuli used and the number and age range of participants. Furthermore it would be of interest to examine saccade movements whilst participants are doing the LDT which may provide further insights into whether words are processed in parallel or series. One of the aims of this paper was to try and clear up some of the confusion in the literature surrounding serial and parallel processing models; unfortunately we were somewhat unsuccessful in doing so. However the findings from this study do provide several interesting thought questions and future research expansions and suggest that maybe word recognition relies on elements of serial and parallel processing. References; Coltheart, M (1981). The MRC Psycholinguistic Database, Quarterly Journal of Experimental Psychology, 33A, 497-505. Connine, C.M., Mullennix, J., Shernoff, E., Yelen, J. (1990). Word familiarity and frequency in visual and auditory word recognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16 (6), 1084-1096. Davis, C. J. (2005). N-Watch: A program for deriving neighbourhood size and other psycholinguistic statistics. Behavior Research Methods,. 37, 65-70. Forster, K. I., Chambers, S. M. (1973). Lexical access and naming time. Journal of Verbal Learning and Verbal Behavior, 12,627-635. Fredriksen, J. R., Kroll, J. F. (1976). Spelling and sound: Approaches to the internal lexicon. Journal of ExperimentalPsychology: Human Perception and Performance, 2, 361-379. Gerhand, S., Barry, C. (1998). Word frequency effects in oral reading are not merely age-of-acquisition effects in disguise. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 267-283. Gilhooly, K.J. and Logie, R.H. (1980). Age of acquisition, imagery, concreteness, familiarity and ambiguity measures for 1944 words. Behaviour Research Methods and Instrumentation, 12, 395-427. Gilhooly, K. J., Logie, R.H. (1981). Word Age-Of-Acquisition And Lexical Decision Making. Acta Psychologica. 50; 21-34. Gough, P.B. (1972). One second of reading. In Kavanagh Mattinglys Language by ear and by eye. Cambridge, MA: MIT Press. Heron, W. (1957). Perception as a function of retinal locus and attention. American Journal of Psychology, 70, 38-48. KuÄ era, H., Francis, W. (1967). Computational analysis of presentday American English. Providence, RI: Brown University Press. McClelland, J. L. Rumelhart, D. E. (1981). An interactive activation model of context effects in letter perception: Part 1. An account of Basic Findings. Psychological Review, 88, 375-407 Mishkin, N., Forgays, D. G. (1952). Word recognition as a function of retinal locus. Journal of Experimental Psychology, 43,43-48. ORegan, J. K., Jacobs, A. M. (1992). Optimal viewing position effect in word recognition: A challenge to current theory.Journal of Experimental Psychology: Human Perception and Performance, 18, 185-197. Paap, K. R., Newsome, S. L., McDonald, J. E., Schvaneveldt, R. W. (1982). An activation-verification model for letter and word recognition: The word-superiority effect. Psychological Review, 89 (5), 573-594. Richardson, J. T. (1976). The effects of stimulus attributes upon latency of word recognition. British Journal of Psychology, 67,315-325. Seidenberg, M.S., Harm, M.W (1999). Phonology, Reading acqusition, and dyslexia: insights from connectionist models. Psychological review, 106 (3), 491-528. Sperling, G. (1963). A model for visual memory tasks. Human Factors, 5, 19-31. Woodworth, R.S. (1938). Experimental psychology. New York; Holt. Toglia, M.P. and Battig, W.R. (1978). Handbook of Semantic Word Norms. New York: Erlbaum. Whaley, C. P. (1978). Word-nonword classification time. Journal of Verbal Learning and Verbal Behavior, 17, 143-154.