Scott McDonald

Eye Movements Reveal the On-Line Computation of Lexical Probabilities During Reading

Skilled readers are able to derive meaning from a stream of visual input with remarkable efficiency. In this article, we present the first evidence that statistical information latent in the linguistic environment can contribute to an account of reading behavior. In two eye-tracking studies, we demonstrate that the transitional probabilities between words have a measurable influence on fixation durations, and using a simple Bayesian statistical model, we show that lexical probabilities derived by combining transitional probability with the prior probability of a words occurrence provide the most parsimonious account of the eye movement data. We suggest that the brain is able to draw upon statistical information in order to rapidly estimate the lexical probabilities of upcoming words: a computationally inexpensive mechanism that may underlie proficient reading.

Low-level predictive inference in reading: the influence of transitional probabilities on eye movements.

We report the results of an investigation into the ability of transitional probability (word-to-word contingency statistics) to account for reading behaviour. Using a corpus of eye movements recorded during the reading of newspaper text, we demonstrate both the forward [P(n/n-1)] and backward [P(n/n+1)] transitional probability measures to be predictive of first fixation and gaze durations: the higher the transitional probability, the shorter the fixation time. Initial fixation position was also affected by the forward measure; we observed a small rightward shift for words that were highly predictable from the preceding word. Although transitional probability is sensitive to word class, with function words being generally more predictable from their context than content words, the measures accounted equally well for the data for both classes.

An Anatomically Constrained, Stochastic Model of Eye Movement Control in Reading

This article presents SERIF, a new model of eye movement control in reading that integrates an established stochastic model of saccade latencies (LATER; R. H. S. Carpenter, 1981) with a fundamental anatomical constraint on reading: the vertically split fovea and the initial projection of information in either visual field to the contralateral hemisphere. The novel features of the model are its simulation of saccade latencies as a race between two stochastic rise-to-threshold LATER units and its probabilistic selection of the target for the next saccade. The model generates simulated eye movement behavior that exhibits important characteristics of actual eye movements made during reading; specifically, simulations produce realistic saccade target distributions and replicate a number of critical reading phenomena, including the effects of word frequency on fixation durations, the inverted optimal viewing position effect, the trade-off between first and second fixation durations of refixated words, and the dependence of parafoveal preview benefit on eccentricity.

Rethinking the Word Frequency Effect: The Neglected Role of Distributional Information in Lexical Processing

Attempts to quantify lexical variation have produced a large number of theoretical and empirical constructs, such as Word Frequency, Concreteness, and Ambiguity, which have been claimed to predict between-word differences in lexical processing behavior. Models of word recognition that have been developed to account for the effects of these variables have typically lacked adequate semantic representations, and have dealt with words as if they exist in isolation from their environment. We present a new dimension of lexical variation that is addressed to this concern. Contextual Distinctiveness(CD), a corpus-derived summary measure of the frequency distribution of the contexts in which a word occurs, is naturally compatible with contextual theories of semantic representation and meaning. Experiment 1 demonstrates that CD is a significantly better predictor of lexical decision latencies than occurrence frequency, suggesting that CD is the more psychologically relevant variable. We additionally explore the relationship between CD and six subjectively-defined measures: Concreteness, Context Availability, Number of Contexts, Ambiguity, Age of Acquisition and Familiarity and find CD to be reliably related to Ambiguity only. We argue for the priority of immediate context in determining the representation and processing of language.

Optokinetic therapy improves text reading in patients with hemianopic alexia A controlled trial

Objective: An acquired right-sided homonymous hemianopia can result in slowed left-to-right text reading, called hemianopic alexia (HA). Patients with HA lack essential visual information to help guide ensuing reading fixations. We tested two hypotheses: first, that practice with a visual rehabilitation method that induced small-field optokinetic nystagmus (OKN) would improve reading speeds in patients with HA when compared to a sham visual rehabilitation therapy; second, that this therapy would preferentially affect reading saccades into the blind field. Methods: Nineteen patients with HA were entered into a two-armed study with two therapy blocks in each arm: one group practiced reading moving text (MT) that scrolled from right to left daily for two 4-week blocks (Group1), while the other had sham therapy (spot the difference) for the first block and then crossed over to MT for the second. Results: Group 1 showed significant improvements in static text reading speed over both therapy blocks (18% improvement), while Group 2 did not significantly improve over the first block (5% improvement) but did when they crossed over to the MT block (23% improvement). MT therapy was associated with a direction-specific effect on saccadic amplitude for rightward but not leftward reading saccades. Conclusion: Optokinetic nystagmus inducing therapy preferentially affects reading saccades in the direction of the induced (involuntary) saccadic component. This is the first study to demonstrate the effectiveness of a specific eye movement based therapy in patients with hemianopic alexia (HA) in the context of a therapy-controlled trial. A free Web-based version of the therapy used in this study is available online to suitable patients with HA.

Determinants of adjective-noun plausibility

This paper explores the determinants of adjective-noun plausibility by using correlation analysis to compare judgements elicited from human subjects with five corpus-based variables: co-occurrence frequency of the adjective-noun pair, noun frequency, conditional probability of the noun given the adjective, the log-likelihood ratio, and Resniks (1993) selectional association measure. The highest correlation is obtained with the co-occurrence frequency, which points to the strongly lexicalist and collocational nature of adjective-noun combinations.

Hemispheric asymmetries in the split-fovea model of semantic processing

We report a series of neural network models of semantic processing of single English words in the left and the right hemispheres of the brain. We implement the foveal splitting of the visual field and assess the influence of this splitting on a mapping from orthography to semantic representations in single word reading. The models were trained on English four-letter words, presented according to their frequency in all positions encountered during normal reading. The architecture of the model interacted with the training set to produce processing asymmetries comparable to those found in behavioral studies. First, the cueing effects of dominant and subordinate meanings of ambiguous words were different for words presented to the left or to the right of the input layer. Second, priming effects of groups of related words were stronger in the left input than the right input of the model. These effects were caused by coarser-coding in the right half compared with the left half of the model, an emergent effect of the split model interacting with informational asymmetries in the left and right parts of words in the lexicon of English. Some or all of the behavioral data for reading single words in English may have a similar origin.

LATER predicts saccade latency distributions in reading

When saccades are evoked by suddenly presented visual stimuli, the stochastic distribution of their reaction times is typically recinormal, in conformity with the LATER model of decision-making, sometimes with an additional sub-population of early responses. In the real world, saccades are more often spontaneous responses to static features of the visual world; nevertheless, the time between the end of one saccade and the start of the next may be regarded as a reaction time to the new image that is presented to the retina. The distribution of these times is qualitatively similar to that of evoked saccades, but typically with a longer median and more of the early responses. Here, we analyse the statistical distributions of fixation times from a large database of spontaneous saccades made while reading and show that the distributions are altered in characteristic ways by particular features of the current fixation: likely familiarity of the currently fixated word, and its proximity to the preceding fixation. These alterations are of the kind predicted by LATER: familiarity appears to influence the mean rate at which the decision signal approaches completion, whereas proximity to the previous fixation, presumably because it provides partial prior information about the upcoming word, appears to increase prior probability. We conclude that spontaneous saccades may be successfully described by the same decision-making model that can be used for evoked ones.

Parafoveal preview benefit in reading is only obtained from the saccade goal

Previous research has demonstrated that reading is less efficient when parafoveal visual information about upcoming words is invalid or unavailable; the benefit from a valid preview is realised as reduced reading times on the subsequently foveated word, and has been explained with reference to the allocation of attentional resources to parafoveal word(s). This paper presents eyetracking evidence that preview benefit is obtained only for words that are selected as the saccade target. Using a gaze-contingent display change paradigm (Rayner, K. (1975). The perceptual span and peripheral cues in reading. Cognitive Psychology, 7, 65-81), the position of the triggering boundary was set near the middle of the pretarget word. When a refixation saccade took the eye across the boundary in the pretarget word, there was no reliable effect of the validity of the target word preview. However, when the triggering boundary was positioned just after the pretarget word, a robust preview benefit was observed, replicating previous research. The current results complement findings from studies of basic visual function, suggesting that for the case of preview benefit in reading, attentional and oculomotor processes are obligatorily coupled.

Evaluating Smoothing Algorithms against Plausibility Judgements

Previous research has shown that the plausibility of an adjective-noun combination is correlated with its corpus co-occurrence frequency. In this paper, we estimate the co-occurrence frequencies of adjective-noun pairs that fail to occur in a 100 million word corpus using smoothing techniques and compare them to human plausibility ratings. Both class-based smoothing and distance-weighted averaging yield frequency estimates that are significant predictors of rated plausibility, which provides independent evidence for the validity of these smoothing techniques.