Marslen-Wilson and Warren (1994) found that lexical-decision latencies to cross-spliced word sequences whose initial CV portion had been excised from an existing word (e.g., jo(g)b) did not differ from latencies to word sequences excised from a nonword (e.g., jo(d)b). They interpreted this result as strong evidence against models incorporating lexical competition via lateral inhibition. McQueen, Norris, and Cutler (in press), who replicated this result, argued that lexical competition between 'jog' and 'job' is resolved before the activation level for the target sequence 'job' reaches the threshold that triggers lexical-decision responses. However, to simulate the lexical-decision data, their model depends on choosing a threshold within an extremely restricted range. Here, we argue that mean lexical-decision latencies may not be an appropriate measure of activation for the target 'job', because the activation of its competitor 'jog' also influences the probability of responding 'yes', making it difficult to relate lexical decisions to the underlying activation functions. By contrast, tracking eye movements can provide a continuous measure of lexical activation over time. Participants' eye movements to pictured objects were recorded as they followed instructions to click on one of the four objects (e.g., 'click on the net'). Allopenna, Magnuson, and Tanenhaus (1998) demonstrated that the proportion of fixations to each picture over time, as the target word is heard, can be mapped onto lexical-activation functions using a simple linking hypothesis.

In the present study, the name of the target object was cross-spliced from another token of itself (ne(t)t), from another word (ne(ck)t), or from a nonword (ne(p)t). In Experiment 1, the competitor picture (neck) was displayed along with the target picture and two unrelated pictures. Fixations to the target over time indicated a fast rise in the ne(t)t condition, intermediate in the ne(p)t condition, and slowest in the ne(ck)t condition. The competitor picture was fixated most in the ne(ck)t condition, intermediate in the ne(p)t condition and least often the ne(t)t condition. The time course of fixations to the target and competitor pictures mirrored predictions generated from activations in a TRACE-style interactive activation model, using the Allopenna et al. (1998) linking hypothesis. Eliminating lexical competition significantly reduced the fit between the model and the data. In Experiment 2, the competitor picture was not displayed. Under these conditions, the model predicted that the time course of fixations to the target would be similar in the ne(p)t and ne(t)t conditions and slower in the ne(ck)t condition. The data closely matched these predictions. Finally, the model simulates the lexical-decision data across a range of response thresholds under the assumption that a 'yes' response is triggered probabilistically when either the target or competitor reaches threshold.