The present studies addressed the issue of whether the phonological mediation of visual word recognition proceeds through an assembled or an addressed representation. In Experiment 1, subjects judged whether pairs of words were semantically related. Both homophone and "false homophone" stimuli were used. The set of "false homophones" consisted of words with the following characteristics: (1) They have neighbors that share its orthographic body but not its pronunciation (B$\underline{\rm EARD}$ - H$\underline{\rm EARD}$ and (2) when an alternate pronunciation of the body is attached to the pronunciation of the onset, another word is produced (e.g., if BEARD were pronounced like HEARD, then the word "bird" would result). Experiment 1 demonstrated that reaction times in a semantic relatedness judgment task were longer to homophones (e.g., SAND - BEECH) and "false homophones" (e.g., ROBIN - BEARD) of a semantic associate than to visually similar controls. Subjects also made more errors to homophone pairs than to visually similar controls. Since the false homophone pairs were related through a phonological representation not specified in the word's lexical entry, it was concluded that the phonological representation responsible for the effect was an assembled representation. In a second experiment, a parafoveal preview paradigm was used in order to determine whether the phonological representation integrated across fixations in reading is an assembled or an addressed representation. As in Experiment 1, subjects made semantic relatedness decisions to the stimulus pairs. In the most interesting condition, it was expected that a "biasing" preview (one that specified the spelling-to-sound correspondence that, when applied to the false homophone, would produce the phonological representation of a word related to the other member of the to-be-judged pair) would increase reaction times to the false homophone targets. The failure to observe the expected "biasing" effect is discussed in terms of the characteristics of a neighborhood based on the onset and following vowel cluster of the preview. While the expected preview effect was not observed, the effect of the target words essentially replicated those of Experiment 1. These results argue that phonological mediation proceeds through an assembled phonological representation.

Two experiments investigated how textual context is used to disambiguate lexically ambiguous words. Previous research had suggested that context did not guide access toward the contextually appropriate meaning but instead selected this meaning from multiple activated meanings at a later stage of processing. The experiments reported here developed and used a new technique to explore the very early stages of word recognition. Eye movements were measured during reading. In both experiments a "prime" word was briefly displayed during the initial part of the fixation on the "target" word. Priming was measured by comparing fixation times on targets preceded by semantically Related versus Unrelated primes. Experiment 1 showed significant priming effects at a 35 ms prime duration but not at 30 or 25 ms prime durations. In Experiment 2, lexically ambiguous words were used as primes to targets in short passages and were presented for 35 ms. The type of preceding context (Consistent vs. Inconsistent), type of ambiguous prime (Biased vs. Balanced), and strength of instantiated meaning (Dominant vs. Subordinate) were varied. Only when the preceding context was Consistent with the Dominant meaning of a Biased ambiguous word were significant priming effects obtained. These results supported a model of lexical access in which context does guide access toward the contextually appropriate meaning of an ambiguous word.

The purpose of this research was to investigate the role of scene backgrounds on object identification. Previous research with brief presentation of scenes indicated that scene context facilitated object identification. Experiment 1 replicated this finding with longer display durations. Experiments 2 and 3 were designed to investigate the time course of background information acquisition using an eye movement paradigm. Although the results from Experiment 2 were inconclusive, Experiment 3 demonstrated that scene background information was acquired on both the first and second fixations on a scene. It was concluded that background information acquired from the first and second fixations facilitates object identification.

Two experiments investigated the effect of a congruent sentence context on processing time for a target word. Subjects eye movements were monitored as they read sentences presented on a Cathode Ray Tube. Processing time on the target word, as measured by first fixation duration and gaze duration was shorter when the target was preceded by a congruent sentence context containing a noun and verb that were only weakly related to the target word than when it was preceded by a context in which either the noun or verb had been replaced by a more neutral word. In addition, the fully congruent contexts were modified to either preserve or disrupt the original syntactic relation between the noun and verb. No difference in processing time on the target was observed when these two conditions were compared to one another. These results replicate findings obtained by Duffy et al. (1989) using the Rapid Serial Visual Presentation method to present the sentence contexts and naming time as the dependent measure and demonstrate that these effects generalize to normal silent reading. The observed facilitation for the full congruent contexts exceeded what could be accounted for by summation of activation from the individual lexical items contained in that context. However, there was no evidence that the syntactic relations among the items was critical to this effect. In a second eye movement experiment, lexical and message-level information in the sentence context were manipulated independently to explore potential sources of the facilitation. Evidence of facilitation from the message-level representation of the sentence context was obtained. These results, in conjunction with the previous results from the naming task are interpreted as support for an interactive view of lexical access.

A series of experiments explored the nature of the memory representation of numerical information. Two distributions were presented to each subject as three-digit numbers paired with distribution labels. Three stimulus presentation conditions were used in Experiment 1: a rapid serial presentation as used in Malmi and Samson (1983); a task which requires the subject to retype each stimulus item; and a classification task in which the subject must supply the category name when presented with the stimulus number. In Experiment 1, subjects estimated the averages of the distributions they were presented. In Experiment 2, subjects classified an additional 40 items, chosen to enable discrimination between two classes of models of memory representation. Subjects in Experiment 3 made estimates of the frequency of scores per decade for each distribution. The results strongly favor the category density model (Fried and Holyoak, 1984), a model which assumes that the subject abstracts distributional information and uses a default 'normal' distribution to organize the incoming information. The Nosofsky (1988) exemplar similarity model did not predict subject classification behavior or subject frequency estimation as accurately as the category density model. Reasons for these findings are discussed.

Four different tasks were used to investigate if readers reinstate information which is no longer in focus when it is needed to resolve a break in causal coherence. In five experiments an inference condition was included in which passages contained a causal coherence break which could be resolved by reinstating a backgrounded concept. In Experiment 1, the results of a recognition task provided evidence that readers were able to integrate the targeted cause more easily with the inference version than the control version of the passage, either because of processes occurring while reading or at the time of test. In Experiments 2 and 3, the results of a word naming task provided evidence that the backgrounded cause was reactivated during reading in the inference condition after encountering the coherence break. In Experiment 4, the results of a reading time measure suggested that readers did not only reactivate a single concept, but used this concept to form a new proposition which acted as a cause for the action in the focal sentence. The causal link was maintained in working memory. According to the results of the recall test in the final experiment, the causal link was also included in the long-term memory text representation. The results were interpreted as support for a fast, direct access, "resonance" process rather than a slow, deliberate search.

This dissertation examines auditory language processing with two emphases: the steps in the processing of an auditory input which identify those characteristics which enable the listener to match the incoming signal to a lexical item, and the morphological and phonological form of lexical entries. Evidence is presented to support the Morphological Parsing Hypothesis, a proposal that inflectional morphemes are recognized prelexically. Moreover, it is shown that the analysis into stem and inflectional affix is evaluated for phonological well-formedness prior to a lexical search for the hypothesized stem (the Prelexical Phonological Checking Hypothesis). Results from an experiment using German surface homophones (with final obstruent devoicing) strengthen the claim of Lahiri & Marslen-Wilson (1991) that lexical entries are stored in the form of underlying phonological representations. Suggestive evidence that lexical entries may be composed of radically underspecified featural representations leads to the proposal of the Specified Feature Priority Hypothesis, a means by which listeners may attend to those portions of the acoustic/phonetic input which provide cues to the underlyingly specified phonological features of the language.

Research in human factors engineering has recently begun to focus on the role of computers and powerful graphics display technology as a means for enhancing the information processing abilities of the human decision maker. This experiment evaluates different display formats (ranging from an integral polygon display to a separable digital display) for presenting system data in a process control task that requires diagnosing system state. The effect of both system state uncertainty and data reliability on classification performance (response time and accuracy) across the different display formats are explored. System state uncertainty was manipulated by creating instances within each system state that systematically vary from the system state prototype. Data reliability refers to the diagnosticity of each of four system cues. Highly significant performance differences emerged across the different display formats, uncertainty and data reliability conditions. Perhaps even more noteworthy, however, were the findings relating to individual differences in classification strategies used by operators across all display conditions. These findings are important for human factors engineers to consider when making display design recommendations for process control environments where operators must integrate system data to make diagnostic decisions.

The literature contains a distinction between iconic causal relations as because he studied hard, he got a good grade and evidential causal relations as in because he got a good grade, he studied hard. The current research presents a theoretical analysis of the of these two categories and introduces a third category, the deductive causal relations as in because grading a paper is a subjective process, the teacher made some mistakes. It is argued that iconic causal relation is a relation between two actual, specific events and requires gap-filling inferences. Evidential and deductive relations are inferential and they represent a relation between evidence (in evidential) or a premise (in deductive) and a conclusion, hypothesis, or belief that the reader reaches depending on the evidence or premise. A series of 4 experiments were conducted to verify the predictions derived from the characteristics of each type of causal relations. Experiment 1 showed that deductive relation is a distinguished category that is more difficult than the iconic relation but easier than evidential ones. It was also found that adding an epistemic marker (e.g., I think) facilitated the interpretation of deductive and evidential relations but harmed the iconic relations. Experiments 2 and 3 tested the hypothesis that inferential, non-directly observable events such as those expressed in future tense or in psychological state verbs are more consistent with inferential relations (because they have to be inferred) than the events expressed in past tense or action verbs. The results of Experiments 2 and 3 showed that state verbs and future tense reduced the difficulty associated with inferential relations. Experiment 4 investigated the effect of the presupposition-assertion distinction on iconic and evidential relations. It was found that while the distinction is context-dependent in iconic relations, the main clause is preferred to be the presupposed in evidentials. The results were discussed in terms of the conditional nature of deductive relations and its being based on general, enabling conditions rather than upon real causes. Finally, a processing mechanism was suggested on the basis of the current results.

The effects of onset interaural time differences (ITD) on lateralization and detection of binaural broadband click trains with alternating and non-alternating ITDs were examined under earphones. Three ITDs were employed: 0 us, or $\pm$ 500 us (right-leading and left-leading). Within each train, ITDs were the same (all-left, all-right, and centered) or alternated between two of the three values. The interclick interval was 2 milliseconds and the train duration was 250 milliseconds. Lateralization has investigated with a broadband acoustic pointer with variable delay. Listeners' pointer adjustments were essentially dominated by onset ITD. Detection thresholds were obtained with two broadband continuous maskers, left-leading and diotic. MLDs of nearly 12 dB were seen for the right-leading non-alternating probe when the masker was left leading. For the alternating ITD probes, thresholds improved when one or both ITDs in the probe differed from that of the masker. Threshold improvement was independent of the onset ITD. Similar results were obtained with the diotic masker, with the magnitude of the MLD being somewhat less. To rule out the possibility of the onset being masked by the noise masker at threshold, MLDs were obtained for trains with an exponentially decaying envelope with the onset 20 dB above the trailing portion of the trains. Detection thresholds obtained in a duration discrimination task revealed MLDs of smaller magnitude, but even with the onset enhanced, no effect of onset ITD was seen. These results showed that while onsets strongly influenced lateralization of the relatively long duration signals used in these experiments, MLDs were determined by the ongoing interaural differences. Subjects' performance in lateralization showed some inter-subject variability and variability across signals. Three out of four subjects also lateralized more strongly to the left than to the right. When lateralization was determined as a function of signal sensation level (SL), the effect of onset ITD became weaker or disappeared at low SLs. These results are generally consistent with the statistical decision theory based hypothesis of Houtgast and Plomp (1968) and suggested that while onset cues dominated lateral position, detection depended on the time-varying steady-state cues in the signals.