Recognizing Spoken Words:The Neighborhood Activation ModelPaul A. Luce and David B. PisoniPresented by: Yinan FanFeb. 2nd, 2007Neighborhood Activation ModelSpoken Word Recognition• Structural organization in mental lexicon• Sensory and perceptual informationWhy?• Spoken word is not perfectly content-addressable– Various noises– Signal itself• Listener’s recongnition system also noisyÎAlternative: stimulation on a number of candidatesDefining Similarity• Computational:– Metrics based on specific computational terms• Behavioral (operational):– metric based on the result of series of perceptual experimentsNeighborhood Probability Rules (NPR)• Confusability of individual speech sound determined from all components• Incorporating the probability of identifying the stimulus word and the probabilities of confusing its neighborsStimulus Word Probabilities (SWPs)• Based on the probabilites of the individual phonemes of the stimulus word•Neighbor Word Probabilities (NWPs)• Conditional probabilities for each neighbor of the stimulusFrequency-Weighted Neighborhood Probability Rule (FWNPR)Word Decision UnitExperiment 1•CVC words• Manipulated SN ratios• Subjects instructed to guess the wordsPerformanceExperiment 2•CVC words• Subjects instructed decide as quickly as possible whether a given stimulus item is a word or nonwordPerformance -1Perfomance-2Experiment 3•CVC words• Subjects presented with a spoken word and required to repeat or pronouce the word as quickly as possiblePerformanceConclusion and Discussion• Spoken word recognition systems is capable of considering numerous alternatives in parallel• Cost: accuracy and reaction
View Full Document