Title:
Discriminative Learning and Implicit Morphology
Date:
23/04/2015
Town:
Pisa
Venue:
ILC-CNR – Aula Seminari IBF SG 5
Description:
I will present a computational model built on a Rescorla-Wagner network which successfully learns to activate most strongly those lexical contrasts that are actually encoded in the orthographic signal, without decomposing that signal into a non-overlapping sequence of morphemes.
Next, I will show that a measure of prior availability that can be derived from the Rescorla-Wagner network correlates strongly with standard word frequency counts. This measure provides an explanation for the word frequency effect without invoking counters in the head.
Finally, I will demonstrate that the product of network activation and network prior availability provides a single measure that simplifies regression models predicting response latencies in primed lexical decision, as compared to a range of standard lexical predictors.
The presented learning model makes optimal use of the rich potential of sublexical cues that serve to discriminate the various experiential contrasts encoded in writing systems and exploited during word recognition.
Next, I will show that a measure of prior availability that can be derived from the Rescorla-Wagner network correlates strongly with standard word frequency counts. This measure provides an explanation for the word frequency effect without invoking counters in the head.
Finally, I will demonstrate that the product of network activation and network prior availability provides a single measure that simplifies regression models predicting response latencies in primed lexical decision, as compared to a range of standard lexical predictors.
The presented learning model makes optimal use of the rich potential of sublexical cues that serve to discriminate the various experiential contrasts encoded in writing systems and exploited during word recognition.
Programme:
- Introduction
Vito Pirrelli, ILC-CNR (5′) - Discriminative Learning and Implicit Morphology
Petar Milin, FF-UNS (45′) - Discussion (10′)
Promotional Material: