"Quasiregularity and its discontents"
« previous post | next post »
Suggestion for your weekend reading: Mark Seidenberg and David Plaut, "Quasiregularity and Its Discontents: The Legacy of the Past Tense Debate", Cognitive Science 2014. The abstract:
Rumelhart and McClelland’s chapter about learning the past tense created a degree of controversy extraordinary even in the adversarial culture of modern science. It also stimulated a vast amount of research that advanced the understanding of the past tense, inflectional morphology in English and other languages, the nature of linguistic representations, relations between language and other phenomena such as reading and object recognition, the properties of artificial neural networks, and other topics. We examine the impact of the Rumelhart and McClelland model with the benefit of 25 years of hindsight. It is not clear who “won” the debate. It is clear, however, that the core ideas that the model instantiated have been assimilated into many areas in the study of language, changing the focus of research from abstract characterizations of linguistic competence to an emphasis on the role of the statistical structure of language in acquisition and processing.
Some previous LLOG coverage of related issues:
"The theology of phonology", 1/2/2004
"The curious case of quasiregularity", 1/15/2004
"Who let the 'n' in?", 1/22/2006
"The evolutionary psychology of irregular morphology", 4/10/2008
Ran Ari-Gur said,
September 14, 2014 @ 1:21 pm
According to Seidenberg and Plaut (if I'm understanding them correctly), Pinker (and other scientists) treated Rumelhart and McClelland's model as though it were the connectionist model, and presented any limitations of the model as though they were fatal flaws in connectionism.
From the standpoint of 2014, that seems like a very silly mistake; was it a more plausible one at the time?
Brett Reynolds said,
September 16, 2014 @ 3:04 am
@Ran: No, I remember reading RM, and thinking, "and just imagine what you could do if you ramped this up." That in itself was a bit naive because simply adding nodes to a network doesn't always improve its performance. But the point is, it was easy to imagine a different setup getting better results.