Message 1: Re: 17.1317, Disc: Starling Study: Recursion
Date: 17-May-2006
From: Geoffrey Pullum <pullumsoe ucsc edu>
Subject: Re: 17.1317, Disc: Starling Study: Recursion
Tim Gentner and his colleages recently published a paper in Nature about
learning of artificially constructed recursive "grammars" by songbirds,
with patterns like AAAABBBB (Gentner TQ, Fenn KM, Margoliash D, Nusbaum
HC, 2006: Recursive syntactic pattern learning by songbirds, Nature
440: 1204-1207). The work has been widely reported in newspaper articles.
We (Ray Jackendoff, Mark Liberman, Geoff Pullum, and Barbara Scholz) wrote
a letter to Nature to warn against over-interpretation of the Gentner
group's results. The letter presented four reasons for thinking the
paper's conclusion, that the experiment "opens a new range of complex
syntactic processing mechanisms to physiological investigation", was
not sufficiently supported.
Within 18 hours, Nature declined to publish the letter. (In our
experience, this is what usually happens when linguists write to
general science journals like Nature and Science commenting on the
content of papers with linguistic content that have been published
by non-linguists.) Readers of the LINGUIST List might like to read
here the four points that were expressed in the letter, which were
expressed thus:
(i) Even if it were true that starlings could grasp a recursive grammar,
this could hardly provide direct evidence about the evolution of human
language. It would be at best an analogous rather than homologous
capacity -- surely not an inheritance from some common ancestor of
birds and mammals.
(ii) It is not clear that the starlings learned a recursive rule.
Becoming habituated to a pattern like AAABBB does not necessarily imply
grasping recursively center-embedded structures like A(A(A(...)B)B)B.
This pattern could equally be detected by comparing the number of A's
and B's, given that some birds such as pigeons can subitize numbers up
to 4 or more (Dehaene S, The Number Sense, Oxford University Press, 1997).
(iii) Humans do not perform well on center-embedded syntax. Even for
n = 2, they are often baffled by A^nB^n structures ("People people
cheat cheat" is all but unintelligible). Few can handle n = 3. Are
starlings outperforming humans syntactically?
(iv) Recursion is not the unique core property of the human language
faculty anyway (Pinker S & Jackendoff R, `The faculty of language:
What's special about it?', 2005, Cognition 95, 201-36). Recursion is
arguably involved in comprehension of complex visual fields, planning of
action, and understanding social environments. These human capacities are
shared with other primates. Unique to human language is a very large
learned vocabulary consisting of long-term memory associations between
meanings and structured pronunciations plus varied phrasal syntax.
The Gentner experiment may help us understand animal pattern recognition
and learning abilities, some of them possibly prerequisites for linguistic
abilities; but the implications are being considerably exaggerated,
especially in popular media accounts with headlines like "Songbirds
May Be Able to Learn Grammar."
Ray S. Jackendoff
(Center for Cognitive Studies, Tufts University)
Mark Y. Liberman
(Institute for Research in Cognitive Science, University of Pennsylvania)
Geoffrey K. Pullum, Barbara C. Scholz
(Radcliffe Institute for Advanced Study, Harvard University)
Автор
Тема: изчислителна лингвистика (Прочетена 23176 пъти)