Page 460 - Deep Learning

Notes to Pages 297–303 443

(1995, 1997), Dunbar and Fugelsang (2005), Fugelsang, Stein, Green and Dunbar
(2004) and Trickett, Schunn and Trafton (2005) report empirical studies that add
detail to the story of how scientists react to unexpected findings. They have found
that when evidence contradicts a core hypothesis, scientists do attend to it, largely
because they are pushed to do so by their colleagues in the context of laboratory
meetings and other public and semi-public interactions. Another finding is that the
character of the conflict between theory and data, and the frequency with which
the conflict appears both determine the extent and the type of response.
12. Kuhn (1970, p. 152).
13. Major handbooks of psycholinguistics include Gaskell (2007), Graesser,
Gernsbacher and Goldman (2003) and Traxler and Gernsbacher (2006). In con-
junction, these handbooks cover most aspects of discourse.
14. In a famous article, H. P. Grice (1975) proposed that if conversations are viewed
as a form of collaboration, it follows that people strive to moderate the quantity,
quality and relevance of what they say, as well as their manner of speaking, so as
to be maximally helpful to the listener. This includes saying neither more nor less
than what the listener needs to be told, i.e., to state only what is new to him or her.
See http://plato.stanford.edu/entries/grice.
15. Gaskell (2007) and Traxler and Gernsbacher (2006).
16. See, e.g., Swinney (1979) for an early paper that supported parallel activation of
multiple meanings. As researchers dig deeper, the details become more compli-
cated (Swaab, Brown & Hagoort, 2003). For present purposes, the general idea
of parallel activation of more than one meaning, followed by resolution through
context (perhaps among other factors), is sufficient.
17. The spy and the bird are visitors from Goldstein (2008, pp. 370–376).
18. The turtles on the log were a gift to psycholinguistics from Bransford, Barclay
and Franks (1972). Like other turtles, they have turned out to be long-lived (Jahn,
2004). They demonstrate the ubiquitous presence of bridging inferences in com-
prehending belief-congruent texts. For an example of the diametrically opposite
case, with bridging inferences lacking in the processing of a belief-incongruent
text, see Graesser, Kassler, Kreuz and McLain-Allen (1998).
19. Nussbaum (1985), Vosniadou (1994a, 1994b) and Vosniadou and Brewer (1992).
20. Johnson, Moher, Ohlsson and Gillingham (1999), Johnson, Moher, Ohlsson and
Leigh (2001) and Ohlsson, Moher and Johnson (2000).
21. Vosniadou (1994a, 1994b) and Vosniadou and Brewer (1992).
22. Duhem (1914/1991, especially pp. 83–88; see p. 187 for the quote).
23. Quine (1951) and Quine and Ullian (1978).
24. Quine and Ullian (1978, p. 16).
25. Quine and Ullian (1978, p. 79).
26. Quine and Ullian (1978, p. 22).
27. Quine and Ullian (1978, p. 79).
28. Quine (1951, p. 39).
29. See Rokeach (1960, 1970). The idea of grouping beliefs or knowledge elements
by topic or theme is consistent with the idea that the mind strives for local
rather than global coherence. Global coherence is not attainable in a natural
system, in part because assessing the coherence of every piece of knowledge