Pragmatics models inspired by the Iterated Best Response Model. From this google code SVN repository.
This is an R toolkit for studying signaling systems for grounded language understanding. The code provides:
- A variety of speaker and listener agents (and flexible tools for defining new ones).
- Iterators for specifying the Iterated Best Response model and its variants.
- Tools for exhaustively exploring large semantic spaces to find empirical differences between systems.
- Tools for generating experimental stimuli for reference games.
To run classical IBR (as in Jäger 2012) on a standard scalar-implicatures case:
source('ibr.R')
scalars = matrix(
c(1, 0, 0,
0, 1, 0,
0, 1, 1), byrow=T, nrow=3,
dimname=list(
c('w_no', 'w_somenotall', 'w_all'), # Row names; worlds.
c('NO', 'SOME', 'ALL'))) # Column names; messages.
IBR(scalars)
For this example, this is equivalent to
Sstar(Lstar(S0(scalars), scalars))
though IBR allows iteration to arbitrary depth (until convergence, which is guaranteed).
## Scenario from figure 1A:
fg = matrix(
c(1,0,1,0,
1,0,0,1,
0,1,1,0), byrow=T, nrow=3,
dimnames=list(
c('r_bs','r_bc','r_gs'), # Row names; objects.
c('blue','green','square','circle')) # Column names; messages.
)
## Approximate from figure 1C:
bets = c(20,40,38)
prior = bets/sum(bets)
FG(fg, prior=prior)
This is equivalent to
Lbayes(S(L(S0(fg), sem=fg)), sem=fg, prior=prior)
Michael C. Frank, Noah D. Goodman, and Christopher Potts.
Frank, Michael C. and Goodman, Noah D.. 2012. Predicting pragmatic reasoning in language games. Science 336(6084): 998.
Franke, Michael. 2009. Signal to Act: Game Theory in Pragmatics. Institute for Logic, Language and Computation, University of Amsterdam.
Jäger, Gerhard. 2012. Game theory in semantics and pragmatics. In Claudia Maienborn, Klaus von Heusinger, and Paul Portner, eds., Semantics: An International Handbook of Natural Language Meaning. Berlin: Mouton de Gruyter.
Lewis, David. 1969. Convention. Cambridge, MA: Harvard University Press. Reprinted 2002 by Blackwell.