nucleation

THEORY-816: Learners inferring what a novel word means from its use reason about a speaker who chose the word to be informative about the referent, so the size principle in word learning is a consequence of that inference and not a separate rule of generalisation

2 min read

Properties5
StatusProposed
Tagspragmatics, cognition, probabilistic-modeling
SourceLIT-824 — Informative Communication in Word Production and Word Learning
ExtendsTHEORY-472 — A speaker who chooses words to minimise the information lost about a single intended referent, with words' literal meanings uniform over their extensions, picks each true word with probability inversely proportional to a power of its extension, so the size principle follows from informative communication; in one betting experiment, learners who knew the referent bet on a novel word's meaning in line with the inversion of that speaker (r = .93 on condition means), and in one free-description study speakers' word use followed it only weakly (r = .19)
AnswersQUESTION-167 — Do word learners reason about an informative speaker, or only apply the size principle, when displays are built so that the two predict different bets?

Graph View

Source

Frank, Goodman, Lai and Tenenbaum (2009), LIT-824, read in full in NOTE-627. They write the informative speaker down (utility −D_KL from the intended to the literal meaning), derive the size principle from it, and present the learner as inverting that speaker: a learner who sees a speaker point at the one red circle among blue shapes and say a novel word should lean towards “red”, because “red” singles the object out. The size principle, justified elsewhere as a rule about generalisation, “here becomes a consequence of the speaker’s goal” (NOTE-627, claim C2).

What is asserted, and by whom

The thesis is that learners who infer a novel word’s meaning from its use reason about a speaker who chose the word to be informative about the referent, so the size principle in word learning is a consequence of that pragmatic inference and not a separate rule of generalisation. Frank, Goodman, Lai and Tenenbaum propose it.

What the record holds toward it

  • THEORY-472 (Active): the derivation (for a point referent and uniform literal meanings, word choice ∝ |extension|^(−α)), and, in one betting experiment, mean bets in line with the inverted speaker (r = .93 on condition means).
  • In those displays the informative-speaker learner and a plain size-principle learner coincide by construction, so the fit does not separate this thesis from the rule it would explain.
  • THEORY-785 states the same inversion for listeners interpreting utterances in reference games, from the rational speech act papers; this entry is its word-learning counterpart.

What this does not say

  • Not that speakers are informative. In the paper’s free-description study speakers followed informativeness only weakly (r = .19), and colour terms not at all; the thesis is about the learner’s assumption.
  • Not that children learn this way; the participants were adults on the web.

What would confirm or reject it

Confirmed by the promote_when: displays in which the two learners predict different bets, with both models and a frequency baseline fitted on part of the data and tested on the rest, and the informative-speaker learner winning. Rejected if the plain size-principle learner does as well where they differ. QUESTION-167 asks the same thing as a question; this entry is the answer the authors propose.

Backlinks