Notes

The Importance of a Pragmatic Agent Slot
We live in an era of unprecedented access to information, communication tools and computational power. Despite this, many human interactions still feel frustratingly imprecise and unsatisfactory. Pragmatics seeks to understand the reasons behind these frustrations and provide guidance for avoiding or mitigating them. planetliga has a broad range of topics including turn-taking timing, gaze behaviour, gesture, topical coherence, repair initiation and backchannel usage (to name just a few). In all of these areas it is mainly concerned with accounting for the ways in which pragmatic factors exert pressure on grammar to produce utterances that have desirable functional properties.
The most important insight for achieving this is that functional utterances are built from the bottom up: pragmemes are general situational prototypes (Petersson 1999) that have multiple possible realizations in context. Thus, the key to understanding a speech act lies in identifying its ‘slot’, the role it plays in its corresponding social situational context. The goal of a slot-based approach is to build an agent’s internal representation of the world by assembling slots into a ‘functional model’, i.e. an inference network that predicts the consequences of a given action for both speaker and hearer.
In order to construct such a model, the agent consults all word-object associations in its lexicon when inference is needed. For this reason, the learning signal created by an inference affects all lexicon entries, as illustrated for the fixed lexicon implementation in Fig. 1. In particular, a one-to-one mapping between the current word and its object is strengthened, whereas associations between words and other objects are weakened.
As a result, the mean of all lexicon entries-with which new slots are initially initialized-moves further and further away from true association weights. This means that the likelihood of choosing a novel word to fill a given slot decreases, and the ME index in both the general and specific evaluation drops.
The general-context ME bias declines because one-to-one mappings between familiar objects and a newly presented novel word are not strong enough to exclude them as referents in a literal reasoning process. The specific-context ME bias increase contrasts with the inverse pattern observed for the dynamic lexicon: since only associations between familiar words and objects are strengthened, this reduces the number of available options from which to select the novel word.
The emergence of these effects suggests that the ME bias in pragmatic agents with dynamic lexicons can be explained by the fact that they collect evidence for one-to-one mappings through their initialization mechanism during learning. This is in contrast to the findings of previous studies, which suggest that pragmatic inference induces a lexical ME bias only when it leads to a change in a lexicon entry’s meaning. Nevertheless, it is also clear that pragmatic online inference does induce an ME bias throughout training without needing additional lexical pressures. The article is part of the Elements in Pragmatics series that showcases dynamic and high-quality original, concise and accessible scholarly works in the field of pragmatics. The series is intended to promote dialogue across different perspectives and to encourage cross-fertilization of ideas.

Homepage: https://sneakerprojects.com/