In particular, in TSOMs acquisition strategies are analysed by focussing on emergent relations between stored word forms and on dynamic expectation/competition of incoming input.
In the present paper, we intend to show that computational models of serial memories can provide a principled account of at least some of these effects, by grounding them on some basic mechanisms of co-activation and competition between concurrently stored words.
A non-word with high-frequency chunks is likely to overlap with many stored words at different positions in the string (corresponding to different n-grams.), thus being more likely to be repeated by the map as quickly and correctly as possible.
Likewise, when the symbol 'A' is shown immediately after 'M', the strength of expectation for the corresponding 'a' node to be activated will be a direct function of the cumulative frequency of all stored words beginning with 'MA'.
The search data are then broadcast into the memory banks through n pairs of complementary search-lines SL and directly compared with every bit of the stored words
using comparison circuits.
The gadget then selects the corresponding expression from some 200 stored words
and either displays them on a liquid crystal display or ''speaks'' them to concerned pet owners.
This is formalized by a theory of the lexicon in which complex words inherit properties from their dominating schemas (note that this presupposes accessibility of internal morphological structure of stored words) (Booij, in press).
Therefore, the notion 'lexicon', traditionally conceived as the list of stored words, does not suffice, and must be seen as a constructicon, in which both morphological and syntactic units are stored, together with the schemas that they instantiate.