Words can be represented as temporal patterns (strings of acoustic/written symbols), and are produced by and input to a TSOM one symbol at a time on the input layer.
a sound occurs in word final position, or it is preceded by a vowel), a TSOM is likely to develop many different nodes for the same stimulus, with each node being specialised to respond to a particular contextual realisation of the stimulus.
1995) the top 50 high-frequency paradigms and 15 inflected forms for each paradigm (the full set of present indicative (6) and prateritum (6) forms, the past participle, the infinitive and the present participle); (ii) for each of the 50 paradigms one verb form was not administered to a TSOM during training, and used to form a set of paradigmatically-related test words; (iii) an additional German test set was selected from CELEX, by picking up 25 more high-frequency paradigms forming a set of novel words.
For all the four datasets we first evaluated the TSOM behaviour on two tasks: WORD recoding and WORD RECALL.
Accuracy decreases to 51% for German novel words (novel verb paradigms) and drops to 18.6% for Italian words (non-words from the TSOM perspective).
With these goals in mind, we tested a TSOM trained on German verbs by observing its behaviour in recalling words from three different test sets not included in training, and defining a natural gradient of word familiarity to a German TSOM: unknown forms of known verb paradigms (set ii), unknown German paradigms (set iii) and Italian paradigms (set iv).
As expected, they are recalled consistently more poorly (Table 1), but their degree of perceived familiarity can be accounted for by their fitting a recurrent morphological pattern exhibited by other German words the TSOM is already familiar with.
in this perspective, Temporal Self-Organising Maps (TSOMs) provide a computational framework to test fundamental mechanism underpinning serial cognition and for simulating processes of lexical organisation.
In TSOMs, alternative chains of memory nodes may be co-activated by the same input sequence, but only one chain is preferred over the others during processing, depending on the degree of probabilistic support received by the network of long-term associative relations among stored word forms.
Here we want to empirically assess the role of wordlikeness in the overall organisation of lexical items by TSOMs, with particular emphasis on issues of processing, storage, access, and recall.
In this section, we describe wordlikeness effects by relating them to specific aspects of TSOMs' behaviour.