This work aims at defining an explanatory model of the morphological lexicon as a dynamic system of word learning and processing in both mono- and bilingual contexts. The main focus is on exploring some relevant aspects of the paradigmatic organisation of the mental lexicon in language learning, based on a dynamic analysis of mono- and bilingual contexts. The proposed interdisciplinary approach to lexical acquisition combines theoretically-motivated accounts, psycho-cognitive evidence and methodologies, and machine learning technologies. In particular, I will take into account those basic psychological and cognitive mechanisms that are considered as crucial in language acquisition: (i) the ability to perceive recurrent morphological structures (invariances) in varying temporal contexts, (ii) the capability to access/activate time series of symbols in the short term memory and to selectively integrate them with long term memory expectations, (iii) the attitude towards building novel forms through analogical extension of intra- and inter-paradigmatic relations (generalisation). This investigation is pursued through a computational model based on a recurrent Self-Organising Map, with Hebbian connections defined over a temporal layer (Temporal Self-Organising Map, TSOM), providing a principled algorithmic account of effects of lexical acquisition, processing and access. The computational simulation of a biologically inspired neural architecture of the mental lexicon offers the possibility to reproduce a wide range of conditions of mono- and bi-lingual input exposure, and to illustrate the dynamic of word acquisition and the emergence of morphological organisation. The proposed model provides an adaptive multifactorial account of morphology acquisition affected by a variety of input factors, such as word frequency distributions, paradigm regularity and wordlikeness, whereby lexical perception and organisation are grounded in memory-based processing strategies. In addition, it suggests a processing-based notion of morpheme, as a by-product of processing dynamics, with paradigms emerging as specialised surface relations between inflected forms.
Il presente lavoro si propone di definire un modello esplicativo del lessico morfologico come sistema dinamico di acquisizione ed elaborazione in contesti mono- e bi-lingui. L'attenzione è rivolta all'esplorazione di alcuni degli aspetti principali dell'organizzazione paradigmatica nell'acquisizione evolutiva del lessico mentale, attraverso un'analisi dinamica delle tappe nelle quali essa si articola in contesti mono- e bilingui, partendo dai contributi più recenti della riflessione teorica contemporanea sul lessico mentale, dell'evidenza psico-linguistica sperimentale, e della modellistica computazionale su base neurale. In particolare, si approfondiscono alcuni dei meccanismi psico-cognitivi di base ritenuti determinanti in questo processo: (i) la capacità di percepire strutture invarianti che ricorrono in diversi contesti temporali, (ii) la capacità di attivare sequenze simboliche nella memoria a breve termine e di integrarle selettivamente con le aspettative della memoria a lungo termine, (iii) la capacità di produrre forme nuove attraverso processi di estensione analogica intra- ed inter-paradigmatica (generalizzazione). La ricerca si avvale del contributo di mappe ricorrenti auto-organizzanti dotate di un livello di connettività hebbiana (TSOM, mappe temporali auto-organizzanti), che offrono una modellizzazione algoritmica di effetti di acquisizione, elaborazione e accesso lessicali. La simulazione computazionale di architetture neuro-biologicamente ispirate del lessico mentale consente di riprodurre artificialmente un ampio ventaglio di condizioni di esposizione a input mono- e bi-lingue, e di analizzare la dinamica dell'acquisizione lessicale e l'insorgenza di una organizzazione morfologica basata sui paradigmi. Il modello si propone di illustrare una dinamica multifattoriale del processo di acquisizione morfologica, risultante dall'interazione tra fattori di frequenza, regolarità paradigmatica e familiarità lessicale, e meccanismi di percezione e organizzazione lessicali basati su strategie di elaborazione e memorizzazione. Nella prospettiva suggerita, la nozione di morfema acquisisce una identità funzionale in relazione alle dinamiche di percezione delle relazioni superficiali tra forme pienamente flesse, e la nozione di paradigma emerge come risultato dell'interazione di principi di auto-organizzazione emergente.
Modelling the morphological lexicon - A computational approach to mono-and bilingual learning and processing of verb inflection
Marzi C
2021
Abstract
This work aims at defining an explanatory model of the morphological lexicon as a dynamic system of word learning and processing in both mono- and bilingual contexts. The main focus is on exploring some relevant aspects of the paradigmatic organisation of the mental lexicon in language learning, based on a dynamic analysis of mono- and bilingual contexts. The proposed interdisciplinary approach to lexical acquisition combines theoretically-motivated accounts, psycho-cognitive evidence and methodologies, and machine learning technologies. In particular, I will take into account those basic psychological and cognitive mechanisms that are considered as crucial in language acquisition: (i) the ability to perceive recurrent morphological structures (invariances) in varying temporal contexts, (ii) the capability to access/activate time series of symbols in the short term memory and to selectively integrate them with long term memory expectations, (iii) the attitude towards building novel forms through analogical extension of intra- and inter-paradigmatic relations (generalisation). This investigation is pursued through a computational model based on a recurrent Self-Organising Map, with Hebbian connections defined over a temporal layer (Temporal Self-Organising Map, TSOM), providing a principled algorithmic account of effects of lexical acquisition, processing and access. The computational simulation of a biologically inspired neural architecture of the mental lexicon offers the possibility to reproduce a wide range of conditions of mono- and bi-lingual input exposure, and to illustrate the dynamic of word acquisition and the emergence of morphological organisation. The proposed model provides an adaptive multifactorial account of morphology acquisition affected by a variety of input factors, such as word frequency distributions, paradigm regularity and wordlikeness, whereby lexical perception and organisation are grounded in memory-based processing strategies. In addition, it suggests a processing-based notion of morpheme, as a by-product of processing dynamics, with paradigms emerging as specialised surface relations between inflected forms.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.