This paper deals with Domain Adaptation for automatic syntactic annotation. Until the half of the 1980s, automatic linguistic annotation was based on algorithms built on groups of hand-written rules, defined a priori on the basis of the knowledge of the system to formalise. Subsequently, thanks to the progress of research in the field of Artificial Intelligence and to the development of linguistic resources, algorithms based on machine learning techniques began to be employed. The major difficulties of those algorithms were due to certain aspects of natural language such as ambiguities, diachronic evolutions, or language variations from the original domain of knowledge. More specifically, the issue of Domain Adaptation can be put in the following terms: "can an annotated corpus [which is representative of a specific linguistic variety] be used for the syntactic analysis of a second corpus [which is representative of a different linguistic variety]?". The author answer presenting an algorithm called ULISSE (Unsupervised LInguistically-driven Selection of dEpendency parses), which selects in an optima way the most representative sentences of a new target domain and feed them to the parser in addition to the original training set.
ULISSE: una strategia di adattamento al dominio per l'annotazione sintattica automatica
Dell'Orletta F;Venturi G
2016
Abstract
This paper deals with Domain Adaptation for automatic syntactic annotation. Until the half of the 1980s, automatic linguistic annotation was based on algorithms built on groups of hand-written rules, defined a priori on the basis of the knowledge of the system to formalise. Subsequently, thanks to the progress of research in the field of Artificial Intelligence and to the development of linguistic resources, algorithms based on machine learning techniques began to be employed. The major difficulties of those algorithms were due to certain aspects of natural language such as ambiguities, diachronic evolutions, or language variations from the original domain of knowledge. More specifically, the issue of Domain Adaptation can be put in the following terms: "can an annotated corpus [which is representative of a specific linguistic variety] be used for the syntactic analysis of a second corpus [which is representative of a different linguistic variety]?". The author answer presenting an algorithm called ULISSE (Unsupervised LInguistically-driven Selection of dEpendency parses), which selects in an optima way the most representative sentences of a new target domain and feed them to the parser in addition to the original training set.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.