National and Kapodistrian University of Athens
Αυτόματη διάκριση μεταφορικής και κυριολεκτικής σημασίας σε σώματα κειμένων της ελληνικής με υπολογιστικές μεθόδους μηχανικής μάθησης
Abstract
dc:descriptionThe object of this thesis is the automatic differentiation between literal and metaphorical meaning in authentic non-annotated phrases from the Corpus of Greek Texts by means of computational methods of machine learning. For this purpose the theoretical background of distributional semantics is discussed and employed. Distributional Semantics Theory develops concepts and methods for the quantification and classification of semantic similarities displayed by linguistic elements in large amounts of linguistic data according to their distributional properties. In accordance with this model, the approach followed in the thesis takes into account the linguistic context for the computation of the distributional representation of phrases in geometrical space, as well as for their comparison with the distributional representations of other phrases, whose function in speech is already “known” with the objective to reach conclusions about their literal or metaphorical function in the specific linguistic context. This procedure aims at dealing with the lack of linguistic resources for the Greek language, as the almost impossible up to now semantic comparison between “phrases”, takes the form of an arithmetical comparison of their distributional representations in geometrical space.The vast majority of theoretical approaches to metaphor conceive it as a means of mapping and connecting different conceptual areas; there have also been several attempts to identify the criteria according to which these mappings are made. The approach to metaphor as a concept and its assumed correlation with cognition, albeit useful, cannot contribute to the automatic recognition of metaphorical meanings and their differentiation from literal meanings. In parallel, various theories on the automatic detection and recognition of metaphorical phrases have been formulated. However, their majority requires access to linguistic resources and use of tools or expensive and time-consuming manual annotation in order to reach semantic mappings. Taking into consideration the limitations of Greek as regards resources and tools, it is preferable to follow the principles of distributional semantics, which pursue the affiliations of a word with the rest of its linguistic context without explicitly taking into account the connections from between this word and its related concepts. Distributional semantics have been paramount in shifting research interest towards neural language models, which can attribute hidden statistical characteristics of the distributional representations of word sequences in natural language. In this thesis neural language models are preferred to traditional statistical models, as they are in position to render a large amount of training data with a relatively small number of parameters. Thus, the goal is to match every word with a constantly evaluated distributional representation in geometrical space, since the probability of a lexical sequence is expressed as the result of the appearance probabilities, under certain conditions, of the next word offering the probabilities of previous ones. In order to achieve the matching of linguistic data with their distributional representations in geometrical space, the deep learning algorithm word2vec is utilized. After the neural model has been appropriately trained, based on distributional data representations, a support vector machine attempts the binary classification of phrases between literal and metaphorical meaning with criteria of arithmetical proximity of the distributional representation of the phrase under investigation to the distributional representations of other phrases whose function is already known.For the distinction between the literal and metaphorical function of a phrase a discriminative model is recommended, which, through appropriate training, is able to identify the optimal separating hyper plane of the vector representation word combination and will have the ability to generalize to new, unseen data. Then an attempt is made for the automatic detection and recognition of the metaphorical meaning in an automatic and dynamic way through the arithmetical comparison of the distributional representations of phrases with those of others, already characterized in terms of their function.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2017
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Florou, Eirini
- Φλώρου, Ειρήνη
Subjects
dc:subject × 18- Μεταφορά
- Νευρωνικό γλωσσικό μοντέλο
- Κατανεμητική σημασιολογία
- Υπολογιστική σημασιολογία
- Διανυσματικές αναπαραστάσεις λέξεων
- Περικείμενο
- Αλγόριθμος βαθιάς εκμάθησης word2vec
- Metaphor
- Neural language model
- Distributional semantics
- Computational semantics
- Word embeddings
- Context
- Deep learning algorithm word2vec
- Ανθρωπιστικές Επιστήμες και Τέχνες
- Γλώσσα και Λογοτεχνία
- Humanities and the Arts
- Languages and Literature
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/41714