In the field of Natural Language Processing, supervised machine learning is commonly used to solve classification tasks such as sentiment analysis and text categorization. The classical way of representing the text has been to use the well known Bag-Of-Words representation. However lately low-dimensional dense word vectors have come to dominate the input to state-of-the-art models. While few studies have made a fair comparison of the models' sensibility to the text representation, this thesis tries to fill that gap. We especially seek insight in the impact various unsupervised pre-trained vectors have on the performance. In addition, we take a closer look at the Random Indexing representation and try to optimize it jointly with the classification task. The results show that while low-dimensional pre-trained representations often have computational benefits and have also reported state-of-the-art performance, they do not necessarily outperform the classical representations in all cases.

published_or_final_version / abstract / Linguistics / Master / Master of Arts

published_or_final_version / Philosophy / Master / Master of Philosophy

No description available.

No description available.

This dissertation uses syntactic, semantic and morphological evidence from English nominalization to probe the interaction of event-structure and syntax, develop a typology of structural complexity within nominalization, and test hypotheses about the strict ordering of functional items. I focus on the widely assumed typology of nominalization found in Chomsky (1970). In particular, I show that derived nominals are structurally more complex than nominal gerunds; this has long been assumed to be the opposite. I provide a structural and morphological account of these forms of nominalization. In doing so, I explore a number of disparate topics such as: the importance of syncretism in apparently unrelated morphological elements for theories like Distributed Morphology; the role of prepositions in allowing or preventing binding relations and NPI-licensing, the exact nature of root-object union that allows idiomatic interpretations; the morphological reflexes of Case in the nominal system; the syntactic structure of verb particle constructions; the nature of events in nominalization; and the role syntactic operations play in determining morphological regularity. The dissertation also explores the nature of the English verb particle construction, arguing that it has (at least) three distinct structural configurations. Using these three distinct structures I am able to explain a number of distinct behaviors from predicate-object relationships, particle modification and argument loss in particle construction. I also discuss the relationship between particles (and results) and the different forms of nominalization. In particular, I show that apparent co-occurrence restrictions between nominal types and particles are not due to event-structure or other semantic restrictions. Rather, these differences are tied solely to the particular, idiosyncratic morphological properties of the constructions. The dissertation shows that certain functional projections may only appear once with a given root, but that there is some freedom of ordering of projections relative to the root in some cases. This work provides a window into the interaction between syntax and event structure as well as the nature of ordering within functional projections.

No description available.

No description available.

No description available.

No description available.