Oskar Gross

Document summarization based on word associations

In the age of big data, automatic methods for creating summaries of documents become increasingly important. In this paper we propose a novel, unsupervised method for (multi-)document summarization. In an unsupervised and language-independent fashion, this approach relies on the strength of word associations in the set of documents to be summarized. The summaries are generated by picking sentences which cover the most specific word associations of the document(s). We measure the performance on the DUC 2007 dataset. Our experiments indicate that the proposed method is the best-performing unsupervised summarization method in the state-of-the-art that makes no use of human-curated knowledge bases.

Lexical Creativity from Word Associations

A fluent ability to associate tasks, concepts, ideas, knowledge and experiences in a relevant way is often considered an important factor of creativity, especially in problem solving. We are interested in providing computational support for discovering such creative associations. In this paper we design minimally supervised methods that can perform well in the remote associates test (RAT), a well-known psychometric measure of creativity. We show that with a large corpus of text and some relatively simple principles, this can be achieved. We then develop methods for a more general word association model that could be used in lexical creativity support systems, and which also could be a small step towards lexical creativity in computers.

Data mining and machine learning in computational creativity

Creative machines are an old idea, but only recently computational creativity has established itself as a research field with its own identity and research agenda. The goal of computational creativity research is to model, simulate, or enhance creativity using computational methods. Data mining and machine learning can be used in a number of ways to help computers learn how to be creative, such as learning to generate new artifacts or to evaluate various qualities of newly generated artifacts. In this review paper, we give an overview of research in computational creativity with a focus on the roles that data mining and machine learning have had and could have in creative systems. WIREs Data Mining Knowl Discov 2015, 5:265–275. doi: 10.1002/widm.1170

On Creative Uses of Word Associations

The ability to associate concepts is an important factor of creativity. We investigate the power of simple word co-occurrence analysis in tasks requiring verbal creativity. We first consider the Remote Associates Test, a psychometric measure of creativity. It turns out to be very easy for computers with access to statistics from a large corpus. Next, we address generation of poetry, an act with much more complex creative aspects.We outline methods that can produce surprisingly good poems based on existing linguistic corpora but otherwise minimal amounts of knowledge about language or poetry. The success of these simple methods suggests that corpus-based approaches can be powerful tools for computational support of creativity.

Language-independent multi-document text summarization with document-specific word associations

The goal of automatic text summarization is to generate an abstract of a document or a set of documents. In this paper we propose a word association based method for generating summaries in a variety of languages. We show that a robust statistical method for finding associations which are specific to the given document(s) is applicable to many languages. We introduce strategies that utilize the discovered associations to effectively select sentences from the document(s) to constitute the summary. Empirical results indicate that the method works reliably in a relatively large set of languages and outperforms methods reported in MultiLing 2013.