Creativity is what we say it is: constructing an ontology of creativity

Tracking #: 430-1568

anna jordanous
Bill Keller

Responsible editor: 
Krzysztof Janowicz

Submission type: 
Full Paper
This paper describes a novel application of cognitive science methods and approaches to the analysis of creativity and to the development of a machine-readable ontology of creativity. As a complex, ambiguous and multi-faceted concept, creativity has proved resistant to satisfactory and comprehensive definition, despite numerous attempts to provide one. Using techniques from the field of statistical natural language processing and cognitive modelling techniques, this paper describes the construction of a representation of creativity on the Semantic Web, built through an analysis of what is considered important when talking about creativity. Words were identified which appeared significantly often in connection with discussions of the concept. Using a measure of lexical similarity to cluster these words, a number of distinct themes emerged, which collectively contribute to a comprehensive and multi-perspective representation of creativity. The components provide an ontology of creativity in the form of a set of building blocks that collectively make this subjective concept more tractable, increasing semantic clarity and depth of information available on the concept while accommodating different instantiations of creative activity. The ontology has application to research into the nature of creativity in general and to the evaluation of creative practice, in particular. The provision of a machine readable conceptualisation of creativity also provides a small but significant step towards addressing the problem of automated evaluation, 'the Achilles' heel of AI research on creativity'. From a broader perspective, building the creativity ontology illustrates how adoption of cognitive science methods enables the modelling and representation of highly subjective, semantically contestable concepts on the Semantic Web: a significant step forward for Semantic Web research.
Full PDF Version: 


Solicited Reviews:
Click to Expand/Collapse
Review #1
Anonymous submitted on 17/Jun/2013
Review Comment:

This paper proposes a conceptualization for the term creativity in form of ontological knowledge. The authors use natural language processing and, in particular, corpus-based methods in order to find a multi-dimensional and interdisciplinary analysis of creativity in terms of key components. Section 1 (Introduction) sketches, first, the general goals of the paper and second, attempts to show that creativity is a complicated and not well-defined concept in psychology, cognitive science, and computer science. The authors discuss lengthy various proposals for definitions in Subsection 1.1 in order to conclude that none of these attempts to define creativity suffices. They continue in Subsection 1.2 to discuss the very possibility of the semantics of subjective concepts by reference to philosophers such as Wittgenstein and Waismann. Subsection 1.3 puts the subjective aspects into a semantic web perspective. Then, they conclude in Subsection 1.4 that a possibility to gain a clearer understanding of creativity could be the construction of an ontology for the term in question through an empirical study of natural language. Section 2 (Identifying the key components of creativity) describes the strategy to build the ontology for creativity and mentions the methods used in the construction. First, the authors explain the corpus (Subsection 2.1) and then they mention some natural language processing methods, such as lemmatization or POS tagging (Subsection 2.2). In Subsection 2.3, the authors explain their method to find words in the corpus that are associated with creativity. In particular, they use the log-likelihood ratio to compute candidates. Subsection 2.4 introduces the method to compute the distributional similarity of two words using grammatical relations as major factor in order to form clusters as key components. Section 3 (Results: an ontology of creativity) discusses the computed key components in Subsection 3.1 and continues with describing the implementation of the ontology. Section 4 (Critical evaluation and discussion) roughly sketches some experiments related to the evaluation of the components and concludes the paper.

In general, I think this is an interesting paper. The problem of how to define creativity, and in particular, of how to define computational creativity, is in fact an open problem with many disputes and competing views. The authors propose to evaluate the usage of words in articles that deal with creativity and the co-occurrence of words related to creativity in order to abstract an ontology of creativity. I think this is an interesting approach in general. Clearly, a crucial issue is the choice of the underlying corpus, in particular, the balance of texts from different disciplines (in the current corpus there is a clear dominance of psychological papers).

I do not understand to which extent the paper is related to cognitive science. As far as I can see this, the whole business of computing the ontology uses more or less standard techniques of corpus linguistics, semantic web technologies, and statistical data analysis. The few, not very precise, and perhaps preliminary remarks in Subsection 4.1 concerning experimental studies with subjects are the only hint where proper cognitive (or perhaps psychological) aspects are mentioned. I am not sure whether this suffices to count as a proper candidate in a special issue of Cognitive Science and the Semantic Web.
It should be noted that the paper does not contain any new methods or techniques. All methods are well-known and taken from the literature. Therefore, the theoretical significance concerning technical content is rather low. The practical aspect concerns the construction of an ontology for creativity, in particular the specification of key components of creativity. To which extent these key components result in a better understanding of creativity is less clear to me. In order to understand creativity better, I think more structure and relations between the clusters need to be added, simply because they describe creative acts on rather different levels: "Originality" is a property of the result of a creative process, whereas "emotional involvement" describes a property of the creator. Or to mention another example: "Domain competence" and "general intellectual ability" are prerequisites or necessary abilities of the creator, whereas "value" is a judgment of others about the usefulness of the result of a creative act. In other words, I fear that the connections, dependencies, and interactions between the clusters would be of interest and necessary to understand creativity better. Unfortunately, I cannot see such aspects in the paper.

An issue that is related to the need for more structure: I do not fully understand why the collection of clusters pointing to the concept "creativity" is called "ontology". Ignoring for a moment that the term ontology is perhaps similarly unclear as the term creativity, I do think that ontological knowledge presupposes a hierarchical structure of concepts and specified relations between individuals of certain concepts. The resulting "ontology" in Fig. 4 has a very limited degree of hierarchical structure and not many illuminating relations between individuals. Therefore, I have problems to see the ontology in the graph simply because of the lack of more structure. Connecting this point with the remarks above, I would say that more structure would probably yield a better understanding.

Although, the first section gives a nice introduction to the overall problem, namely how to define and how to understand creativity, I am not sure whether a shorter introduction would also suffice to provide the necessary background to the reader. The saved space could be used to better explain the rather informal evaluation of the components in Subsection 4.1.

Concerning the formal aspects of the paper there is not much to criticize. The paper is very well structured and well written. There are only some minor errors and mistakes. Here are some issues that can be easily resolved:

Page 2: "Currently, no one definition..." -> Probably: "Currently, not one definition..."

Page 6: "...[15] defines various terms to do with bibliographic records..." -> Probably: "...[15] defines various terms having to do with bibliographic records..."

Page 8: "...factors that are persistent persist across different..." -> Probably: "...factors that are persistent across different..."

Page 11, caption Fig. 2: "Each word i a node in the graph..." -> Probably: "Each word is a node in the graph..."

Page 12: Fig. 3 could be increased in size. In the current version it is hard to read on a printout.

Page 15: "Feedback so far has shown..." -> Something seems to be wrong with this sentence.

Page 18: Formatting problems in reference [31].

All in all, I think that the paper could be potentially of interest. Nevertheless, I see some shortcomings and problems of the paper. The relation to cognitive science is not clear to me, there are no new technical insights, the resulting ontology is only to a minimal degree structured, and whether this ontology yields a better understanding of creativity is unclear. Insofar, I am not completely sure what I have learned after reading the paper, although the topic seems to be of interest to me.

Review #2
By Werner Kuhn submitted on 30/Jun/2013
Review Comment:

This is a very well written paper on a fascinating question, applying an interesting method. But it is not a paper for the Semantic Web Journal.

The authors have built a (very small) "corpus" of scientific papers describing creativity and one of papers NOT describing creativity. From these, they extract terms related and not related to creativity. This allows them to identify and then cluster terms that occur with more than the expected frequencies in research on creativity.

The question of what creativity is, what its components are, how it can be assessed and evaluated is obviously interesting and important. There are some areas where answers matter a lot, such as intellectual property or scientometrics. However, the authors do not state a real problem to solve. They argue at length that there is no clear and agreed upon definition, let alone a formalization of creativity. Yet, they do not make a convincing case that such a definition is needed and how it would be used in specific reasoning tasks. The many quoted claims of others are mainly hand waving and fail to convince me. There are many core topics of science that do not have a standard definition, consider "life" or "interoperability" (not to mention concepts in the humanities and social sciences). While one can argue that this is detrimental for certain goals and tasks, one does need to make this case specifically. The authors do not give us any use case for their work, much less one for a semantic web ontology (as opposed to a broader conceptual analysis, which is what they have done, though limited to textual and scientific sources).

The authors seem to imply that ontology (or ontologies in the semantic web) are capturing generally accepted definitions. This is a common misconception, even held by some prominent figures in the field. Semantic web ontologies can only constrain interpretations of terms, so that these terms get used and interpreted by machines in ways that *some* identified information communities use them. These ways are often far from "standard" or "general" and reflect particular terminological conventions.

One could ask what communities use the concept of creativity (or various aspects of it, as identified in the paper) for what tasks, and how one could support them through the semantic web. The paper does not raise this question in any specific way and remains at a rather general level of "we believe ontologies are a good tool to clarify terms".

The title of the paper is the first hint that no clear problem has been addressed. The distinction, early in the paper, of "objective" vs "subjective" concepts is not clear and probably not very useful. Almost all concepts are neither strictly objective (what ever that would mean) nor subjective (to whom?).

I expected some interesting insights from the 14 component concepts that were identified by clustering creativity-related terms. However, the clustering (and the vagueness of creativity-related discourses) has produced component concepts that appear to be even less clearly definable than creativity. Also, the claim that they are "building blocks" appears not justified to me.

The paper offers an insight that was less clear to me before: creativity, like beauty, is probably mainly in the eye of the beholder (in this case, the eyes of both, the creative agent and the one who finds something creative, possibly in different ways). Fortunately or unfortunately, this kind of concept is just not one we are ready to encode in OWL and let machines reason about it.

How could the research be presented differently? Having made all these critical remarks, I do like the fact that an empirical, text-based method was followed and the results are available for inspection and improvements. The method (if properly documented and made available as a workflow for ontologies from corpora, with associated tools) could be a valuable contribution, even to this journal. It will have nothing to do with creativity as such, which might just serve as an application, though more useful and convincing ones would need to be presented. If such a path should be taken by the authors, I suggest they consider whether a limitation of text sources to the very small samples they took (with rather non-transparent criteria) is really necessary and admissible.


Please consider this paper under the call for the Special Issue on Cognitive Science and the Semantic Web. Many thanks,
Anna Jordanous and Bill Keller