Ontology Verbalization using Semantic-Refinement

Tracking #: 2048-3261

Vinu Ellampallil Venugopal
P Sreenivasa Kumar

Responsible editor: 
Philipp Cimiano

Submission type: 
Full Paper
This paper presents an inference-based verbalization approach for OWL axioms that aims at removing redundancy to provide more helpful and more concise sentences to show formal facts to users. We focus on verbalization of OWL axioms about individual and (atomic) concepts as it is relevant in the context of validating the correctness of the formalized domain knowledge by domain experts. The approaches that are currently adopted in the existing verbalization tools generally consider all axioms that are associated with the individual (or the concept) under consideration and then translate them to corresponding natural language texts. Further refinement (mainly, grouping and aggregation) of these texts would be done at the natural language level to yield a more fluent and comprehensive form. However, we observed that human-understandability of such descriptions is affected by the presence of repetitions and redundancies which can be removed easily at the semantic level. We propose a novel technique called semantic-refinement which fulfills this requirement. This technique utilizes a predefined set of rules that are repeatedly applied over the restrictions that are associated with an individual (or a concept) in a meaning-preserving manner to get a refined set of restriction that can be verbalized to get a concise description. Our experiments on two ontologies show that semantic-refinement technique could significantly improve the readability of the natural language descriptions on comparing to those descriptions that are generated without employing the semantic-refinement. We have also tested the effectiveness and usefulness of the the generated descriptions in validating the correctness of ontologies and found that the proposed technique is indeed helpful in that context.
Full PDF Version: 


Solicited Reviews:
Click to Expand/Collapse
Review #1
By Basil Ell submitted on 21/Jan/2019
Review Comment:

In my previous review I stressed the point that redundancy removal is the main contribution of the paper, that works by other authors exist that address redundancy removal in ontologies, and that these works were not cited or discussed. I consider it particularly important that the authors make clear the similarities and differences with existing works on redundancy elimination in ontologies. Although in the current submission under review, the reference to a paper that I provided is mentioned, there is no detailed discussion. That work is only briefly mentioned on pages 2, 3, and 4. Unfortunately, the authors did not seem to make enough effort to actually look into related works. In the current state, I cannot recommend the paper to be published.

Review #2
By Tobias Kuhn submitted on 27/Feb/2019
Major Revision
Review Comment:

The paper has improved in some respects from the previous version that I reviewed, but I am not convinced by how several of my main points of criticisms have been addressed (or not addressed). I will focus here on the main points of criticism from my previous review, quoting them with the authors' response below:

> - The restriction to OWL "individuals and concepts" is not well motivated. OWL
> axioms involving properties can be expected to be harder to understand, given
> that properties are logically more complex than individuals and classes.
> Therefore the restriction to individuals and classes is a serious one that
> limits the potential impact and usefulness.

> Restrictions on concept implicitly represent the TBox axioms in an ontology.
> We also agree that the axioms involving properties are harder to understand.
> For this reason we have refinement rules meant for property restrictions.
> Required rewritings are done in the Introduction (Page-3).

I am not satisfied with this answer. This restriction is still not well motivated. The authors only copy-pasted (you could also say "plagiarized") my sentences above ("OWL axioms [...] and usefulness") to their introduction, only replacing "classes" with "concepts".

> - The use of axioms like "cat subClassOf animal" to remove "has as pet an
> animal" if we already know "has as pet a cat" implies that we are starting
> from a consistent ontology. This means that we cannot use the presented
> approach as a method to find the errors in inconsistent ontologies, which
> actually is an important problem that such verbalization tools could help us
> with. This assumption and the resulting limitation are not discussed.

> We currently assume that the ontology is consistent. This details is included
> in Section 4.2. Considering the reviewer's comment, the mentioned point is
> included in the Future Work section.

The restriction to consistent ontologies is now mentioned but still not motivated. We only read once that "Currently we consider only those consistent ontologies for verbalization" in the technical parts of the paper and the evaluation section ends with a sentence about future work and inconsistent ontologies, which is again almost a verbatim copy of my review text. While we are now explicitly (but not very prominently) told that the approach assumes a consistent ontology, we still don't know why the authors think that this is a good restriction to make in the first place.

> - "if a role restriction R1 is implied by another role restriction R2 (i.e.,
> R2 ⇒ R1), then R1 can be said as a stricter version of R2.": That doesn't seem
> to be right. If R1 follows from R2 then R1 is less strict and not more.
> Whenever R1 is violated so is R2, but not necessarily vice versa. So, R2 can
> be violated while R1 is not, therefore R2 is "stricter" in its intuitive sense.

The response to this seems to be missing, and the text seems unchanged in the article.

> - The empirical evaluation doesn't seem to be a fair one, as the "traditional
> approach" is not really what is normally done. [...]

> In the empirical evaluation we wanted to show that even if we consider all the
> implicit axioms our systematic way of reduction could express the OWL
> statement in a much understandable way. It is true that existing approach
> generally considers a subset (mainly the explicit axioms) from the label-set
> for generating the NL text, however, this would not be equivalent to the
> verbalized text by the proposed approach.

It's true that the approaches are different, but the paper is written in a misleading way by claiming that it compares what is traditionally done with their approach. The baseline called "traditional approach" is not what other approaches are doing, and it seems quite obvious that the chosen baseline should be expected to perform much worse than what is normally done in existing work. So, it's really unclear what the current evaluation is telling us, and the paper hasn't improved in this respect. At the least, the label "traditional approach" is highly confusing and misleading.

My other main points of criticism from my first review with respect to unnecessary complexity and the use of formal language have been partly addressed or answered and can partly be seen as a matter of taste. Given that the other reviewers didn't report these as serious issues, I will not insist on them either and I can agree to consider them resolved, in contrast to the unresolved issues explained above.