Hi Colin,

recently I came across the open multilingual wordnet initiative http://compling.hss.ntu.edu.sg/omw/. I didn't read it yet in full detail, but maybe it could be helpful for you.

Hi Thomas,

I appreciate your response, which then made me realise that I had not made my main need clear. I think the multilingual aspect is, and will become more important, but my focus at this time is the existence of cloud-based resources that I can either consider as ontologies (knowledge bases?) in themselves, or resources that will help me build my own (ontologies). I look for syntactic support - lemmas, "is a" relationships, synonyms, etc., as well as (hopefully) semantic support - such as started by FrameNet, which I will use to build my own models using a graph database.

Wordnet, the whole Wiki series, BabelNet etc. offer partial help - and I have also discovered ConceptNet, which looks promising. What I would really appreciate from any resource I find is reliability and comprehensive coverage, as well as easy REST API access. I will work with English for the foreseeable future.

I am not sure that I understand your needs but perhaps can you look at these and give me a feedback :

Ontology from the Library of Congress : http://id.loc.gov/ontologies/lcc.html

Thesaurus from the Unesco : http://vocabularies.unesco.org/browser/thesaurus/en/

The base structure is made up of SKOS relationships (https://www.w3.org/2004/02/skos/).

Thank you very much for these two links. I will need to investigate them further, but they both look to be important resources for any natural language understanding work (mine included), particularly in domain-specific contexts.

I expect that I will be using "general" ontologies ("upper level ontologies") together with domain-specific ontologies in any natural language understanding task,.

I am interested in both types of ontologies that are cloud-based and accessible.

Not sure if I understand the question, but maybe http://conceptnet.io/

Hi Muhammad,

I did mention ConceptNet above, but thank you anyway. As you are not the first to say "Not sure if I understand the question", perhaps I should try to clarify.

It is not that I know, at this point, exactly how I will proceed - I am feeling my way. My overall objective is to build an "intelligent agent" (for want of a better word) that will be able to understand natural language. I am using the word "understand" in the deepest sense, and I will focus on this capability in the context of conversations - not big data or huge text corpora mining.

I will lean heavily on the use of one or more cloud-based "ontologies" in order to determine the meaning of words and phrases. This typically means that I will discover "pieces" of taxonomies/ontologies that contain the words and phrases looked for, and incrementally add them to a local knowledge base/ontology. This would be done sentence by sentence. This is unlike the large-scale NLP that I read about everywhere.

I would also like to find resources that support semantic processing - for example the kind of thing I read about on the FrameNet site. This would enable me to add representations of meaning to my knowledge base, to be used to understand received sentences, and hopefully to be able to construct meaningful responses. As far as I can see, though, FrameNet, with all its promise, has not grown into a usable resource for my purpose. I am not aware of any other similar accessible resources.

I would love to know what progress has already been made in the same direction.

Colin Goldberg

Given that this is a topic that does not deserve a shallow answer, I will do my best to provide updates to my own investigations as they occur.

Please correct me if I am wrong or have missed something:

- As powerful and comprehensive as the Library of Congress ontology classification database is (http://id.loc.gov/ontologies/lcc.html), it is a classification system more that an ontology as I see it. It is available via REST API, so that is one good thing. But their documentation is still in its early stages, and it requires a librarian's understanding of Subject Headings and Authorities, etc., so it is difficult to assess. It does not seem to apply, for example, to my need to submit a single word or phrase - that may not exist as a "publication", and determine where it sits as a "concept" in a hierarchy of concepts.

- The Unesco Thesaurus (http://vocabularies.unesco.org/browser/thesaurus/en/) does appear to be a usable ontology, but it does not have a REST API - they provide a link to download the database. So it is not as usable as I would expect from a modern, web-based ontology.

- BabelNet (https://babelnet.org) is a comprehensive (and multi-lingual) ontology, and it has a REST API. I was able to obtain an API key and access it. However, although it provides the kind of ontology information that is useful, it appears that it may not be a resource that I could rely on in a "production" capacity - it is not ready for "primetime". For example, it provides no support if I have a question.

- ConceptNet (http://conceptnet.io) is an available ontology resource which does have a REST API. It also has a forum where questions can be asked. I still have some questions about its ability to provide for a complete traversal in both directions, but I consider this a valuable ontology resource.

- SUMO (https://en.wikipedia.org/wiki/Suggested_Upper_Merged_Ontology) is a very interesting resource, which incorporates logic programming in its provision of an ontology. I believe this is one of my most interesting finds so far. I am still investigating.

I hope you find this useful. Please let me know if you find any glaring errors or omissions.

Colin Goldberg

For those who are following this thread, my original question was about web-based ontologies that are accessible via a REST API - a requirement that was quite specific, and also asked when I had no perspective on what is "out there".

I have since come across other resources, in addition to the ones above, only one of which has REST API access, which I list below.

Yago (http://yago-knowledge.org), built by the Max Planck Institute for Informatics, is available for download, and not via an API. Described as "a huge semantic knowledge base, derived from Wikipedia WordNet and GeoNames. Currently, YAGO has knowledge of more than 10 million entities (like persons, organizations, cities, etc.) and contains more than 120 million facts about these entities", I would say it focuses on named entities, and is not what I was looking for as an ontology.

Umbel - Upper Mapping and Binding Exchange Layer (http://umbel.org, http://www.mkbergman.com) - is available via an API. However, it seems to have been succeeded by KBpedia (http://kbpedia.org), which, although open-source, is not accessible via an API, only downloadable. KBpedia has an upper ontology (KKO), and is described as "an amalgam of Wikipedia, DBpedia, Wikidata, GeoNames, OpenCyc, UMBEL". KKO is described as "based on the universal categories and knowledge representation theories of the great 19th century American logician, polymath and scientist, Charles Sanders Peirce". However, I found KKO confusing.

Adimen-SUMO (http://adimen.si.ehu.es/web/AdimenSUMO), described as "an off-the-shelf first-order ontology that has been obtained by reengineering out of the 88% of SUMO", is available for download, just like SUMO. I look forward to investigating it and learning more about how it - like SUMO - can be used by First Order theorem provers for formal reasoning.

Cyc (http://www.cyc.com/researchcyc) is described as available to the research community. Although I have not applied for a license to use it, a look at it's upper ontology and articles about the use of higher order logic make it an interesting candidate for future investigation.

If you know of other web-based, accessible ontologies, I would appreciate pointers to them.

Kind regards

Colin Goldberg

I have been researching many of the same sources of knowledge bases albeit for a different purpose. I am also looking for a full-featured ontology. My focus is on finding an ontology with a rich set of relations between entities. I've found most focus on hierarchies (hypernyms/ hyponyms aka sub-types / super-types relations), or other WordNet relations (holonyms / meronyms aka part-whole).

KBPedia seems focused on the topologies only, with connections to the sources in Wikipedia and other sources.

SUMO (http://www.adampease.org/OP/) seems to be closer to an ontology.

Babelnet.org and Conceptnet.io seem to be richer as ontologies and perhaps easier to understand than SUMO. I'm focusing my investigation on these.

Lastly I would not bother with Cyc, since OpenCyc has been shuttered for some time now, and KBPedia still incorporates some of OpenCyc.

Regarding APIs and technology, I'd expect that a serious web-based application of a conversation agent would need its own API, with its own knowledge base for the domain(s) you're interested in addressing. Even a general agent like IBM's Watson of Jeopardy fame had a specific but large knowledge base, and inference engine, NLP etc. It shouldn't matter if an ontology does not have a REST API or just a database, since you'll want to use it on a server and behind your own API. Neo4j would be excellent in this regard, since you can import many of these ontologies or Wikipedia / Wikidata or similar knowledge bases using SPARQL and the neosemantics (n10s) plugin.

Thanks very much for your input. I have spent some time with SUMO, which I like, as well as conceptnet.io. The restrictions that I came across with babelnet put me off a little - not to say that this won't change.

I agree with your assessment on the whole API question. (It's a lot of work). And thanks for your comments on Neo4j.