RE: From PSG to OWL

Thanks! I’ll get to wrok...

 

From: Aldo Gangemi <aldo.gangemi@cnr.it> 
Sent: dinsdag 26 november 2019 00:49
To: hans.teijgeler@quicknet.nl
Cc: Aldo Gangemi <aldo.gangemi@cnr.it>; Mikael Pesonen <mikael.pesonen@lingsoft.fi>; Semantic Web <semantic-web@w3.org>
Subject: Re: From PSG to OWL

 

Hello Hans, as far as I understand, you need to match a data structure (e.g. a knowledge graph extracted from text as in the case of applying FRED to definitions) to another one (ISO templates). I assume you can transform templates and/or input graphs appropriately.

Now, the problem is that templates contain general notions such as possessor, weight, scale etc., while the text may not use exactly that terminology (kilogram instead of scale, pump for the general type, etc.). And I assume that you do not have enough templates to cover all possible definition patterns.

 

My suggestion is to use some pattern matching technique to compare a sequence (a schema pattern emerging from a text e.g. through FRED) to a set of template sequences, and to rank them. Of course, you’ll need an expansion of template elements in order to provide enough material for the matching. I am pretty sure a seq2seq algorithm could perform pretty well with some background knowledge for the templates, which of course should be provided by you as domain expert. Alternatively, ontologies and lexical resources may be used for inferring the matching when available for the domain addressed.

 

My 2 cents

Aldo

 

PS when submitting special performatives to FRED (e.g. commands, questions, etc.), since the basic parser has been trained on declarative sentences, it might help paraphrasing the performative, e.g. “You should list all the elements” instead of “List all the elements”. This is a typical rewriting made at pre-processing time to optimise results without retraining the basic syntactic parser.

 





On 24 Nov 2019, at 12:02, <hans.teijgeler@quicknet.nl <mailto:hans.teijgeler@quicknet.nl> > <hans.teijgeler@quicknet.nl <mailto:hans.teijgeler@quicknet.nl> > wrote:

 

Hi Aldo,

 

Please read this and give me advice on how we should, roughly, proceed. I would be much obliged!

 

Regards,

Hans 

 <http://data.15926.org/> 15926.org

 


MAPPING WIZARD FOR ISO 15926


Introduction


In the ecosystem of ISO 15926 most things have been defined, but we miss a tool that makes use of these definitions in order to assist in the mapping from any source system of plant life-cycle information to the templates of ISO 15926-7/8.


Information Sources


Information that is to be mapped is structured and resides in data bases and spreadsheets. It will be required that for each data element a simple data dictionary with definitions is available or will be made available.


Defined Classes


The following is defined:

1.  <http://15926.org/topics/data-model/index.htm> ISO 15926-2 – the upper ontology of ISO 15926 in OWL, with:

a. 93 generic Classes
b. 129 generic Relationships classes

2.  <http://15926.org/topics/reference-data/index.htm> ISO 15926-4 Reference Data Library (RDL) in RDF ( <http://data.15926.org/rdl/RDS327239> here is an example), with endpoint-based extensions that can be federated, all in strict taxonomies that are instances and/or specializations of ISO 15926-2 entity types. Alltogether approx. 40,000 Classes and Metaclasses.
3. The RDL and its extensions are gradually being extended with ontologies, in RDF format.


Destination


The above source information needs to be mapped to  <http://15926.org/15926_template_specs.php> ISO 15926-7/8 templates in RDF, along with the required declarations of the predicate objects. 

There are 225 Template Specifications, including a FOL definition in terms of ISO 15926-2 entity types. In this context it is useful to mention that each template class has a label like this one:

[EssentialType][hasPossessor] has a [hasPropertyType] of [valPropertyValue] [hasScale]

where the text between brackets is replaced with the labels of the objects in the predicates e.g.:

[PUMP] [P-101] has a [WEIGHT] of [137.5] [KILOGRAM]

where  <http://data.15926.org/rdl/RDS327239> PUMP,  <http://data.15926.org/rdl/RDS7285420> WEIGHT and  <http://data.15926.org/rdl/RDS1328669> KILOGRAM are defined in the RDL.


Tool


The question is: how can we (semi-)automate the mapping from source data to template instances and related object declarations? With so much defined in the background that should be possible (I think :)  )

 

____________________________________

From: Aldo Gangemi <aldo.gangemi@cnr.it <mailto:aldo.gangemi@cnr.it> > 
Sent: donderdag 21 november 2019 22:36
To: Mikael Pesonen <mikael.pesonen@lingsoft.fi <mailto:mikael.pesonen@lingsoft.fi> >
Cc: Aldo Gangemi <aldo.gangemi@cnr.it <mailto:aldo.gangemi@cnr.it> >; Semantic Web <semantic-web@w3.org <mailto:semantic-web@w3.org> >
Subject: Re: From PSG to OWL

 

Hello Mikael, if you need a semantic analysis of text into OWL (modulo some approximations needed when dealing with formal semantics of natural language), FRED [1][2] is a solution: it extracts RDF/OWL triples integrating multiple parsing, linking, and disambiguation components, and applying ontology design patterns.

 

Aldo

 

[1]  <http://wit.istc.cnr.it/stlab-tools/fred/> http://wit.istc.cnr.it/stlab-tools/fred/

[2]  <http://semantic-web-journal.org/system/files/swj1379.pdf> http://semantic-web-journal.org/system/files/swj1379.pdf






On 21 Nov 2019, at 16:52, Mikael Pesonen < <mailto:mikael.pesonen@lingsoft.fi> mikael.pesonen@lingsoft.fi> wrote:

 

Hi,

given phrase structure grammars of some text, are there any theories, work or utilities done for extracting knowledge from them into semantic (OWL) form?
There are lots of steps involved, for example identifying the entities and relations between entities.

Br