Re: lime module

Hi Philipp,

please find my comments below.

2015-07-15 8:06 GMT+02:00 Philipp Cimiano <cimiano@cit-ec.uni-bielefeld.de>:

>  Hi Manuel,
>
>  replying to this, todos from last Friday....
>
> Am 07.07.15 um 15:55 schrieb Manuel Fiorelli:
>
>   Dear Philipp, All
>
>  here are my preliminary comments. Most of them are minor typos, while
> other may seed further discussion.
>
> -----
>
> In the introduction to example 1, the spec says:
>
> "As an example we may describe a simple lexicon using this property as
> well as properties from Dublin Core and VoID: "
>
> The example then contains also the actual lexical entries that constitute
> the lexicon. This is good for what concerns the self-explanatory nature of
> the example. However, we should make clear that in general the metadata
> only deals with the description of the lexicon as a whole, while the
> representation of its actual content is in the scope of other modules. This
> is particularly relevant to "lexicon catalogs", which may only be
> interested in indexing lexicons without the need to also host the actual
> content.
>
>    I kept the example as is but added a sentence that makes clear that
> the metadata describes the lexicon as a whole as suggested by you.
>
>
It seems OK to me.


>   -----
>
> In the definition of LexicalizationSet, the classes Lexicon and Dataset need,
> respectively, the prefix ontolex and void.
>
>
> Fixed
>

Good

>
> -----
>
> I am not sure about this statement:
>
> "The lexicalization set object should be unique for a given
> lexicon-ontology pair"
>
> Indeed, the statement above imply that there cannot be two different
> lexicalization sets for FOAF using the WordNet RDF lexicon. I think that
> this conclusion is false, so the previous statement should be retracted.
>
>    This has been removed.
>

Good


>
>   -----
>
> In the definition of lexicalizationModel, the disjunction is spelled OR,
> whereas in other cases it is spelled in lowercase.
>
>
> has been fixed by you I guess, thanks.
>

Actually, I think that ontolex:Lexicon should not be part of the domain of
lexicalizationModel, unless we want to support lexica not expressed through
the OntoLex model. Indeed, this proeperty is meant to hold values such as
RDFS, SKOS, SKOS-XL, OntoLex,... while the use of a catalog of linguistic
annoatations is reported by the property linguisticModel.

>
> -----
>
> The definition of lime:references does not mention the fact that in a
> lexical linkset an ontology reference can be associated with a lexical
> concept.
>
>
> In order to avoid overlading, I would prefer to keep "references" as
> referring to the distinct number of resources ?o, that is:
>
> # of different ?o such that (?s,reference,?o)
>
>
I noticed that in the wiki the domain still contains lime:LexicalizationSet.
What have you decided to do? Maybe you want to coin another property, but I
have no suggestion for its name.

>
> -----
>
> Concerning Example2:
> - we should add the language "ja" to the lexicalizationSet resource
> - we may say that the ontology is an instance of voaf:Vocabulary, which
> is a subclass of void:Dataset to represent vocabularies (both RDFS
> Schemas and OWL Ontologies)
> - I would extend the introduction to the example. This is my attempt:
>
>  <cite>
> In the following example, we describe a lexicalization set expressing how
> elements of an ontology can be verbalized in Japanese by means of entries
> from a supplied lexicon. The metadata clearly tells which ontology and
> lexicon are involved in the lexicalization sets, as well as the relevant
> natural language. The knowledge of these facts about the lexicalization set
> allows us to assess the usefulness of a lexicalization set for a given task
> as well to discover relevant lexicalization sets, when we are constrained
> by the choice of an ontology, lexicon or natural language.
>
> We model the ontology as an instance of the class voaf:Vocabulary that is
> a kind of void:Dataset representing vocabularies (bot RDFS Schemas and
> OWL Ontologies). We benefit from the more specific distinctions made by
> VOAF, by breaking down the total number of entities in the ontology (held
> by the property void:entities) into separate counts for the classes and
> properties (held by voaf:classNumber and voaf:propertyNumber,
> respectively).
>
> Similarly, we use terms from the Lime vocabulary to represent statistics
> about the linguistic content of the lexicon and the lexicalization set.
> Overall, the ontology defines 80 entities and the lexicon 100 lexical
> entries; however, only 20 entities from the target ontologies have been
> associated with a total of 50 lexical entries.
>  </cite>
>
> -----
>
> Great, I have added your text to the example.
>
>
Good. However, I have just noticed that my introduction assumed the use of
some VOAF terms in the example: I will add the missings statements in the
example late this afternoon.


>
> In the definition of avgNumOfLexicalizations, it occurs the word "define"
> while it should be "defines".
>
>
> I can not find this, sorry.
>
> But this brings me to another issues. The formula for
> avgNumOfLexicalizations could be improved to make it clearer as follows:
>
> avgNumOfLexicalizations = # lexicalizations / # ontology entities in the
> reference dataset
>
> What do you think? Can you possibly update the formula? That would be
> great. Thanks.
>
>
I agree that the formula should be updated. I will try this afternoon to
produce a new formula, but I don't know if I can replicate exactly the same
appreance.


>
> -----
>
> I would postpone example 3 to end of the section, and I would modify it
> as follows:
> - reuse the same data as in example 2, and make this clear in the
> introduction to the example
> - then, use the properties lexicalizations, avgNumOfLexicalizations and percentage
> to "analyze" the scenario depicted in example 2. For instance, it is now
> possible to tell explicitly that only 25% of the reference ontology has
> been lexicalized.
>
>  We can make the example more interesting playing with polisemy so that
> the ratios are not "obvious".
>
>
> Actually, I think that example 3 makes definitely sense here. The ratios
> are rather obvious, true, but this is good as a simple and clear example.
>

I think we already agreed on this during the last call.

>
> -----
>
> In the definition of LexicalLinkset, the class dataset needs the prefix
> void.
>
> -----
>
>   OK, this has been fixed as far as I see.
>
>
Good

Best regards

Manuel Fiorelli