acoli-repo / olia Goto Github PK

cf. LOV recommendations

• submit OLiA ontologies to LOV, integrate feedback

With milestone 1.0, we will apply numerical version numbering. At the moment (= as long as we stay backward-compatible), we operate with 0.*x* (without a numerical value for x).

At the moment, OLiA is focusing on corpora with morphological or syntactic annotation. Semantic annotation has been excluded (unless addressed in such corpora or dictionaries) as we assume that this requires an agreement among developers of semantic resources such as WordNet, PropBank and FrameNet. Such efforts are underway, but current results (e.g., SemLink v.2) are too unstable to incorporate them into OLiA.

Most likely, these specifications should move into a separate reference model and current specifications for semantics should move there, too. This will break backward compatibility, hence milestone 1.0.

• Rename top-level categories (MorphosyntacticCategory) to units (MorphosyntacticUnit)
• Integrate OLiA Discourse Extensions
  [please document/discuss other requirements in this direction under this issue]

(breaks backward compatibility => milestone 1.0)

Right now, annotation models are a plain list. To more easily find relevant ones, they should have a shallow hierarchical structure:

• /lang/ for language-specific schemes for morphology, morphosyntax or syntax, using BCP47 language codes
• /multi/ for multi-/crosslingual/language-independent schemas for morphology, morphosyntax or syntax
• /discourse/ for discourse
  etc.

Breaks backward compatibility => milestone 1.0

at the moment, OLiA uses CamelCase URIs, but rarely rdfs:labels. To be added.

The OLiA Reference Model is a monolithic knowledge graph for annotation terminology. However, this makes it relatively large and hard to use, and for a future OLiA releases, we suggest to break into different levels of description (syntax, parts of speech, [morphosyntactic] features, morphology, semantics, discourse) with corresponding sub-URIs, e.g., http://purl.org/olia/pos#Noun, http://purl.org/olia/feats#Singular, http://purl.org/olia/syntax#Subject, http://purl.org/olia/morph#Diminuitive, etc.). Then, the corresponding parts of the current olia-top.owl should be distributed across these, as well.

This breaks downward compatibility, hence milestone 1.0.

Aside from linguistic annotations, OLiA now (2022) also provides models

• for grammatical features in machine-readable dictionaries (e.g., LexInfo), and
• for general linguistic terminology (BLL thesaurus)

The term "annotation model" is not really adequate for these applications.
The suggested change pertains to the documentation, only, but it sets it apart from all publications on OLiA so far. Its implementation should thus be aligned with milestone 1.0.

Extension to make sure that word-specific tags can be reproduced when mapping tags via OLiA.

Suggestion:

• If hasLemma is defined in an annotation model, a particular instance is restricted to words with this exact lemma. hasLemma must be unique, if multiple lemmas are to be matched, use hasLemmaMatching
• if hasLemmaMatching is defined in an annotation model, a particular instance is restricted to words whose lemma matches this exact regular expression.

From a review to an LREC-2020 paper:

. Some annotation schemes devise classes so that particular words
will always have the same tag, even though in particular sentences
they have different uses.  In LOB, for example (which I take as an
example because the manual is handy), the occurrences of "all" in
the two sentences "All mothers go there" and "Let all pray for
peace" are both tagged ABN.

Other annotation schemes may be devised which attempt to tag the
attributive and pronominal uses of "all" with different tags.

In the one scheme, the concepts of the tag set relate (at least in
these cases) to word forms like "all", and the intension of the
tag is, roughly "a word form which may sometimes be used as a
determiner and sometimes as a pronoun and ... (further
specification)".  In the other, the concepts of the tag set
relate, in the same cases, not to word forms but to word
occurrences, and the intension of the tags will be "a word token
used as determiner of a noun ..." and "a word token used
pronominally ...".

• create a central Makefile
• automated XML, RDF and OWL validation (#5)
  • XML validation with xmllint (=> make release)
  • RDF validation with rapper (=> make release)
  • drop tools/validate_with_bash.sh
  • drop tools/validate_with_eyeball.sh and tools/eyeball (Eyeball is deprecated)
  • integrate OWL validation with Pellet (=> make validate)
• automated validation of ontology metadata (#1)
  • check rdfs:labels (#4)
• copy valid ontologies (from stable/, core/ and experimental/) to docs/owl
• update Purl links to point to https://acoli-repo.github.io/olia/owl (cf. #10)
• automated generation/refresh of docs/html
  • update/replace/automatize tools/olia2html
• automated generation/update of docs/models.md

Via purl.org, I created a partial redirect for http://purl.org/olia to replace earlier direct redirects for individial files. In order for this to work for all files in the stable/ folder, their individual entries must be removed. For many, this has been applied already. However, the API is not reliable, and in many cases, the DELETE operation (despite reporting success) was not applied.

Note: This is a severe bug, as the original URLs do not resolve anymore.

TODO:

• Perform weekly checks at purl.org as to whether all direct entries for stable/ ontologies have been deleted.
• If not, apply the deletion for the top 10 files.

at the moment, this seems to be in the ontologies, only.

• request: mirror, link, and (optionally) revise the named entity categorization and linking from NERD
• objective: NERD is a valuable resource but hasn't been maintained for 9 years; as of Sep 2022, the original website is down, but the code is still available via GitHub
• suggestion: integrate into OLiA Discourse Extensions (rather than OLiA Reference Model) because it has a relationship with (co-)reference

• rename to unimorph:label (used in paper)
• develop vocabulary to automatically map this to system:hasTagStartingWith, system:hasTagContaining, system:hasTagEndingWith [idea: additional property that defines separator symbols => generate all combinations of separators and the label to enable partial matches)

Remove deprecated URIs (will break backward compatibility, do not apply to the current master branch).

to be developed and applied.
must include

• XML validity (make release)
• RDF validity (make release)
• OWL-DL/2 validity (make release, more feedback with make validate)
• URI resolution (make checks)
• licence declaration

Purl URIs are extremely hard to maintain. Migrate to W3ID or another solution for persistent URIs.

as part of the revised validation process, this ontology (and some others) failed

• SemAF discourse core: linking model
• PDTB3 (!)
• https://www.ims.uni-stuttgart.de/forschung/ressourcen/korpora/dirndl/
• https://github.com/duncanka/BECauSE
• https://github.com/UniversalAnaphora/UniversalAnaphora/
• https://github.com/chardmeier/parcor-full
• https://lrec-virtual.org/presentation/lecture/openel-annotated-corpus-entity-linking-and-discourse-open-domain-dialogue#
• https://lrec-virtual.org/presentation/poster/causal-news-corpus-annotating-causal-relations-event-sentences-news
• https://github.com/merelscholman/DiscoGeM
• http://corpora.aifdb.org/qt30 (https://confcats-event-sessions.s3.amazonaws.com/lrec22/papers/final/605/605_Paper.pdf)
• https://confcats-event-sessions.s3.amazonaws.com/lrec22/papers/final/1169/1169_Paper.pdf

https://github.com/duncanka/BECauSE

OpenMinTeD interoperability criteria: A number of quality criteria, most of which can be easily improved

• Licensing information must be included in the metadata
• Content resources must include metadata on their language(s)
• It should be possible to determine the source of an annotation/assigned category
• Licensing information should be expressed in a machine-readable form
• Classification metadata should be included, where applicable, in the metadata record of the resource
• S/W (tools, web services, workflows) must indicate whether they are language-independent or the language(s) of the resources they take as input and output
• S/W (tools, web services, workflows) must indicate format of their output
• Documentation references should be versioned
• Documentation of GATE components
• Encoding citable publications (for scholarly attribution) in resource metadata records
• Including license text in resource packages
• Making models and annotation resources accessible as entities distinct from the components they are compatible with
• Encoding funding information in the metadata descriptions of all resources
• Encoding metadata on domain/subject/ classification for all resources when applicable

move from Frankfurt to GitHub, update Purls

For integration in NLP2RDF/NIF workflows, there used to be a live mirror under http://olia.nlp2rdf.org. However, this seems to be defunct. As publication now involves a different process operating directly on GitHub, the following files should be removed or moved into a different sub-folder:

• generate_csv_for_annotationmodel.sh
• generate_csv_for_linkingmodel.sh
• generate_java_map_from_csv.sh
• OLiAMap.java
• publish.sh

At the moment, these are being preserved because it is unclear whether there are NIF or Stanbol workflows depending on that.

Make sure / provide validator for

• The ontology is automatically retrievable and parses (make release)
• License statement provided as separate document (make release)
• The license statement achieves minimial interoperability, can be validated with Archivo SHACL tests, cf. https://archivo.dbpedia.org/rating
• LODE conformity (w. SHACL)
• successfull consistency check by reasoner (make release)
• submit ontologies to archivo