Re: Call on the 10th of October, 14:00 CEST

Dear Tom,

Thanks for sharing your thoughts. Let me comment on a couple of points you
raised:

* about context, sense-ordering, and examples

> In the future, where a electronic personal assistant holds a conversation
with the user,
> the context is known and should re-prioritize word choices and samples,
even withhold
> dictionary entries that are clearly not in context.
> The computers can do that, I would not spend much time on in what
sequence the
> entries should be displayed/stored. If any, I would focus on the ability
of the computer
> to select based on context.

In my view, this translates into a model that supports ANY order criteria,
being agnostic of the particular sequence in which the entries should be
displayed/stored, which is left to the dictionary creator's choice or to a
NLP service that dynamically assigns it according to the context. In
principle we should focus on the modelling side (which will be discussed in
issue 6).

* about provenance

> What I think is important is that the sources of the dictionary entries
are tracked,
> in order to track back where entries came from, until they are confirmed
from
> several reputable sources.

There are "standard" metadata vocabularies (PROV-O, DC-Terms) that support
provenance description. I do not think that more extra properties are
needed in the case of dictionaries. However I think it might be useful to
describe a new issue or "best practice" to clarify how to do that.

Best,

Jorge




2017-10-10 1:19 GMT+02:00 Tom Knorr <tknorr@neurocollective.com>:

> I have some observations to the current discussion on the open issues.
>
> I1: The German ‘Leiter’ can be female gender and translate into en:ladder,
> a noun or it can be male gender and translate into en:leader, either noun
> or a role, if you support that. Both are not or if any, very esoteric
> related, one could maybe construct that if you move up the corporate ladder
> you end up being a leader. Other than the article they have a common
> morphology but they should not be the same lexical entry.
>
> A bit more interesting is the German ‘Bank’. ‘Die Bank’, the bench has a
> plural of ‘Bänke’ while ‘die Bank’, the place you bring your money to, has
> a plural of ‘Banken’. Entirely different morphology but same gender.
> Definitely different lexical entries.
>
> Of the top of my head these are examples of many I encountered processing
> Europarl and the German Wikipedia dump. It required us to modify some code
> to properly map the morphology to the corresponding lexemes.
>
>
> I have a opinion about the usage examples and ordering of dictionary
> entries.
>
> Usage examples are useful but they should come from the semantically
> linked data that is behind the dictionary entries. I think we need to take
> a hard look at how the systems are going to be used in the future. In a
> printed dictionary, even if it is a translation app, the publisher does not
> know what the user intends to look up, therefore the ‘printed’ dictionary
> needs a representative set of examples and a ordering that is based of some
> acceptable scientific reasoning. In the future, where a electronic personal
> assistant holds a conversation with the user, the context is known and
> should re-prioritize word choices and samples, even withhold dictionary
> entries that are clearly not in context. The computers can do that, I would
> not spend much time on in what sequence the entries should be
> displayed/stored. If any, I would focus on the ability of the computer to
> select based on context.
>
> What I think is important is that the sources of the dictionary entries
> are tracked, in order to track back where entries came from, until they are
> confirmed from several reputable sources.
>
> I will try to catch the call tomorrow, if I get out of bed in time. The
> call is at 5am PST for me. If not I will read the transcript and answer
> through the list.
>
>
> regards
>
> Tom
>
> On 10/02/2017 07:28 AM, Philipp Cimiano wrote:
>
> Dear all,
>
>  I propose we have another ontolex telco on the 10th of October, 14:00 CEST.
>
> I propose we continue discussing the concrete examples that Julia and
> Jorge have been preparing.
>
> I think the conclusion we had is that we wanted to continue working
> bottom-up from examples of current lexica and then try to get an
> abstract model that is able to accomodate future dictionaries that are
> native LLD dictionaries.
>
> Let's try!
>
> We will again the meeting via skype. It worked quite well last time.
>
> Greetings,
>
> Philipp.
>
>
>
>
> --
> Tom Knorr
> Independent Consultant currently working on The NeuroCollectivewww.NeuroCollective.com
>
> Blog: http://www.neurocollective.com/blog/
>
> LinkedIn: https://www.linkedin.com/in/tom-knorr-4406406/
>
>


-- 
Jorge Gracia, PhD
Ontology Engineering Group
Artificial Intelligence Department
Universidad Politécnica de Madrid
http://jogracia.url.ph/web/