Modification on the Generalization Graphs feature

28 views
Skip to first unread message

Maxence Robin

unread,
Jul 12, 2018, 8:28:46 AM7/12/18
to Unitex-GramLab
Hello everyone,

We are currently working on an update of the Generalization graphs, and we would like to know if some of you are using the "negation of a category" feature (see figure 12.13 of the manual) ?

Bests regards,

Maxence ROBIN


Daniel Stein

unread,
Jul 12, 2018, 9:10:15 AM7/12/18
to maxence...@gmail.com, Unitex-GramLab
Hello Maxence,

I do use the feature "negation of categories".

Kind regards,
Daniel

--
You received this message because you are subscribed to the Google Groups "Unitex-GramLab" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unitex-gramla...@googlegroups.com.
To post to this group, send email to unitex-...@googlegroups.com.
Visit this group at https://groups.google.com/group/unitex-gramlab.
To view this discussion on the web visit https://groups.google.com/d/msgid/unitex-gramlab/0f1a2517-b2d8-4899-9524-519e045cbdbf%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--
Daniel Stein
Frasenweg 61
34128 Kassel

Maxence Robin

unread,
Jul 12, 2018, 10:03:14 AM7/12/18
to Daniel Stein, Unitex-GramLab
Hi Daniel,

To add some context, the modifications we are working on are the following :

Currently it is possible to restrict the research by putting a category in the second box (the box that follows the 'G' box), as it's shown in the figure 12.14 of the manual. This works well but only on a sub token that has a "depth" of 1 in the main token, it doesn't work if the restriction target a sub token that is part of a token which is itself a sub token (a depth of 2 or more).

From now, it will be possible to apply restriction targetting sub tokens of any depths. The rule is that it keeps only the matching tokens that have the least depth, for example, on such a token :

{{sub one,.a} {{sub sub one,.a} {sub sub two,.b},.b},.main}

If the restriction is "a", both "sub one" and "sub sub one" will be kept, but if the restriction is "b", only "sub sub one sub sub two" (which correspond to the top level token that have the category "b") is kept.

The problem is that this rule is well defined for restrictions like this, but not for negative restrictions (negation of categories), because it is not clear if the research has to keep digging into the sub tokens if a top level token is rejected.

It seems that a graph written with "negative restrictions" can be rewritten with "positive restrictions" such that it still matches the same tokens.
Could you please send us an example of graph where you used the "negation of categories" feature ?

Bests regards,

Maxence

2018-07-12 15:09 GMT+02:00 Daniel Stein <daniel...@gmail.com>:
Hello Maxence,

I do use the feature "negation of categories".

Kind regards,
Daniel

On Thu, Jul 12, 2018 at 2:28 PM Maxence Robin <maxence...@gmail.com> wrote:
Hello everyone,

We are currently working on an update of the Generalization graphs, and we would like to know if some of you are using the "negation of a category" feature (see figure 12.13 of the manual) ?

Bests regards,

Maxence ROBIN


--
You received this message because you are subscribed to the Google Groups "Unitex-GramLab" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unitex-gramlab+unsubscribe@googlegroups.com.
To post to this group, send email to unitex-gramlab@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Unitex-GramLab" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unitex-gramlab+unsubscribe@googlegroups.com.
To post to this group, send email to unitex-gramlab@googlegroups.com.

eric.laporte

unread,
Jul 17, 2018, 4:32:06 AM7/17/18
to Unitex-GramLab
Dear Maxence,
Here at UPEM we use negative right contexts (Section 6.3.1 of the manual) without CasSys, and I think many other users do, even if they don't reply to your post.


On Thursday, 12 July 2018 16:03:14 UTC+2, Maxence Robin wrote:
<<
It seems that a graph written with "negative restrictions" can be rewritten with "positive restrictions" such that it still matches the same tokens.
>>
In theory, with a finite alphabet, yes, but in practice with Unitex, not always, because of ambiguities and lexical masks recognizing several tokens. Here are 2 examples of problems you can face if you try to replace a negative right context with normal boxes.
- If the content of a negative right context begins with <DIC>, you can't obtain exactly the same effect by replacing it with something that begins with <!DIC>, because <DIC> can match a multiword unit with several tokens, whereas <!DIC> matches only simple words (manual Section 4.3.5). Without negative right contexts, I can't imagine any way of checking that some position is not followed by a <CDIC>.
- If a negative right context is followed by other boxes, these boxes may match again a part the negative right context already successfully matched, and the limits of the objects matched in the two passes will not necessarily coincide. For example, one of them can check simple words and the other a multiword unit.

<<
Could you please send us an example of graph where you used the "negation of categories" feature ?
>>
You have an example of this style of graph in Fig. 6.17 in the manual.

Best regards,
Eric

Maxence Robin

unread,
Jul 17, 2018, 8:25:03 AM7/17/18
to Unitex-GramLab
Hello everyone,

It seems that there has been a misunderstanding with my first message, I was not talking about negative right contexts but about "negation of a category", which is a specific feature of the generalization graphs in CasSys (see figure 12.13 of the manual).

@Daniel Stein
About which one of these two features were you talking about in your message, negative right contexts or "negation of a category" ?

Thanks by advance.

Best regards,
Maxence

Daniel Stein

unread,
Jul 17, 2018, 8:36:18 AM7/17/18
to Maxence Robin, Unitex-GramLab
Hello Maxence, 
sorry, indeed a missunderstanding:I was talking about negative right context.

Kind regards
Daniel

--
You received this message because you are subscribed to the Google Groups "Unitex-GramLab" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unitex-gramla...@googlegroups.com.
To post to this group, send email to unitex-...@googlegroups.com.
Reply all
Reply to author
Forward
0 new messages