Wordnet Integration
Integrated Semantic Framework (Bond, Tuan Anh)
Leader: Francis
Scribe: Glenn
Glenn: Pre- vs post-processing… Your nomenclature when dealing with Victorian English?
Francis: Post. Removing exuberant hyphenation: better to let ERG get a chance first and then fix afterwards.
John: Is the wordnet tagging from semcore?
Francis: No there’s not enough semcore treebank.
Dan: Tim Baldwin challenged me to treebank 3000 sentences [story…] Lost in the mists of time.
Francis: Chris [?] Doesn’t like people to use semcore. Because the human tagger went mad.
John/Francis: It wasn’t sable enough. Things shifted underneath the annotators.
Dan: Would there be value is a chunk of semcore that is stable and work with that?
John: Not many people are doing WSD these days.
Francis: Trying to persuade people to use wordnet, but even more so the MASC annotated portion of the ANC.
Ping: Do you have a way to traverse wordnet hyper/hyponyms
Francis: Yes. We use such to do cross-lingual alignment.
Ping: Which wordnet languages?
Francis: English, Chinese, Indonesian, Japanese
Ping: Where did you get the Chinese Wordnet?
Francis: We made it.
Ping: HouNet.
Francis: Very good, but proprietary; we can’t use it.
Ping: What about granularity considerations when you built your Chinese wordnet?
Frances: Yes and no, for Chinese we largely copied the structure of the English wordnet, but that may not have been a good choice.
Ping: First-level concept vs. meta-concept…?
Francis: Justice is an abstract concept; I think it’s good enough if it can exist in your head. I consider the things spoken by scholars to be part of the language. Post-modern literary examples were weird.
Woodley: I think you partially solved the mundane problem of correlating … with … so are we now ready to drop this into our paraphrasing machinery. To use some of our tools for more non-trivial entailment. Are you doing this by hand..
Francis: We are doing this by hand; we’re working towards it.
Emily: You said it would be useful to have give_up_v2 as “give” + “up”; Seems like that would be something better off put to the side, i.e. SEM-I, rather than mucking with pred values vs. stem forms.
Dan: “Turned on his friends”
Francis: “Give-up” will be indexed by … in wordnet. We need to be able to find it.
Dan: There are verb predicates that say it could be selected by either of two.. “Either of,” “either about”.. I’m with Emily, this doesn’t belong in the grammar. We should figure out how to store it in the SEM-I. It’s not a very mysterious mapping.
Francis: It can be.
Dan: We should keep straight what’s inside vs. what’s outside here and not make mistakes as before.
Francis: As a user-grammarian, the process of learning to maintain the SEM-I is an extra hurdle.
Mike: There are some tools in the logon dist that can help maintain this stuff outside the grammar (SEM-I)
Dan: There’s a lot of chaff here, we need to capture patterns for these particle associations.
Oe: Conceptually, what are the two pieces we trying to associate?
Francis: For this lexicon entry, if I’m looking it X dictionary, here’s how you should look it up (under). “Give up” must/should be looked up thusly.
Oe: You are saying there’s a conventional citation form for many words…
Francis: …all languages…
Oe: A conventional predominant citation form…
John: Different types of dictionaries have different conventions for this.
Francis: Yeah, yeah, it’s always going to be slightly messy, but in my experience, a well-known, well-understood citation form is sufficient.
All: it could be a separate table
Oe: Do we associate semantic predicates in the SMI with these aforementioned citation forms? Then there’s also …. ? … There are different arguments to be made.
Francis: […]
Emily: Why do you want to do this? What are all these use cases…
Francis: Finding frequencies, I don’t want to throw information away. But meanwhile, I’m happy to use a standalone table. And try to link it through the SEM-I.
Woodley/Emily: Mapping to wordnet synsets contain lemmas or canonical form already probably.
Francis: I was thinking of two tables. One that has the give_v_up to “give_up” and one that gives “give up” to “give_up_238948345”. A single synset might have multiple…
Woodley: The pivot seems to naturally be through wordnet, maybe not the table you’re introducing?
Francis: No, I prefer to keep them under my control as two separate things because I may not agree with wordnet and want to override. Sounds like some people might like to see these when they’re ready…
Woodley: What degree is the mechanism for maintaining these usable by other grammars…
Francis: Not much, unfortunately.
Last update: 2017-08-08 by GlennSlayden [edit]