{"items": [{"author": "Daniel", "source_link": "https://www.facebook.com/jefftk/posts/622183028102?comment_id=622200982122", "anchor": "fb-622200982122", "service": "fb", "text": "I feel that \"at least\" should be grouped as a single word, not split into two. Drawing the tree as (ADVP (RB at) (RBS least)) would seem to imply that other RBs could be swapped in for \"at\", which doesn't seem to be the case. Consider \"There were {quickly**|mostly**|usually**|at} least five people wearing plaid shirts at the party today.\" I also don't know what the RBS category is supposed to represent - superlative adverbs? What?", "timestamp": "1375154050"}, {"author": "Daniel", "source_link": "https://www.facebook.com/jefftk/posts/622183028102?comment_id=622201680722", "anchor": "fb-622201680722", "service": "fb", "text": "Also, what function is the trace thing serving in those parses? My parser has no notion of \"trace\" and can handle those WH-ADVPs without a problem. \"At least three analysts have issued forecasts in the last 30 days\" is just a regular sentence, there's no need to suggest that there is some strange ghostly implied structure hidden in it.", "timestamp": "1375154332"}, {"author": "Nicolas", "source_link": "https://plus.google.com/105722903272192091597", "anchor": "gp-1375194129437", "service": "gp", "text": "Are you working with OntoNotes again, or just looking for fun?", "timestamp": 1375194129}, {"author": "Jeff Kaufman", "source_link": "https://plus.google.com/103013777355236494008", "anchor": "gp-1375194774927", "service": "gp", "text": "@Nicolas\n\u00a0looking back at it for fun.\n
\n
\nYou're still at BBN? \u00a0More NLP?", "timestamp": 1375194774}, {"author": "Nicolas", "source_link": "https://plus.google.com/105722903272192091597", "anchor": "gp-1375196110664", "service": "gp", "text": "Yup! Working on a slightly different domain extraction task mostly right now.", "timestamp": 1375196110}, {"author": "Jeff Kaufman", "source_link": "https://www.facebook.com/jefftk/posts/622183028102?comment_id=622558815022", "anchor": "fb-622558815022", "service": "fb", "text": "@Daniel: \" Drawing the tree as (ADVP (RB at) (RBS least)) would seem to imply that other RBs could be swapped in for 'at'\"
I don't think that's their criteria. I think it's just \"assign part of speech to every space separated token\".
\"superlative adverbs?\"
Yes. For example, \"most\" in \"That's the most angry I've every seen you.\"
\"trace\"
I mostly understand traces with wh-movement, but not elsewhere.", "timestamp": "1375358629"}, {"author": "Daniel", "source_link": "https://www.facebook.com/jefftk/posts/622183028102?comment_id=622560511622", "anchor": "fb-622560511622", "service": "fb", "text": "But are there any members of the RBS category other than \"most\" and \"least\"? I would consider \"most angry\" to be just a conjugation of \"angry\". I'm skeptical of the idea that every space separated token has its own part of speech.", "timestamp": "1375361040"}, {"author": "Daniel", "source_link": "https://www.facebook.com/jefftk/posts/622183028102?comment_id=622560751142", "anchor": "fb-622560751142", "service": "fb", "text": "The point about separate tokens and parts of speech is a good example of one reason why I think the annotation paradigm doesn't work very well. If the annotators make a mistake in their conception of what the parse trees look like, this mistake will be replicated by all the high-scoring automatic parsing systems. It is kind of like a teacher who gives an A to the student that can most precisely regurgitate exactly what the teacher himself said, while flunking the student who sits in the back and actually understands the subject well enough to correct the teacher's own mistakes.", "timestamp": "1375361354"}]}