Tags used in KnowBuddy™
This document addresses how to configure KnowBuddy to present parse tree and part of speech disambiguation dialogs.
The following table documents the labels that KnowBuddy may present in part of speech or parse tree dialogs:
When disambiguating text using KnowBuddy, the Settings Dialog includes the following options on its “Analysis” tab:
When the “choose” options are checked, the Linguist will prompt for your help with disambiguating among multiple parts of speech that may occur for some words and among the various tree structures that typically arise in any sentence longer than just a few words.
There are various blog posts at http://www.haleyai.com that discuss and demonstrate the use of these dialogs for various sentences. Here we review the tags used in this dialogs and offer some suggestions as to how you might use these dialogs in practice.
To start, let’s review the tags you might see in a typical “parse tree”, as rendered below:
The tags abbreviate the following:
· NP noun phrase
· VP verb phrase
· V verb or verbal construct (other than a verb phrase)
· DET determiner (e.g., a, some, the)
· N noun or nominal construct (other than a noun phrase)
· P preposition (other than a prepositional phrase)
· PP prepositional phrase (including particles, such as a particle of a verb)
People often find the difference between “P” and “PP” difficult at first.
· in which case, it’s best simply to avoid the decision (i.e., make other selections or cancel the dialog)
Consider the particle “up” in “look up this word”:
· it acts as a PP and “this word” acts as NP that is not part of a PP
Here is an example that may be confusing, for example:
One can see someone in and one can see someone perform and one can see someone’s bid. The computer does not know enough about these things nor about parks, unfortunately.
In this case, the correct choice would be P but you are unlikely to be presented with this dialog after using the dialog to approve or vetoes parts of parses. It’s quite likely that you would either have vetoed “saw(the(man))(in)” or done so indirectly by approving “in(the(park))” or “in(the(park(with(the(telescope)))))”.
The following dialog arose in the course of disambiguating a sentence for which natural language processing initial returned no parses (because of the complexity of the untagged and unstructured sentence, most due to its use of many connectives and prepositions):
Here the additional tags are as follows:
· XP can be interpreted as a phrase with something “extra” or “extraordinary”
o e.g., a dangling “and”, as may occur in informal communication, such speech or tweets
· ADV means adverb or adverbial
o the use of “and” as an adverb is “extraordinary” here
o when parses are found (e.g., after vetoing or approving some parts of the returned parse structures) this is unlikely to recur
Generally, labels having “X” in them are rarely affirmed, but consider the following:
Here, the extraordinary phrase (XP) complements a verb phrase (VP), hence the label “VP/XP”.
· The same label applies to “drink” interpreted as a standalone phrase (but this is only one interpretation as we will see in a moment).
In general, “-C” appended to a label means “conjunct” as in one or more phrases “conjoined” together (e.g., using “and”). Thus, “VP-C (and(be))” depicts the verb phrase (not just the verb!) “be” as a “conjunct” ready to be “conjoined” with something to the left of it in the sentence.
Here, we right click on “VP-C(and(be))” since we want “be(merry)”. After another click, we get to the following:
With is more like it. Note that “AP” means “adverbial or adjectival phrase”.
Here is a dialog asking about the parts of speech of some words in a sentence for which no derivations (a.k.a., parses) were immediately found by the natural language processing system:
The parts of speech here are as follows:
· N noun
· V verb
· DET determiner (unusual for “as”)
· ADJ adjective (unusual for “as”)
· P preposition (unusual for “and”)
· ADV adverb (unusual for “as” and “and”)
The unusual cases here arose only because the NLP tried everything it could, including the most remote possibilities, to parse this sentence.
Selecting “employed” as a verb, “and” as a conjunction, and cancelling the dialog before reparsing is recommended in this example.
As it turns out, all it needs is a little guidance about the structure, such as bracketing the following phrase (or increase the numbers in “Parsing” tab of the Settings Dialog).
· student aides in the special schools of the State Department of Education and in the public schools of the state