- Notifications
You must be signed in to change notification settings - Fork15
ayoungprogrammer/Lango
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
Lango is a natural language processing library for working with the building blocks of language. It includes tools for:
- matchingconstituent parse trees.
- modeling conversations (TODO)
Need help? Ask me for help onGitter
pip install langoMake sure you have Java installed for the Stanford CoreNLP to work.
Extract to any folder
Run the following command in the folder where you extracted Stanford CoreNLP
java -mx4g -cp "*" edu.stanford.nlp.pipeline.StanfordCoreNLPServerMatching is done by comparing a set rules and matching it with a parse tree. Youcan see parse trees for sentences from examples/parser_input.py.
The set of rules is recursive and can match multiple parts of the parse tree.
Rules can be broken down into smaller parts:
- Tag
- Token
- Token Tree
- Rules
A tag is a POS (part of speech) tag to match. A list of POS tags used by the Stanford Parser can be foundhere.
Format:tag = stringExample:'NP''VP''PP'A token is a string comprising of a tag and modifiers/labels for matching. We specify a match_label to match the tag to. We can specify opts for extracting the string from a tree. We can specify eq for matching the tree to a string.
Example string:The red caropts:-o Get object by removing "a", "the", etc. (Ex. red car)-r Get raw string (Ex. The red car)Format: (only tag is required)token = tag:match_label-opts=eqExample: 'VP''NP:subject-o''NP:np''VP=run''VP:action=run'A token tree is a recursive tree of tokens. The tree matches the structure of a parse tree.
Format:token_tree = ( token token_tree token_tree ... )Examples: '( NP ( DT ) ( NP:subject-o ) )''( NP )''( PP ( TO=to ) ( NP:object-o ) )'Rules are a dictionary of token trees to dictionaries of matching labels to anested set of rules.
Format:rules = {token_tree: {match_label: rules}}Example: { '( S ( NP:np ) ( VP ( VBD:action-o ) ( PP:pp ) ) )': { 'np': { '( NP:subject-o )': {} }, 'pp': { '( PP ( TO=to ) ( NP:to_object-o ) )': {}, '( PP ( IN=from ) ( NP:from_object-o ) )': {}, } },}When matching a rule to a parse tree, the token tree is first matched. Then, allmatching tags are matched to nested rules corresponding to their matching label.
All nested match labels must have a subrule match or the rules will not match.
The first rule to match is returned so the order of match is based on keyordering (use OrderedDict if order matters). Once a rule is matched, it callsthe callback function with the context as arguments.
Suppose we have the sentence "Sam ran to his house" and we wanted to match thesubject ("Sam"), the object ("his house") and the action ("ran").
Sample parse tree for "Sam ran to his house" from the Stanford Parser.
(S (NP (NNP Sam) ) (VP (VBD ran) (PP (TO to) (NP (PRP$ his) (NN house) ) ) ) )Simplified image of tree:
Matching:Parse Tree: (S (NP (NNP Sam) ) (VP (VBD ran) (PP (TO to) (NP (PRP$ his) (NN house))))Matched token tree: '( S ( NP:np ) ( VP ( VBD:action-o ) ( PP:pp ) ) )'Matched context: np: (NP (NNP Sam)) action-o: 'ran' pp: (PP (TO to) (NP (PRP$ his) (NN house)))Rule for '( S ( NP:np ) ( VP ( VBD:action-o ) ( PP:pp ) ) )':
Matching 'NP' matches the whole NP tree and converts to a word:
Matched token tree for np: '( NP:subject-o )'Matched context: subject-o: 'Sam'Matching 'PP' requires matching the nested rules:
Match token tree for pp: '( PP ( TO=to ) ( NP:to_object-o ) )'Match context: object-o: 'his house'Match token tree for pp: '( PP ( IN=from ) ( NP:from_object-o ) )'No match foundPP of the sample sentence:
Nested PP rules:
Only the first rule matches for 'PP'.
Now that we have a match for all nested rules, we can return the context:
Returned context: action: 'ran' subject: 'sam' to_object: 'his house'Full code:
fromlango.parserimportStanfordServerParserfromlango.matcherimportmatch_rulesparser=StanfordServerParser()rules= {'( S ( NP:np ) ( VP ( VBD:action-o ) ( PP:pp ) ) )': {'np': {'( NP:subject-o )': {} },'pp': {'( PP ( TO=to ) ( NP:to_object-o ) )': {},'( PP ( IN=from ) ( NP:from_object-o ) )': {} } }}deffun(subject,action,to_object=None,from_object=None):print"%s,%s,%s,%s"% (subject,action,to_object,from_object)tree=parser.parse('Sam ran to his house')match_rules(tree,rules,fun)# output should be: sam, ran, his house, Nonetree=parser.parse('Billy walked from his apartment')match_rules(tree,rules,fun)# output should be: billy, walked, None, his apartment
About
Language Lego
Topics
Resources
License
Uh oh!
There was an error while loading.Please reload this page.
Stars
Watchers
Forks
Releases
Packages0
Uh oh!
There was an error while loading.Please reload this page.
Contributors2
Uh oh!
There was an error while loading.Please reload this page.