GLRParser

A GLR Parser for Natural Language Processing and Translation

GLRParser is not just a parser. It's

• Natural Language Parser which handles ambiguous grammars
• Unification Engine which handles unification of features
• Translation Engine for Syntax-Based Translation of Natural Languages

For a quick start, you can use following commands to install and run an interactive demo for English to Turkish Translation:

pip install GLRParser
python -m GLRParser.main

In interactive demo, you can enter an English sentence to get Turkish translation(s):

Grammar load time: 806,295 mics
Number of rules: 24915
Number of states: 28047
Number of symbols: 5738
Number of NonTerm symbols: 159
Enter Sent> who do you think you are
  kim olduğunuzu düşünüyorsunuz
Enter Sent> as long as she is happy i will be happy
  mutlu olduğu sürece mutlu olacağım
Enter Sent> his sudden departure had demonstrated how unreliable he was
  ani ayrılışı ne kadar güvenilmez olduğunu göstermişti
Enter Sent> attacks were threatening to destabilize the government
  saldırılar yönetimi istikrarsızlaştırmakla tehdit ediyordu
Enter Sent> if i had come early she wouldn't have missed her bus
  erken gelmiş olsaydım otobüsünü kaçırmış olmazdı
  erken gelmiş olsaydım otobüsünü özlemiş olmazdı

You can also visit following url to try interactive translations: https://mdolgun.pythonanywhere.com/

For a list of sample translations check the file: https://github.com/mdolgun/GLRParser/blob/master/GLRParser/grm/main.out.txt

For detailed information about the features and the grammar syntax, you can refer to wiki page: https://github.com/mdolgun/GLRParser/wiki

Sample code for parsing and translation should be like:

from GLRParser import Parser, ParseError, GrammarError, Tree

try:
    parser = Parser() # initialize parser object

    parser.parse_grammar("GLRParser\grm\simple_trans.grm") # load grammar from a file
    sent = "i saw the man in the house with the telescope" # sentence to parse

    parser.compile() # constructs parsing tables
    parser.parse(sent) # parse the sentence

    tree = parser.make_tree() # generates parse forest
    ttree = parser.trans_tree(tree) # translate the parse forest

    print(ttree.pformatr()) # pretty-print the translated parse forest

    for trans in ttree.enum(): # enumerate and print all alternative translations in the parse forest
        print(trans.replace(" -","")) # concat suffixes
except GrammarError as ge:
    print(ge)
except ParseError as pe:
    print(pe))

Simple grammar for English -> Turkish translation (see simple_trans.grm)

S -> NP VP : NP VP
S -> S in NP : NP -de S 
S -> S with NP : NP -la S 
NP -> i : 
NP -> the man : adam
NP -> the telescope : teleskop
NP -> the house : ev
NP -> NP-1 in NP-2 : NP-2 -deki NP-1
NP -> NP-1 with NP-2 : NP-2 -lu NP-1
VP -> saw NP : NP -ı gördüm  

Given the above grammar and input string:

i saw the man in the house with the telescope

It produces a parse forest, and 5 alternative translations (of which two are identical):

1. teleskopla evde adamı gördüm
2. teleskopla evdeki adamı gördüm
3. teleskoplu evde adamı gördüm
4. teleskoplu evdeki adamı gördüm
5. teleskoplu evdeki adamı gördüm

The semantic interpretations are:

1. saw(in the house) saw(with the telescope)
2. man(in the house) saw(with the telescope) 
3. saw(in the house) house(with the telescope)
4. man(in the house) man(with the telescope)
5. man(in the house) house(with the telescope)