Hi, generally speaking a rule based approach is often simpler to start but harder to get a good accuracy and also annoying to maintain in the long run, so I'd say given you have a corpus that is enough similar to the unseen text that you want the tense to be extracted from then in my opinion a statistical approach should be better.
That said I think it also depends on the structure of the language you want to extract the tense from, so that if there're clear rules defining that (e.g. the sentence structure is different in simple and past tense) then you may choose to leverage such rules, given the exceptions are really just a few (however I think this situation is not very likely). My 2 cents, Tommaso 2014-03-22 18:44 GMT+01:00 swapnil marathe <[email protected]>: > Hi all, > I am a engineering graduate working on an NLP related project.I am new to > Natural language processing and finally understood something about it. My > aim of project is to design a lab for various activities for each grammar > topics, examples here : > > > *Grammar topics examples :* > subject verb agreement > tense > verb forms > reported speech etc > > *Activities for tenses topic examples.* > Fill in the blank with correct tense : I am ______(to play) football > Change the tense of a sentence > Recognize the tense of the sentence > etc.... > > > I will elaborate 1 activity :*Recognize tense of a sentence .* > *Objective of the activity :* > We will give some sentences to student and they will identify tense of the > sentence in this activity. > > > *Our current procedure* > We will come to know tense of sentence using NLP at backend. We have a > corpus of English textbooks that we willl use for similar grammar topics. > > For now I know there are two approaches to this.Rule based NLP and > Statistical NLP. > > So,I can write down rules to identify data that is related to specific > activity or use statistical nlp . > > what should I choose? > > I know for now there exist various NLP api like Stanford NLP,Opennlp etc. > They have models for POS tagging , chunking etc.. > > *So do I need to make model for each grammar topic if i use statistical > approach?* > > I wonder *how can i make a model* for tense or any other topic and get the > data which i require for actiivites. > > Does that model integrate with other NLP like Stanford etc.... > > > *Is there any other approach?Please tell me if I am going wrong somewhere.* >
