Is it possible to use dictionaries like Longman, Oxford as sense inventory about pywsd HOT 7 OPEN

@coiby , to use a custom dictionary, you could use original_lesk() where you can input your own dictionary:

from nltk import word_tokenize
from pywsd.lesk import original_lesk

dictionary = {'reflex.a.1' : 'Physiology. noting or pertaining to an involuntary response to a stimulus, the nerve impulse from a receptor being transmitted inward to a nerve center that in turn transmits it outward to an effector.',
'reflex.a.2' : 'occurring in reaction; responsive.',
'reflex.a.3' : 'cast back; reflected, as light, color, etc.',
'reflex.a.4' : 'bent or turned back.',
'reflex.a.5' : 'designating a radio apparatus in which the same circuit or part performs two functions.',

'reflex.n.1' : 'Physiology. Also called reflex act. movement caused by a reflex response. Also called reflex action. the entire physiological process activating such movement.',
'reflex.n.2' :'any automatic, unthinking, often habitual behavior or response.',

'reflex.n.3' : 'the reflection or image of an object, as exhibited by a mirror or the like.',
'reflex.n.4' : 'a reproduction, as if in a mirror.',
'reflex.n.5' : 'a copy; adaptation.',
'reflex.n.6' : 'reflected light, color, etc.',
'reflex.n.7' : 'Historical Linguistics. an element in a language, as a sound, that has developed from a corresponding element in an earlier form of the language'}

# Tokenize your definitions from the custom dictionary.
# It's a weird step (possibly, i should change the function 
# to do this automatically)
dictionary = {k:" ".join(word_tokenize(v)) for k,v in dictionary.items()}

context = "Virtual assistants also require a conscious decision to stop doing the current task and actively seek out the virtual assistant, which is a reflex many users haven't developed."
context = " ".join(word_tokenize(context))

disambiguated = original_lesk(context, 'reflex', dictionary)
print disambiguated
print dictionary[disambiguated]

[out]:

reflex.n.7
Historical Linguistics . an element in a language , as a sound , that has developed from a corresponding element in an earlier form of the language

Note that the original lesk is flawed, hence the various version of lesk.

But do note that a dictionary has many sense per word like a synset so when you build your custom dictionary you need 1 key-value pair per sense and not per word.

For copyrights reason, I am not able to allow the software to use longman or oxford. But I could use other dictionaries that are open like wiktionary. I will have some free time after april and let me see whether I can move the software to version 1.1. I'll add wiktionary as one of the things to add.

But you're working with a dictionary company that allows me to use the dictionary freely for pywsd, I'll surely add it as an alternative resource to wordnet ;P

from pywsd.

Ah yes, that synset would have been the only adjective indexed with "reflex". But do note that in NLTK using WordNet v3.0, it's POS is a satellite adjective instead of a.

>>> from pywsd.lesk import simple_lesk
>>> from nltk.corpus import wordnet as wn
>>> simple_lesk("Virtual assistants also require a conscious decision to stop doing the current task and actively seek out the virtual assistant, which is a reflex many users haven't developed.", 'reflex', pos='s')
Synset('automatic.s.03')
>>> wn.synset('automatic.s.03').definition()
u'without volition or conscious control'
>>> wn.synset('automatic.s.03').offset()
2522669

from pywsd.

@alvations Thank you for the comprehensive instructions.

My goal is to accumulate vocabulary words by memorizing new words using spaced repetition learning technique. Given a word and the context, I want to find definition which has the same sense together with the example sentence from the dictionary like Longman automatically. So building flashcards for memorization can also be done automatically. That's why I'm interested in this project.

I try to minimize the efforts of manually editing flashcards. It'll be boring. So WSD precision is one of my major concerns.

Unfortunately, I'm not from a dictionary company. But if I buy the CD-ROM and extract the data for personal use, it will not infringe copyright, right?

from pywsd.

We do have in the Princeton Wordnet a sense for "reflex" that is similar of the sense from Longman:

http://wnpt.brlcloud.com/wn/synset?id=02522669-a
http://compling.hss.ntu.edu.sg/omw/cgi-bin/wn-gridx.cgi?synset=02522669-a

Same for impetus:

http://wnpt.brlcloud.com/wn/synset?id=11447851-n

from pywsd.

@arademaker Thank you for providing info about Wordnet.

For another sense of reflex,

without volition or conscious control

it's considered to be a satellite adjective by Wordnet as pointed out by @alvations. But Longman or Oxford consider both two senses as nouns. Btw, I'm curious why Wordnet introduce satellite adjective.

The reason I think Wordnet doesn't provide proper definition for the first sense is that reflex can be unlearned (unconditioned) or learned (conditioned)

noun
an automatic instinctive unlearned reaction to a stimulus (Freq. 5)
• Syn:
reflex response, reflex action, instinctive reflex, innate reflex, inborn reflex, unconditioned reflex, physiological reaction
• Hypernyms: reaction, response

But I think pyWSD make the right choice.

For impetus, you mean "a force that moves something along", right? pyWSD fail to choose the correct one this time. But I think the metaphorical sense ( an influence that makes something happen or makes it happen more quickly) is better explained by Longman/Oxford.

from pywsd.

@coiby IMHO, satellite adjective is a historical issue. This QnA describes it partially: http://stackoverflow.com/questions/18817396/what-part-of-speech-does-s-stand-for-in-nltk-synsets I usually normalize all my s -> a when I use word sense related features for NLP.

I'm not sure about the license of your dictionary that you've purchased so I can't really commment on that.

As for pywsd accuracy, I really need to sit down and evaluate them but note that when using lesk related functions, usually it boils down to what words appear in both the context and the definition (aka "signatures" in the original lesk paper).

from pywsd.

Actually, this is a decision for encoding PWN in text files. See https://wordnet.princeton.edu/documentation/wndb5wn and https://wordnet.princeton.edu/documentation/lexnames5wn. The synset type is not the syntactic category or part-of-speech as we call it nowadays. This is confusing because the same values are used: a, v, n, and r (but nor s). Many adjectives in PWN are organized into clusters and an a synset can be the HEAD of a cluster and s ones, are satellites of a cluster.

from pywsd.