Spacy Pipeline?

Question

So lately I've been playing around with a WikiDump. I preprocessed it and trained it on Word2Vec + Gensim

Does anyone know if there is only one script within Spacy that would generate tokenization, sentence recognition, part of speech tagging, lemmatization, dependency parsing, and named entity recognition all at once

I have not been able to find clear documentation Thank you

CentAu CentAu · Accepted Answer · 2016-08-20T20:38:37

Spacy gives you all of that with just using en_nlp = spacy.load('en'); doc=en_nlp(sentence). The documentation gives you details about how to access each of the elements.

An example is given below:

In [1]: import spacy
   ...: en_nlp = spacy.load('en')

In [2]: en_doc = en_nlp(u'Hello, world. Here are two sentences.')

Sentences can be obtained by using doc.sents:

In [4]: list(en_doc.sents)
Out[4]: [Hello, world., Here are two sentences.]

Noun chunks are given by doc.noun_chunks:

In [6]: list(en_doc.noun_chunks)
Out[6]: [two sentences]

Named entity is given by doc.ents:

In [11]: [(ent, ent.label_) for ent in en_doc.ents]
Out[11]: [(two, u'CARDINAL')]

Tokenization: You can iterate over the doc to get tokens. token.orth_ gives str of the token.

In [12]: [tok.orth_ for tok in en_doc]
Out[12]: [u'Hello', u',', u'world', u'.', u'Here', u'are', u'two', u'sentences', u'.']

POS is given by token.tag_:

In [13]: [tok.tag_ for tok in en_doc]
Out[13]: [u'UH', u',', u'NN', u'.', u'RB', u'VBP', u'CD', u'NNS', u'.']

Lemmatization:

In [15]: [tok.lemma_ for tok in en_doc]
Out[15]: [u'hello', u',', u'world', u'.', u'here', u'be', u'two', u'sentence', u'.']

Dependency parsing. You can traverse the parse tree by using token.dep_ token.rights or token.lefts. You can write a function to print dependencies:

In [19]: for token in en_doc:
    ...:     print(token.orth_, token.dep_, token.head.orth_, [t.orth_ for t in token.lefts], [t.orth_ for t in token.rights])
    ...:     
(u'Hello', u'ROOT', u'Hello', [], [u',', u'world', u'.'])
(u',', u'punct', u'Hello', [], [])
(u'world', u'npadvmod', u'Hello', [], [])
...

For more details please consult the spacy documentation.

Spacy Pipeline?

3 Answers