GraDiAn 0.0.0.1

A grammatical distribution analyser for NLP datasets.

GraDiAn

The Grammatical Distribution Analyser (GraDiAn) is used for analysing grammatical distributions; particularly the distributions of popular NLP datasets.

At the moment, GraDiAn does this by providing two abstract data types: the Syntactic Dependency Counter and the SentTree.

SentTree

SentTree represents a given sentence in a tree structure. Importantly, the SentTree can be used to analyse the parse-tree with regards to different properties of the text including part-of-speech tags, syntactic dependencies and (with the help of spaCyTextBlob) sentiment.

## Syntactic Dependency Counter (SDC) An SDC does what it says on the tin. Inheriting from python's collections.Counter class, it maintains a count of syntactic dependency labels.

Usage

Syntactic Dependency Counter

Syntactic Dependency Counter from text:

>>> from gradian import SDC
>>> sdc = SDC.from_string('This is a test sentence!')
>>> sdc
SDC({'nsubj': 1, 'ROOT': 1, 'det': 1, 'compound': 1, 'attr': 1, 'punct': 1})

Or from a series of texts:

>>> from gradian import SDC
>>> sdc = SDC.from_string_arr(['This is a test sentence!', 'This is another sentence',
                               'How about another?'])
>>> sdc
SDC({'ROOT': 3, 'nsubj': 2, 'det': 2, 'attr': 2, 'punct': 2, 'compound': 1, 'advmod': 1, 'pobj': 1}

SentTree

SentTree from text:

>>> from gradian import SentTree
>>> sent_trees = SentTree.from_string('This is a test sentence! But this is another!')
>>> # Sent_Tree.from_string produces a list of trees; one for each sentence
>>> sent_trees[0].attr_tree('pos')  # Get the Tree with respect to the sentence's POS-Tags
Tree('AUX', ['DET', Tree('NOUN', ['DET', 'NOUN']), 'PUNCT'])

attr_tree can be used with any attribute of the tree including syntactic dependencies, POS-tags and (if spaCyTextBlob is enabled) sentiment.

>>> sent_trees[0].attr_tree('dependency')
Tree('ROOT', ['nsubj', Tree('attr', ['det', 'compound']), 'punct'])

The function can be called with token=True to see the attributes alongside the relevant tokens:

>>> # token is a positional argument so does not need to be explicitly provided by keyword
>>> sent_trees[0].attr_tree('pos', token=True)  
Tree('is:  AUX', ['This: DET', Tree('sentence:  NOUN', ['a: DET', 'test: NOUN']), '!: PUNCT'])

SentTrees also come with the ability to create multi-attribute trees.

>>> sent_trees[0].multi_attr_tree(['pos', 'dependency'], True)
Tree('is:AUX:ROOT', ['This:DET:nsubj', Tree('sentence:NOUN:attr', ['a:DET:det', 'test:NOUN:compound']), '!:PUNCT:punct'])