pylats 0.31

Text preprocessing for downstream linguistic analyses

Python Linguistic Analysis Tools (pylats)

Pylats is designed to perform text pre-processing for further linguistic analyses (e.g., measuring lexical diversity and/or lexical sophistication). Currently, advanced features such as lemmatization and POS tagging are available for English, with other languages to follow (Spanish models will be added in the near future, we have plans to release other models as well). However, pylats CAN currently be used with other languages using basic features (see examples below).

Pylats currently uses spacy as a starting point for many of its advanced features if working with English. Pylats was tested using spacy version 3.2 and be default uses the "en_core_web_sm" model. To install spacy and a language model, see the spacy installation instructions.

Installation

To install pylats, you can use pip:

pip install pylats

Getting Started

Import

from pylats import lats

Using pylats

pylats is designed to be the first step in conducting linguistic analyses using related analysis tools (such as lexical-diversity).

pylats uses the Normalize class to take a raw text string and format it:

teststr = "I love pepperoni pizza."
normed = lats.Normalize(teststr) #processed text string
print(normed.toks)

#output:
['i', 'love', 'pepperoni', 'pizza']

Paragraphs and sentences

The .toks method will provide a flat list of the tokens in a text. However, it can often be useful to conduct analyses at the sentence and/or paragraph level. The .sents and .paras methods provide a representation of text with nested lists.

teststr = """I love pepperoni pizza. Sometimes I like to add feta and banana peppers.
This is a second paragraph. In the original string there is a newline character before this paragraph."""

normedp = lats.Normalize(para_sample)

tokens

print(normedp.toks)

['i', 'love', 'pepperoni', 'pizza', 'sometimes', 'i', 'like', 'to', 'add', 'feta', 'and', 'banana', 'peppers', 'this', 'is', 'a', 'second', 'paragraph', 'in', 'the', 'original', 'string', 'there', 'is', 'a', 'newline', 'character', 'before', 'this', 'paragraph']

sentences

for x in normedp.sents:
	print(x) #print tokens in each sentence

['i', 'love', 'pepperoni', 'pizza']
['sometimes', 'i', 'like', 'to', 'add', 'feta', 'and', 'banana', 'peppers']
['this', 'is', 'a', 'second', 'paragraph']
['in', 'the', 'original', 'string', 'there', 'is', 'a', 'newline', 'character', 'before', 'this', 'paragraph']

paragraphs

for x in normed.paras:
	print(x) #print sentences each paragraph

[['i', 'love', 'pepperoni', 'pizza'], ['sometimes', 'i', 'like', 'to', 'add', 'feta', 'and', 'banana', 'peppers']]
[['this', 'is', 'a', 'second', 'paragraph'], ['in', 'the', 'original', 'string', 'there', 'is', 'a', 'newline', 'character', 'before', 'this', 'paragraph']]

Changing parameters

By default, Normalize simply removes punctuation and converts words in the text to lower case. However, a wide range of customizations can be made by adjusting the parameters class.

For example, it may be useful to exclude particular words for some analyses. In studies of lexical diversity, for instance, we probably don't want to include misspelled words (misspelled words would positively contribute to diversity scores, but probably shouldn't). Pylats includes a default list of "real" words drawn from a large corpus of English language which can be used to filter out misspelled ones. Words can also be added to a list of items to remove OR can be added to a list that overrides other lists.

Below, we create a copy of parameters and then make some changes:

new_params = lats.parameters() #create a copy of the parameters class
new_params.attested = True #set the attested attribute to True

Output with default settings:

#with default settings
msp_default = lats.Normalize("This is a smaple sentence")
print(msp_default.toks)

['this', 'is', 'a', 'smaple', 'sentence']

Output with new settings:

msp_new = lats.Normalize("This is a smaple sentence", params = new_params)
print(msp_new.toks)

['this', 'is', 'a', 'sentence']

Default parameters

class parameters: #default English parameters
	lang = "en"
	model = "en_core_web_sm"
	punctuation = ['``', "''", "'", '.', ',', '?', '!', ')', '(', '%', '/', '-', '_', '-LRB-', '-RRB-', 'SYM', ':', ';', '"']
	punctse = [".","?","!"]
	abbrvs = ["mrs.","ms.","mr.","dr.","phd."]
	splitter = "\n" #for splitting paragraphs
	rwl = en_rwl
	sp = True
	sspl = "spacy"
	pos = None #other options are "pos" for Penn tags and "upos" for universal tags
	removel = ['becuase'] #typos and other words not caught by the real words list
	lemma = False
	lower = True #treat all words as lower case
	attested = False #filter output using real words list?
	spaces = [" ","  ","   ","    "] #need to add more here
	override = [] #items the system ignores that should be overridden
	posignore = []
	numbers = ["NUM"] #pos_ tag for numbers
	nonumbers = True
	connect = "__" #for connecting ngrams
	contentPOS = [] #can be added, blank for now
	contentLemIgnore = [] #can be added, blank for now

Adding part of speech information

If spacy is installed (and activated), part of speech tags can be added to each token, which can be useful in disambiguating homographic tokens (e.g., run as a verb in the sentence I like to run. versus run as a noun in the sentence I went for a run. ). This is helpful in a number of applications, including calculating indices lexical diversity.

pos_params = lats.parameters() 
pos_params.pos = "upos" #for large-grained universal parts of speech
run_sample = lats.Normalize("I like to run. I went for a run.", params = pos_params)
for x in run_sample.sents:
	print(x)

['i_PRON', 'like_VERB', 'to_PART', 'run_VERB']
['i_PRON', 'went_VERB', 'for_ADP', 'a_DET', 'run_NOUN']

Changing the spacy language model

To change the spacy language model that is used by pylats, first make sure that the desired model has been downloaded from spacy. The preferred method to load models is to create a parameter class that includes the appropriate model (for easier replicability), then use the following command:

lats.nlp = lats.load_model(lats.myparameters.model)

Alternatively, a model can be loaded directly:

#loading the "en_core_web_trf" model
lats.nlp = lats.load_model("en_core_web_trf")

Using pylats with languages other than English

Pylats currently has advanced features available for English and Spanish texts and basic features for other languages. As the tool expands, advanced feature support will be added for other languages (let us know what languages you would like to see supported!).

To process texts with basic features, simply change parameters.sp to True. The processor will treat text between whitespace as a token. Accordingly, some pre-processing may be necessary.

Example 1 (Spanish), advanced features: There are currently two pre-made parameter classes for Spanish. One uses a faster (but slightly less accurate) tagging and parsing model (parameters_es, which uses the "es_core_news_sm" spacy model). The second uses a slower (but more accurate) tagging and parsing model (parameters_es_trf, which uses the "es_dep_news_trf" model). To process Spanish texts, first be sure to download the appropriate model from spacy.

Then, load the appropriate Spanish model and be sure to use the appropriate Spanish parameters class.

lats.nlp = lats.load_model(lats.parameters_es.model) #load spacy model for Spanish
span_sample = lats.Normalize("Me gustaría aqua con gas.",lats.parameters_es) #process text
print(span_sample.toks)

['me', 'gustaría', 'aqua', 'con', 'gas']

If we want to create a lemmatized representation of a text and add part of speech tags, we can create a slightly altered version of our parameters class:

parameters_es_lemmas = lats.parameters_es()
parameters_es_lemmas.lemma = True #set lemma to True
parameters_es_lemmas.pos = "pos" #set pos tags to universal pos + fine-grained verb POS tags (universal pos + mood/tense)


span_sample_lemma = lats.Normalize("Me gustaría aqua con gas.",parameters_es_lemmas) #process text
print(span_sample_lemma.toks)

['yo_PRON', 'gustar_VERB_Cnd', 'aqua_ADJ', 'con_ADP', 'gas_NOUN']

Example 2 (Korean), basic features:

whtsp_params = lats.parameters() #copy parameters
whtsp_params.sp = False #turn off spacy processing
kor_sample = lats.Normalize("피자 좀 주세요",whtsp_params)
print(kor_sample.toks)

['피자', '좀', '주세요']