Get started with Natural Language Processing NLP

NLP, Natural Language Processing is an interdisciplinary scientific field that deals with the interaction between computers and the human natural language.

I teach the Natural Language Processing track at AI Saturdays Abuja. I hope this blog post and NLP series would serve as a refresher for my students and other learners! 💖

Some NLP use case scenario includes;
Speech recognition, Natural language understanding (NLU), Natural language generation, Sentiment analysis, Conversational systems, Machine translation (MT), Information Retrieval systems e.t.c

NLP Wins and Bottlenecks

Let’s start by looking at successes and bottlenecks in NLP.

Wins

Modern computer techniques with natural language are being used today to improve lives and increase efficiency at workplaces.

Email providers now have NLP powered apps that can scan emails to correct misspellings, extract information and compose emails with better grammar for us.

A good example is the Google smart compose which offers suggestions as you compose an email.

Another popular example is Grammarly, a writing assistant that helps you compose writings with better grammar and also checks for spelling errors.

Social media platforms with our permission collect bits of our daily lives and feed us tailored advertisements.

Speech powered devices like the Amazon Echo and Google Home can understand and interpret commands in our human natural language.

NLP is super interesting! isn’t it? 🚀 😍

Some of the bottlenecks

Computers are built to follow certain rules of logic flow, but humans are not and so is our communication.

Some of the tasks that are easy for humans, is extremely difficult for the computer.

How can a computer tell the difference between sarcasm and the original intent of a sentence?

Human natural language and communication are ambiguous for the computer.

Our goal is to apply an algorithm to a dataset of natural language in order to train it to understand, interpret and eventually perform a useful task.

For us to achieve this, we need to;

1. Assess the grammatical rule of our language (Syntactic)
2. Understand and interpret its meanings (Semantics)

Text Normalization

Before we can start building anything useful, we have to preprocess/prepare our datasets.

The goal of stemming and lemmatization is to reduce inflected word forms to its base form.

Inflection: letters can be added to adjectives, verbs, nouns and other parts of speech to express meaning and for other grammatical reasons.

For example;

Base words: Bus, City, Criterion, Bureaucracy
Inflection: Buses, Cities, Criteria, Bureaucratization

Lemmatization and Stemming use different approaches to reduce inflection.

Stemming

Stemming cut off letters from a word in order to reduce it to a stem.

PorterStemmer and LancasterStemmer are two popular algorithms for performing stemming on the English language

How to Stem words using NLTK

#import the nltk package
import nltk
#nltk downloader
nltk.download('punkt')

let’s try out both PorterStemmer (Porter’s algorithm 1980) and LancasterStemmer algorithm

from nltk.stem import PorterStemmer
from nltk.stem import LancasterStemmer

#create an object of class PorterStemmer and #LancasterStemmer

porter_stem = PorterStemmer()
lancaster_stem = LancasterStemmer()

#find the stem of these words 

print("Using Porter's Algorithm")
print(porter_stem.stem("cats"))
print(porter_stem.stem("goats"))
print(porter_stem.stem("running"))
print(porter_stem.stem("ran"))

print("Using Lancaster's Algorithm")

print(lancaster_stem.stem("Zebras"))
print(lancaster_stem.stem("democracy"))
print(lancaster_stem.stem("bureaucracy"))
print(lancaster_stem.stem("perfection"))

Notice the base words produced by stemming, after cutting off ‘acy’ in democracy and bureaucracy. ‘democr’ and ‘bureaucr’ is not a meaningful English word.

Lemmatization

Unlike stemming that only cut off letters, lemmatization takes a step further; it considers the part of speech and possibly the meaning of the word in order to reduce it to its correct base form (lemma).

Lemmatization most times produce an actual word

Lemmatization searches a corpus to match inflected words and base words.

A good corpus example is the WordNet corpus

# import WordNet Lemmatizer 
from nltk.stem import WordNetLemmatizer 

#create an object of class WordNetLemmatizer
lemmatizer = WordNetLemmatizer() 

# a denotes adjective, a word that modifies a noun
# v denotes verb
# n denotes noun
  
print("lemma  of 'quickly' is ", lemmatizer.lemmatize("quickly", pos="a")) 

print("lemma of 'asked' is ", lemmatizer.lemmatize("asked", pos="v"))   

print("lemma of 'cables' is ", lemmatizer.lemmatize("cables", pos ="n")) 

Tokenization

Tokenization describes splitting/breaking down a text document into smaller units.

A token is a unit of text in the document.

For example;

“This sentence can be broken down into a small unit of words”

“This” “sentence” “can” “be” “broken” “down” “into” “a” “small” “unit” “of” “words”

#import word tokinize 
from nltk.tokenize import word_tokenize

text = "The Lazy dog JUMPED over the fence"
tokens = word_tokenize(text)
print(tokens)

we can tokenize into sentences too

#import sentence tokenize 
from nltk.tokenize import sent_tokenize

text = "The Lazy dog JUMPED over the fence. The cat was chasing it. Lagos traffic is on another level"
tokens = sent_tokenize(text)
print(tokens)

More examples

Before we can find the root of words in a sentence, we have to use tokenization to split the sentence, then find the root of each individual tokens.

text = '''Beautiful is better than ugly.
Explicit is better than implicit.
Simple is better than complex. 
Complex is better than complicated.
Flat is better than nested.
Sparse is better than dense. 
Readability counts. 
Special cases aren't special enough to break the rules. Although practicality beats purity. 
Errors should never pass silently. 
Unless explicitly silenced.'''

Tokenize the sentence into words

from nltk.tokenize import sent_tokenize, word_tokenize

def tokenize_Sentence(text):
    token_words=word_tokenize(text)
    return token_words

print(tokenize_Sentence(text))

Perform Porter’s algorithm

def porter_algorithm(tokenize_Sentence):
    text = []
    for word in tokenize_Sentence:
        text.append (porter_stem.stem (word))
        text.append(" ")
        stem_words = "".join(text)
    return stem_words
    

print (porter_algorithm(tokenize_Sentence(text)))

Compare results with Lancaster algorithm

def lancaster_algorithm(tokenize_Sentence):
    text = []
    for word in tokenize_Sentence:
        text.append (lancaster_stem.stem (word))
        text.append(" ")
        stem_words = "".join(text)
    return stem_words
    

print (lancaster_algorithm(tokenize_Sentence(text)))

Let’s lemmatize the sentence

def WordNet_Lemmatizer(tokenize_Sentence):
    text = []
    for word in tokenize_Sentence:
        text.append (lemmatizer.lemmatize(word, pos = "a"))
        text.append(" ")
        lemma_words = "".join(text)
    return lemma_words
    

print (WordNet_Lemmatizer(tokenize_Sentence(text)))

Try this: Example on a File

Using shakespeare-hamlet.txt in the Gutenberg corpora

We would look at more concept and topics in other blog posts

Morphological segmentation, Word segmentation, Parsing, Parts of speech tagging, breaking sentence, Named entity recognition (NER), Natural language generation, Word sense disambiguation, Deep Learning (Recurrent Neural Networks)

Glad you made it to the end 🔥

Download the Jupyter notebook from Github

Please share your comments and questions on the comment box

Looking for a hands-on coding Bootcamp to learn new skills in Cloud Computing and Machine learning, visit TreapAI