Learn about part of speech tagging and how it begins the process of translating human language into a form that machines can understand.
Part of speech (POS) tagging is a tool in natural language processing (NLP) that “tags” words in a sentence with its part of speech based on the context around the word. The English language includes a broad range of different parts of speech, including prepositions, nouns, verbs, adverbs, adjectives, conjunctions, pronouns, interjections, and articles. Part of speech tagging becomes necessary because words in English can have multiple parts of speech since words are not static categories. For example, the word eye can act as both a verb and a noun, such as “I’ve got my eye on you” or “Take a look at my eye.” To combat manually tagging sentences, NLP researchers developed methods to automate POS tagging.
Discover the part of speech tagging methods, how they work, the importance of POS tagging, who uses it, the challenges it faces, and how you can start building POS tagging skills.
Many different methods have been developed for speech tagging. You can essentially break POS tagging methods into three general categories:
Rule-based tagging
Stochastic tagging
Machine learning (ML) and deep learning (DL) tagging
Explore each framework and the methods they offer in more detail to understand their uses.
Rule-based tagging uses a specific set of rules that you create. These rules often follow those of linguistics, using sentences' lexical, morphological, and syntactical features. A linguist can create these rules or use machine learning on a classified data set. This type of POS tagging is time-consuming and can lead to errors in the tagging process.
Stochastic or probabilistic tagging is a popular approach to part of speech tagging that uses supervised learning and trains on sample data sets. Two stochastic tagging approaches include:
Hidden Markov Model (HMM): The HMM is one of the most common approaches to POS tagging. This generative sequence model tries to classify each word in a sentence, in this case, labeling it as a part of speech. In an HMM, the hidden part of the model (the part of speech) only depends on the known part of the model (the word) to make its prediction.
Maximum Entropy Markov Model (MEMM): The MEMM uses logistic regression in a discriminative sequence model instead of the HMM's generative sequence model. This allows the model to use the class of the previous word to help it define the present word it’s analyzing.
Machine learning and deep learning in NLP use neural networks to process words by mimicking the function of neurons in the human brain. Some effective and popular kinds of neural networks for POS tagging include:
Recurrent neural network (RNN): RNNs take an input, work on it in a hidden layer, and continually update their outputs using the previous findings.
Bidirectional long short-term memory (BLSTM): A popular deep learning method of tagging, BLSTM uses two hidden layers, one that feeds forward and another that processes backward. These have the most accurate outputs out of all the RNNs available.
Part of speech tagging disambiguates words, classifying each part of speech with respect to its sentence placement and context. It is a fundamental process in creating any NLP pipeline and, thus fundamental to creating a proper NLP parser. NLP is a key technology behind speech recognition and synthesis, which helps many people in their daily lives. Speech translation, AI-powered chatbots, and document summaries all use natural language processing.
Many industries use part of speech tagging as part of using or creating an NLP program. Some NLP applications include:
Finance: NLP mines financial documents, statements, reports, and the news to assist in strong, data-driven decision-making that impacts earnings.
Health care: In health care, NLP searches health records and research to help professionals make informed decisions about patient care, prevention, and diagnosis.
Legal: With the amount of paperwork and documents in a legal proceeding, NLP can help organize, summarize, and review to ensure lawyers receive all relevant information that could help their case.
Below are some careers that research, create, and engineer NLP systems.
Median annual base pay (Glassdoor): $123,332 [1]
Requirements: This role typically requires a bachelor’s degree in computer science or a related field, with some employers requiring graduate experience in AI.
NLP engineers design NLP systems by creating the language training datasets that they use to train machine learning algorithms. In this position, you will process the results and tweak models accordingly. You’ll analyze human speech patterns and turn the words into mathematical models that machines can understand and implement into AI-generated speech.
Median annual base pay (Glassdoor): $122,930 [2]
Requirements: Many machine learning engineers earn a degree in data science or computer science, sometimes earning an advanced degree to progress in the field.
Machine learning engineers create programs and algorithms that analyze and process data to find inherent patterns and use insights to make predictions. As a machine learning engineer, you will design these programs by training them to identify patterns in large swathes of data. Over time, they make predictions with more and more accuracy.
Median annual base pay (Glassdoor): $112,149 [3]
Requirements: Many AI engineers need a master’s degree at a minimum to gain the necessary skills in the field.
AI engineers use machine learning and deep learning to build intelligent AI models. In this job, you’ll combine extensive programming, software development, and data science knowledge to produce effective models for various industries. You will also create and manage the infrastructure the rest of the data science team will follow to create and maintain AI models.
Part of speech tagging is a fundamental process of NLP but is a field that comes with many challenges and research opportunities to improve the accuracy of NLP systems. Some of these challenges include:
Finding balanced corpus data sets to train more accurate POS taggers that can learn more patterns in ML and DL
Having corpus data sets released publicly to increase the amount of resources for research
Optimizing computing resources to require less power-intensive components to do part of speech tagging work
Other challenges include the need for a sufficient corpus in many different languages, especially ones that are morphologically rich and have large vocabulary like Hungarian, Czech, and Romanian.
To build skills in NLP and POS tagging, you need to develop your linguistic skills and gain proficiency in programming languages like C++, Python, and R. These give you the fundamental tools to understand language and translate it into a form that a machine can understand.
If you are already familiar with Python, consider learning how NLP works by using the Natural Language Toolkit (NLTK), an open-source language processing library. It gives you access to over 50 corpora and tools for tokenization, stemming, POS tagging, and semantic reasoning. It’s an effective computation linguistics tool with a large community and sufficient documentation, including an online version of the NLTK book for Python 3.
Part of speech tagging is fundamental to creating effective NLP programs. To build skills in NLP, try the Natural Language Processing Specialization from DeepLearning.AI on Coursera to build in-demand skills in HMMs, tagging, and recurrent neural networks. If you want to learn linguistics to build your understanding of language structure, try the Miracles of Human Language: An Introduction to Linguistics course from Universiteit Leiden, also on Coursera.
Glassdoor. “How much does a NLP Engineer make? https://www.glassdoor.com/Salaries/nlp-engineer-salary-SRCH_KO0,12.htm.” Accessed February 3, 2025.
Glassdoor. “How much does a Machine Learning Engineer make? https://www.glassdoor.com/Salaries/united-states-machine-learning-engineer-salary-SRCH_IL.0,13_IN1_KO14,39.htm.” Accessed February 3, 2025.
Glassdoor. “How much does an Artificial Intelligence Engineer make? https://www.glassdoor.com/Salaries/united-states-artificial-intelligence-engineer-salary-SRCH_IL.0,13_IN1_KO14,46.htm.” Accessed February 3, 2025.
Editorial Team
Coursera’s editorial team is comprised of highly experienced professional editors, writers, and fact...
This content has been made available for informational purposes only. Learners are advised to conduct additional research to ensure that courses and other credentials pursued meet their personal, professional, and financial goals.