Comparing Python’s Stemming with Lemmatization

Python’s lemmatization and stemming features are utilised to reduce text to its most basic elements. These approaches are utilised by conversational interfaces and search engines to comprehend users’ intentions. Both lemmatization and stemming in Python take into consideration the context of the search query, whereas stemming concentrates on the root of the word or the search question itself.

When searching for information, it is beneficial to not only obtain results that are pertinent to the exact search query, but also to related variations of that keyword. This ensures that users are provided with a comprehensive list of results that is relevant to their query.

In Natural Language Processing (NLP), normalisation techniques such as lemmatization and stemming are utilised to guarantee that texts are consistent. These two methods both operate by breaking down words into their essential components.

In this tutorial, we will provide code snippets and examples to demonstrate how to implement stemming and lemmatization. Utilising the Natural Language Toolkit (NLTK) library in Python, we will gain a more comprehensive understanding of these two approaches.

Python’s Lemmatization Function

Lemmatization is the process of converting a term or search query to its standard dictionary form, or lemma. This process looks at only the portion of the word that can be broken down into its constituent parts. By doing so, we are able to identify the underlying lemma, or the root form, of the term or query. This method is especially useful for accurately and efficiently searching through large amounts of textual data.

Now, we’ll take a look at some of lemmatization’s applications.

  1. Indexing systems

    Lemmatization is a process which is widely utilised by chatbots and search engines to comprehend the meaning and intent of user inquiries. Several search engines are the main adopters of this technique.
  2. Condensed cataloguing

    To efficiently store the information, index values are used.
  3. Biomedicine

    The field of biomedicine benefits from lemmatization analysis. It improves the effectiveness of data retrieval and associated processes.

Rooted in the Python Language

Stemming is a process that involves using rules to identify the root of a word and thereby find its stem. This method simplifies the process by removing word endings indiscriminately. Utilising Python stemming further enhances this process by normalising sentence length and condensing search results, thus improving the readability of the text.


When a word is excessively stemmed, its meaning becomes distorted and the resulting stem word is nonsensical. As a result, words with the same root can take on multiple meanings. For example, in this particular context, words can be shortened to their root form, such as “univers”.

  • University
  • Universal
  • Universe

Each of these three terms has a same etymological root, yet in this context their meanings couldn’t be more distinct.


In this example, two distinct concepts are conveyed through the utilisation of words which share the same root. The issue of under-stemming emerges when a text incorporates terms that are synonymous with each other.

The word “alumnus” serves as a good illustration. For the various options, consider:

  • Alumni
  • Alumna
  • Alumnae

This word’s Latin morphology means that these antonyms are not interchangeable.

Differentiating stems and lemmas

Stemming and lemmatization are two techniques that can be used to reduce the complexity of words by reducing them to a single, simplified form. Although the basic concept of both processes is the same, they differ significantly in terms of the process and results they produce. Stemming is a process of reducing words to their root form by removing affixes such as prefixes and suffixes. On the other hand, lemmatization is a process of reducing words to their dictionary form by taking into account their morphological structure. As a result, lemmatization yields more accurate results than stemming.

The stemming method involves removing the initial or final letters of the original word. It remembers which prefixes are the most often used.

Conversely, the process of lemmatization in Python is focused on examining the etymology of a word to discover its root lemma, which is the source of all its inflected forms.

Lemmatization and stemming are both processes that get a word ready for machine translation.

When it comes to classifying data, vectorization is a useful tool to have in your arsenal. By eliminating the requirement of using for loops, vectorization allows for the execution of array operations. This means that the data that you are currently dealing with can be transformed into something entirely new with the help of a vectorizer.

By employing stemming or lemmatization techniques, it is possible to reduce the set of words used in a given corpus. This simplification of the lexicon leads to less ambiguity when training a model, and ultimately yields better performance in the model’s outputs. As such, using these procedures can be a great way to improve the efficacy of the model.

Below, we will provide a comprehensive overview of the Python Natural Language Toolkit (NLTK) and how it can be utilised to carry out stemming and lemmatization in your own projects. We will also explain the advantages and disadvantages of each method and provide instructions on how to use them in your project.

Varieties of Roots

Previously, it has been established that stemming is a rule-based text normalisation approach which removes the prefix and suffix of a word to get to its root form. Compared to lemmatization, which takes the context of the words into account, stemming is a faster process. However, due to its aggressive nature, it may lead to incorrect results in a dataset.

Algorithm of Porter Stemming

The Porter Stemmer, also referred to as the Porter Stemming Algorithm, is a technique for creating word stems through the process of suffix-stemming. To better understand how the Porter Stemming Algorithm works, we can look at the following code sample which makes use of the Natural Language Toolkit (nltk) library to create a stemmer object.

Stemmer for a snowball

The Snowball stemmer strategy, also known as the Porter2 stemmer, is an improved version of the Porter stemmer algorithm, which resolves several of the shortcomings of the earlier algorithm.

Roots in Lancaster

Lancaster stemming is a particularly robust strategy due to its aggressive approach of over-stemming many phrases, which entails condensing each term to its simplest root. This is more aggressive than the Porter stemming and Porter2 stemming approaches, making Lancaster a preferred option when developing language-based software.

The Various Lemmatization Types

Lemmatization in Python is a process that begins by identifying the part of speech (POS) of a word, as opposed to the stemming method which simply reduces a word to its root. This method takes into account the context of the word and then transforms it into its lemmatized form, which preserves the original meaning of the word.

Lemmatizer for WordNet

WordNet, a lexical database, is widely used by major search engines and is regularly employed in information retrieval studies. It is a well-recognised lemmatizer that provides a range of lemmatization capabilities.

Upon inspecting the preceding code snippet, it is evident that the terms “plays” and “pharmacies” have been converted to their base forms, while all other words have not been altered. Generally, a lemmatizer will assume that all words are nouns unless otherwise specified. To prevent this from happening, it is necessary to provide the lemmatizer with the part of speech (POS) tag that is associated with the word.

It is important to note that the application of a part-of-speech tag to the word “better” leads to a different outcome. The tag is able to differentiate between “better” as an adjective and an adverb, recognising it in both senses.

The lemmatizer is required to undertake additional work to accurately identify the appropriate Part-of-Speech (POS) tag for each word. However, when processing a large quantity of messages, providing a POS identifier for each word can be a complex task.

Several lemmatizers are available for use on a given vocabulary.

  • We use the TextBlob lemmatizer
  • San Francisco CoreNLP
  • Extensive lemmatizer
  • Lemmatizer for Gensim

The Spacy lemmatizer approach is the only one that can’t be utilised with POS tags given.

It is possible to create the root form of any word using either lemmatization or stemming techniques. The primary difference between the two is the shape of the resulting stem. In contrast to the potentially unclear outcome of stemming, lemmatization yields a more accurate and meaningful result.

Lemmatization in Python can be a slower process than using a stemming approach, yet it can accelerate the training of a machine learning model. The Snowball stemmer, which is based on the Porter2 stemmer technique, is particularly effective when dealing with large datasets. If the machine learning model you are using relies on a count vectorizer that does not take into account the context of the words or texts being analysed, then stemming is the optimal method to utilise.


  1. Which is preferable? To lemmatize or to stem?

    It’s easier and quicker to install stemming. There might be errors, and they might not even matter for certain uses.

    Instead of simply chopping off the end of words, lemmatization offers more precise results by considering the part-of-speech of the words and thus providing the actual words. Implementing lemmatization is more intricate and time-consuming than Stemming.

    If you value the quality of your results, lemmatization is the most advantageous choice. This approach has a negligible effect on the performance of modern systems. On the other hand, if you prioritise performance, then stemming algorithms are the way to go.
  2. I’m not sure whether I need to use lemmatization or stemming.

    In both cases, the text inputs must be transformed into numerical vectors. An appropriate approach for accomplishing this is to employ Term Frequency Inverse Document Frequency (TF-IDF). This method provides an effective way to quantify the relative importance of words within a document by comparing the frequency at which a word appears with the overall prevalence of the word in the corpus.

    When the vocabulary used in a text is limited but the text is long, using a process called stemming can be beneficial. Additionally, when the vocabulary used in a text is extensive but the text is short, using word embeddings can be suitable. However, the benefit-to-cost ratio of lemmatization is quite low and may not be worth the effort.

Join the Top 1% of Remote Developers and Designers

Works connects the top 1% of remote developers and designers with the leading brands and startups around the world. We focus on sophisticated, challenging tier-one projects which require highly skilled talent and problem solvers.
seasoned project manager reviewing remote software engineer's progress on software development project, hired from Works blog.join_marketplace.your_wayexperienced remote UI / UX designer working remotely at home while working on UI / UX & product design projects on Works blog.join_marketplace.freelance_jobs