The first article in this series introduced the concept of Generative AI as well as its use of Large Language Models. Here the author will cover Natural Language Processing, a key component to Generative AI.
An introduction to Natural Language Processing
A key component to Generative AI is what is known as Natural Language Processing, or NLP for short. This can actually be viewed as yet another subset of Machine Learning, as this is crucial when it comes to creating audio outputs or giving a Digital Person the look and feel of an actual human being.
The Tasks of Natural Language Processing
By itself, an NLP model can perform the following tasks:
- Speech/Voice Recognition: This is where the human voice is converted into text format. Although the algorithms that handle are advancing, there are still some obstacles to be overcome, which include the following:
-
- How people speak: Slurring words together.
- Changes in the annunciation., such as different accents.
- The usage of incorrect grammar.
- Part of Speech Tagging: This is where the NLP model attempts to ascertain that part of speech of what a word is in a sentence. For example, is it a noun, verb, etc.
- Word Sense Disambiguation: This is where Semantic Analysis is used to segregate the context of word if it is used in multiple sentences. For example:
-
- Will I “make” the team? This connotates a sense of “achievement.”
- Will I “make” the cake? This connotates a sense of an “action.”
- Named Entity Recognition: Also technically known as “NEM”, this is a part of NLP that tries to identify words and their corresponding phrases into a particular entity. For example, “Boeing 747-400” would be classified as an airplane, and “Indiana” would be classified as a state.
- Sentiment Analysis: This is where the NLP model tries to ascertain or gauge the emotional mood of a person.
- Natural Language Generation: This is where the NLP model tries to convert a block of text into actual human speech.
- Tokenization: This is where the words in a sentence are broken down into different “Tokens,” which are mathematical representations of them.
- Natural Language Recognition: This is often confused with Speech to Text Recognition, but it is quite different. This method takes large blocks of a structured foreign language and breaks it down into a more concise, and comprehensible block of text.
- Toxicity Classification: This is a technique used by an NLP model to classify any kind of hostile threat, whether it is spoken or in text. From there, it can put them into different subcategories, such as threats, verbal abuse, bullying, insults, obscenities, etc.
- Information Retrieval: This is where an application such as ChatGPT would fit in nicely. For example, once an end user submits a query to it, the GPT4 algorithms will try to find the most relevant answer by using either one or even both of the following:
-
- Indexing: This consists of a Vector Space Model, and the sub process is done using a Two Tower Network Model.
- Similarity: This is a statistical technique where a system of distance scoring is used to determine the degree of similarity between the resources that are used to compute the output to a query.
- Lemmatization: This is a technique where the NLP model breaks down a word into the most basic root form. This will be reviewed in greater detail in the next article in this series.
Up Next: Natural Language Processing
Now that you have a basic grasp of Natural Language Processing (NLP) as used in Generative AI, the next article in this series will delve deeper to examine the steps and nuances of how that processing occurs.
Sources/References: none for this article
Ravi Das is a Cybersecurity Consultant and Business Development Specialist. He also does Cybersecurity Consulting through his private practice, RaviDas Tech, Inc. He also possesses the Certified in Cybersecurity (CC) cert from the ISC2.
Visit his website at mltechnologies.io
