top of page

Artificial Intelligence: NLG, NPL & NLU

Welcome to the AI Hub. In designing bots today, we use a combination of technologies that include Large Language Models (LLMs), Natural Language Processing (NLP), Natural Language Understanding (NLU), and Natural Language Generation (NLG) which all play central roles in facilitating sophisticated, human-like interactions between bots and users. In this hub, we will explore how these technologies are used in building Conversational AI Agents.

Building Bots with LLMs, NLP, and NLU

Old Way

In the early stages of chatbot development, the foundational technology utilized was Natural Language Processing (NLP), a system dedicated to understanding and analyzing human language in a structured manner.


Following this period, the introduction of Natural Language Understanding (NLU) significantly enhanced a chatbot's ability to interpret and respond to nuanced inquiries, by digging deeper into the contextual and semantic intricacies of natural language. However, the deployment of NLP and NLU necessitated substantial investment in terms of time, effort, and resources to meticulously craft numerous conversational flows, essentially mapping out a wide array of potential dialogues.

Fortunately, the advent of Large Language Models (LLMs) has revolutionized this landscape, automating the majority of tasks previously undertaken manually, thereby drastically reducing the labor intensity involved in chatbot development. LLMs like GPT-3 have the capacity to engage in a Q&A format proficiently, managing a broad spectrum of inquiries with human-like adeptness.

Looking ahead, the future of crafting Conversational AI agents lies in harnessing the prowess of LLMs for the bulk of Q&A interactions, while strategically employing NLP and NLU to fine-tune responses in specialized scenarios. This dual approach facilitates the delivery of specific responses where necessary, and the tracking or tagging of certain conversation types to improve performance and understanding in dedicated tasks.


Now, let's look at how to buid bots going forward. 

The New Stack of Building Bots 

As we venture further into the future of building bots, it becomes apparent that the roadmap for conversational AI projects will be firmly grounded in two pivotal components: Large Language Models (LLMs) and combined Natural Language Processing and Understanding (NLP/NLU) strategies.


Firstly, LLMs will be entrusted with overseeing the majority of Q&A interactions, bolstered by connectivity to expansive knowledge bases that facilitate rich and informed dialogues. The efficacy of LLMs will be continually heightened through sophisticated approaches including prompt engineering and fine-tuning methodologies, ensuring a dynamic and adaptive response mechanism that resonates with human conversational patterns.


Secondly, while LLMs will carry the brunt of the interaction load, NLP and NLU will hold significant roles, particularly in scrutinizing areas of strategic and legal paramountcy in business operations, mandating the crafting of specialized flows for precise and accurate responses.


Moreover, the incorporation of NLP/NLU facilitates a granular tracking of conversational dynamics, offering insights that can be leveraged to refine understanding and elevate performance continually.


By symbiotically integrating the vast comprehension ability of LLMs with the fine-grained analytical prowess of NLP/NLU, the horizon looks promising for conversational AI, paving the way for bots that are not just responsive, but intuitively understanding and adaptively intelligent, marking a new pinnacle in AI-human interaction.

Building Bots with LLMs, NLP, and NLU

Using LLMs

Contemporary Large Language Models (LLMs) like GPT-3 have substantially automated the implementation of sophisticated NLP and NLU functionalities, handling a wide array of tasks right out of the box. They can generate text, understand context, and engage in natural language conversations with remarkable proficiency.

However, despite their capabilities, there remains a significant role for specific NLP and NLU processes in fine-tuning these models to specific applications and in maintaining control over their outputs. While LLMs can generate human-like responses, they sometimes generate incorrect or nonsensical responses and can be sensitive to the input phrasing. Employing NLP and NLU systems alongside allows for a more controlled and refined application, where they can help in filtering outputs, correcting errors, and ensuring the maintenance of a certain standard or guideline, a process that can be vital in professional or sensitive contexts.

Furthermore, NLU can be utilized in sentiment analysis and emotion detection more rigorously to add an extra layer of understanding regarding the user’s intent, which can be essential in customer service bots to escalate issues, route queries more accurately, or offer personalized services. Meanwhile, NLP techniques can be leveraged to develop specific solutions, such as extracting precise information from a large corpus of text, summarizing content, translating languages, and more, which may require a level of precision and specialization that generic LLMs can't provide without guided, specialized training or adjustment. Thus, while LLMs offer powerful, broad-spectrum solutions, NLP and NLU retain critical roles in fine-tuning, specializing, and controlling applications to meet specific needs and standards.

Using NLP

LLMs can handle a lot of tasks "out of the box," but employing NLP techniques can often enhance performance or enable specialized functionalities.


Here are a few examples on how NLP could be utilized in each case, and why it might be needed even with an LLM:

  1. Information Extraction: Specific NLP techniques like Named Entity Recognition (NER) can be employed to identify and extract particular types of information more reliably than a general LLM might.

  2. Content Summarization: While LLMs can generate summaries, employing specialized NLP summarization techniques can allow for more controlled summarization, extracting key sentences or even generating abstractive summaries that paraphrase content to maintain the essence while being concise.

  3. Sentiment Analysis: NLP involves the use of sentiment analysis algorithms that can detect subtleties in language to accurately determine sentiment, which might be used to analyze customer feedback more reliably than a generic LLM.

  4. Speech Recognition: Speech-to-text conversion is a specialized NLP task which involves recognizing spoken language and converting it to written text, a fundamental preprocessing step before an LLM can be employed.

  5. Language Translation: While LLMs can translate text, leveraging NLP can ensure better accuracy through techniques such as Statistical Machine Translation (SMT) or Neural Machine Translation (NMT), fine-tuned for specific language pairs or domains.

  6. Keyword Extraction: Specific NLP algorithms can be employed to extract the most relevant keywords from a text, which can be a crucial step in understanding and categorizing content effectively before further processing by an LLM.

  7. Syntax and Grammar Correction: NLP techniques can be employed to analyze the syntactic structure of sentences, identifying and correcting errors more effectively than a general-purpose LLM might.

  8. Question Answering: While an LLM can answer questions, employing NLP can involve breaking down questions to their fundamental components, and understanding the underlying intent more accurately, to fetch the most appropriate answers from a database or corpus.


In each of these cases, while an LLM can perform the task to a reasonable degree, employing NLP techniques allows for greater control, precision, and often, a better-tuned solution, especially when we consider specific use-cases or domains where specialized handling of language is required. It allows developers to leverage the strengths of both the LLM and specific NLP techniques to build a more robust and efficient solution.

Using NLU

Natural Language Understanding (NLU) in bot development using Large Language Models (LLMs) can enhance the system’s ability to grasp the nuances of human communication and provide more intelligent responses.


Here’s how NLU can be employed in specific instances:

  1. Information Extraction: Beyond just recognizing entities, NLU can help in understanding the context around the entities to extract information more accurately, especially when details are embedded in complex, nuanced sentences.

  2. Content Summarization: NLU can aid in understanding the thematic essence of content, ensuring that summaries not only contain key points but maintain the overarching narrative or argument presented in the original content.

  3. Sentiment Analysis: NLU can delve deeper into understanding the subtleties of language, identifying sarcasm, or distinguishing between positive criticism and negative criticism, providing a nuanced understanding of sentiments expressed in text.

  4. Speech Recognition: After converting speech to text, NLU can help in understanding the context, the implied meanings, or the intentions behind the spoken words, providing a richer understanding of the user’s inputs.

  5. Language Translation: NLU can aid in preserving the original sentiment, humor, or cultural references in the translated text, ensuring a more natural and authentic translation.

  6. Keyword Extraction: NLU can enhance keyword extraction by understanding the context and the prominence of topics in a discourse, ensuring the extracted keywords truly represent the core topics discussed.

  7. Syntax and Grammar Correction: NLU can assist in understanding the intended meaning behind poorly constructed sentences, aiding in the generation of corrections that preserve the user’s original intent.

  8. Question Answering: For question-answering systems, NLU can play a crucial role in comprehending the deeper intent behind questions, distinguishing between literal and figurative language, and providing responses that are well-aligned with the user’s true inquiry.


In each of these scenarios, NLU enhances the bot’s ability to comprehend the subtleties and complexities of natural language, facilitating a more sophisticated, nuanced, and intelligent interaction with users, enhancing the user experience significantly.


Integrating NLU strategies can indeed be a cornerstone in building bots that not only understand what is being said but grasp the underlying meanings, intents, and emotions conveyed in natural language communication.

Using LLMs in your Chatbot Projects


Using LLMs requires a new way of thinking and a new set of tools. Here below is an overview of what is needed to ensure the best possible accuracy and performance. 

We have created an entire section with case studies on how to design a knowledge base so that they are easily digestible by LLMs.


In the rapidly evolving landscape of Conversational AI, prompt design, engineering, and tuning have emerged as pivotal factors in optimizing the performance of Large Language Models (LLMs).


Prompt design entails the meticulous crafting of inputs to effectively communicate with the model, setting the stage for more intuitive and productive dialogues.


On the other hand, prompt engineering dives deeper, exploring advanced strategies to fine-tune the prompt's structure, incorporating control tokens, or refining keyword selection to guide the model’s responses more efficiently.


Further refinement can be achieved through prompt tuning, a process that involves adjusting the parameters and configurations of the underlying model, enhancing its capacity to generate desirable and contextually appropriate responses.

The symbiotic integration of these three facets holds transformative potential for Conversational AI projects, aiding in the creation of chatbots that not only understand and respond to queries but do so with a nuanced understanding and adaptive intelligence.


By focusing on tailored prompt strategies, developers can steer LLMs towards a path of deeper understanding and more human-like interactions, thereby setting a new benchmark in user experience and operational efficiency, and ushering in an era of bots that are not merely responsive but demonstratively insightful and remarkably intuitive.

Knowledge Bases

The LLMs can use Retrieval-Augmented Generation (RAG) and get information from our Knowledge Bases. 


With Knowledge Bases, the LLM works the same way a librarian does. The librarian should know everything about what is in her library. She would know exactly which chapter of which book to suggest to a visitor who asked a certain question.

This explains a semantic search engine in a more technical way. In this case, embeddings are vectorial representations of document parts, and they make it possible to describe mathematically what each section actually means. By comparing embeddings, we can figure out which parts of writing have the same meaning as other parts. This is important for the process of recovery shown below.

Based on the question, you first get the most relevant information from your internal knowledge base. You then add to the normal generation part by passing this relevant information directly to the generator component.

Learn more about Knowledge Bases here.

Fine Tuning an LLM

Fine-tuning is the process of training a large language model (LLM) to a specific task or domain of knowledge. It involves re-training a pre-trained model on a smaller, targeted dataset. The process adjusts the model's weights based on the data, making it more tailored to the application's unique needs.

For example, an LLM used for diagnosing diseases based on medical transcripts can be fine-tuned with medical data. This LLM will offer far superior performance compared to the base model, which lacks the required medical knowledge.

Fine-tuning can help you create highly accurate language models, tailored to your specific business use cases.

Fine-tuning a large language model (LLM) can be expensive and complicated. It involves retraining a pre-trained model on a smaller, task-specific dataset. The new dataset is labeled with examples relevant to the target task. The model can then adjust its parameters and internal representations to become well-suited for the target task.

Here are some considerations when fine-tuning an LLM:

  • Your dataset needs to represent the target domain or task.

  • You need enough training examples in your data for the model to learn patterns.

  • You might not be able to mimic GPT with an open source model.

  • Fine-tuning a large language model (LLM) can cost between $0.0004 and $0.0300 per 1,000 tokens. The cost depends on the type of model you're using and the fine-tuning algorithm you choose. Some algorithms are more computationally expensive than others.

There are also a few Disadvantages to fine-tuning:

  • You will need to maintain the upkeep of your model.

  • Your model will essentially be an expert in a domain instead of a librarian that retrieves information. This makes it more difficult to update, change or remove information from the model.

  • The model can integrate concepts, which lead to new ways of communicating ideas which can be inaccurate 

  • It can be difficult to uncover where certain answers come from

Pre-Training Model

Pre-training is the process of training a model on a large corpus of text, usually containing billions of words. This phase helps the model to learn the structure of the language, grammar, and even some facts about the world. It’s like teaching the model the basic rules and nuances of a language. Imagine teaching a child the English language by reading a vast number of books, articles, and web pages. The child learns the syntax, semantics, and common phrases but may not yet understand specific technical terms or domain-specific knowledge.

Training a large language model (LLM) can cost millions of dollars. The cost of training a single model can range from $3 million to $12 million. However, the cost of training a model on a large dataset can be even higher, reaching up to $30 million.

Future of the Internet

Knowledge is more than just information. It can open up a person's eyes to seeing the world in a whole new way. It has the power to reveal. A good example of this is Amazon. Using your smartphone, you can see what a couch would look like in your living room. They have taken the information about the couch and turned into into an experience. This is where the internet us going.

bottom of page