Modern Conversation Agents.

Litha is the inventor of Conversationers – modern conversation agents, replacing chatbots with powerful conversational skills.

Psycholinguistics (psychology of language) is study of language from a psychological perspective.  Psycholinguistic AI is the application of psycholinguistics in an Artificial Intelligence context.

This covers the cognitive processes that make it possible to generate a grammatical and meaningful sentence out of vocabulary and grammatical structures, as well as the processes that make it possible to understand utterances, words, text, etc.

Our focus on cognitive psychology has meant ongoing research and development of AI associated with perception, thinking, memory, attention, language, problem-solving, and learning.

The outcome of our work in psycholinguistic AI means that we have developed a whole new way for human-computer interaction (HCI).

The first use cases of our technology include AI Psychology and AI Psychotherapy.

Are chatbots & conversation agents at their peak?

Level One: chatbots

Level One chatbots were entirely developed by technologists looking to ‘solve the problem’ of talking with computers.  A reflection of 20th Century Interactive Voice Recognition (IVR) – press ‘one’ for sales, etc., no conversation is involved.  This may work if you are ordering a pizza, but useless for any kind of conversation.  This also meant that Natural Language Processing (NLP) was negligible.  The advantage is that they are cheap.

A cursory look at banking and utility websites with a chat window will show the way customers are directed to specific topics.  These one-dimensional, mechanistic, decision-tree tools work to a particular (formal) syntax using limited speech dictionaries.  These are designed purely for the benefit of the organisation and not for the individual.

Level Two: chatbots

An evolution of Level One chatbots, these are restricted-domain speech recognition systems with larger vocabularies for the spoken (formal and informal) words and phrases.  There is some interaction with the user.  Software companies coming at conversation from a technology perspective focused on building some conversational capability but still within domains.  Don’t talk to a retail website and hope for a wide-ranging conversation.

Enterprises wanted feedback (and ideas on new products / services) and started to use NLP as well as Dialog Management allowing follow-on questions.  This is an evolution of Level One (for example, the Decision Tree model at Level Two was already used in principle in Level One).

Level Three: chatbots

Taking the evolutionary path further, we see Level Three.  Artificial Intelligence (AI) and NLP technology developed significantly with the ability to understand nuances and semantics.  Companies started to talk about chatbot personalities and personalised conversations.

This is what chatbot companies are aspiring to offer – free speech recognition; open-ended vocabulary (formal and informal); far-field (remote) sources; full transcription from any audio/video source, and able to deal with noise, echo, accents, disorganised speech, etc.

This is where many chatbot companies will now stop.  Their focus now is on being a ‘no-code’ company: build your chatbots through plug-and-play and accessing open-source technologies.

Level Four: conversation

At Level Four, you can expect generative conversation about anything (unrestricted contexts) by text and voice.  Not only can it recognise dialect and slang but also work deeply with idioms and metaphors.  At Level Four, conversation is omnidirectional (start a conversation by text through Telegram, follow it up in a box on a website, finish it by voice through a speaker).   For this to happen, it also needs the ability to build short- mid- and long-term memory.

Human conversation works at Level Four.

People are expecting fluid conversations held in context, where meaningful answers are given.

The human conversation can be wide-ranging.  Even the most succinct conversation has tangents and sidenotes.  These are a critical part of human communication.

We explain the difficult or complex through metaphor; we use tangents to help explain a point; we seek to be heard and understood; we want resolution.  If you are engaging with domain-specific chatbots, any tangent will be ignored or states, “I’m sorry, I do not understand.”

With so many chatbots operating at Level 1 and Level 2, are they suitable for your needs?  They may help your organisation to fill in a form, but do they build rapport and consumer confidence?

Technology is wasted on technologists

Natural Language Processing

The underlying work conducted by Litha Labs in this field started with Human-Computer Interaction (HCI).  The question we asked was, “how do we create an environment where people will comfortably speak with a therapist?”.



Increasingly, we engage in Human-Computer Interaction (HCI) and Human-AI Interaction (HAII).

These systems depend on Natural Language Processing (NLP) and the subsets of Natural Language Understanding (NLU) and Natural Language Generation (NLG).

Natural Language Processing is a mix of computer science, AI and computational linguistics intended to support people and machines speak in common language, much the same as a human-to-human discussion.  NLP takes unstructured data and converts it into a structured data format.  It does this through identifying named entities and word patterns.

Natural Language Understanding is where the technology uses syntactic (grammatical) and semantic (meaning) analysis of text and speech to determine the meaning of a sentence.  This is where it is possible to carry out sentiment analysis. NLU also establishes a data structure which specifies the relationships between words and phrases.

Natural Language Generation takes data input. and produces a human language text response (this can also be converted into a speech format).

Conversationers hold
Human-like conversations

Level Four is where technology holds a space for human-like conversation.

Litha’s proprietary Conversationer technology creates safe conversational spaces.   It is built around such concepts as user-centred therapy, using psychological approaches including reflection and summary.  These enriched conversations put the user at the centre of the discussion (a listening space) and helps them to orient themselves within a particular situation / issue.

The six traits of a Conversationer are:

  • Curious – putting the other party at the centre of the conversation; asking questions until the situation / issue is fully understood
  • Attentive – paraphrasing to demonstrate (and clarify) understanding
  • Truthful – being authentic and not faking it (don’t pretend to be human!)
  • Consistent – short- mid- and long-term memory of all conversations
  • Contextual – putting conversations into context
  • Helpful – reaching mutually satisfactory outcomes


All of Litha’s SaaS-based solutions user Conversationer technology.