https://www.agents-united.org/
Academic multi-agent conversational research platform
Showing posts with label chatbots. Show all posts
Showing posts with label chatbots. Show all posts
An Evaluation of a Linguistically Motivated Conversational Software Agent Framework
A cracking paper from Dr Kulvinder Panesar of York St John University on a design for a "Conversational Software Agent" (aka chatbot) which seems to have all the elements we've been championing for some time:
- Grammatical parsing
- Speech acts
- Dialogue management
- Triples based memory
- BDI motivation model
You can read the paper Academia.edu, and the work dates back to at least 2017, so a few papers to catch up on!
Meena - the latest Google ML Chatbot
There's been quite a bit in the press recently about Meena, Google's new chatbot powered by a "Evolved Transformer seq2seq" neural network model which is even more powerful than the recently touted GPT2. The bot has been trained on "341 GB of text, filtered from public domain social media conversations".
Whilst the results for general conversation seem impressive, I can't help but think that this is a bit of a dead end as the bot really has no idea and is in no way directing the conversation, its just coming up with the "best" response to each input, albeit taking the conversation to date into account. It just doesn't seem to have any intent or goals of its own.
More interestingly though Google has been measuring the performance of the bot (and using that as part of its reward function) with just two variables - how sensible is a reply, and how specific is it. Lots of bots try to go for sensibility and the expenses of specificity, they come back with general statements ("why do you say that") rather than actually trying to drive the conversation forward. For a while now we've been using Grice's maxims as a way of assessing bot performance (quantity, quality, relation, manner) which aim to cover similar ground, but the Google model may be even simpler and sharper - particularly when trying to train a human to do the assessing!
California Law Bans Bots From Pretending to Be Human
Effective July 1, 2018 it became illegal for bots interacting with California consumers to pretend they're human if they're trying to sell goods, services, or to 'influence a vote in an election.'
https://uk.pcmag.com/news-analysis/117718/california-law-bans-bots-from-pretending-to-be-human
W3C Conversational Interfaces Community Group
W3C has set up a Conversational Interfaces Community Group "to enable web developers to collaborate and share conversational experiences for a variety of domains. Most dialogue systems serve interactive experiences in their own domain specific language, causing a fragmented zoo of proprietary formats. For example, Google Home or Alexa do not share a common intermediate representation, which makes writing wide-spread content inaccessible to the mass audience. We study existing specs and design standards to harness proven techniques into common agreement. "
More details at:
Why chatbots still leave us cold - ZDNet
Interesting article on ZDNet about a recent competition administered by Facebook's AI researchers, with help from Microsoft and others, to show the state of the art in chatbots. The overall result was "soul-crushingly boring and nonsensical" according to ZDNet.
But an interesting perspective was then posted by one of the stalwarts on chatbots.org, who, recognising that most of the bots tested were "modern" ones developed by commercial start-ups and based around machine learning/neural net algorithms said "This reinforces my belief that using a neural network for a chatbot instead of being rule based is simply not an option. I have yet to see any neural net based bot that was any good.".
Think I'd tend to agree.
But an interesting perspective was then posted by one of the stalwarts on chatbots.org, who, recognising that most of the bots tested were "modern" ones developed by commercial start-ups and based around machine learning/neural net algorithms said "This reinforces my belief that using a neural network for a chatbot instead of being rule based is simply not an option. I have yet to see any neural net based bot that was any good.".
Think I'd tend to agree.
Arckon Chatbot
Interesting looking chatbot called Arckon from Don Patrick which seems to take a development line far closer to what I'm after than most I've seen. He reports using a grammatical parser to understand the input, then rather than focus on Parts of Speech he extracts the implied triples in the sentence, and then uses a triples base and inferencing to derive the response, apparently he "formulates his own sentences dynamically by feeding the relevant facts through a complex grammatical template. The only pre-programmed answers are a few dozen standard manners, like "Thank you" and "You're welcome"."
The following diagram shows how the triples extraction works:
And an example exchange:
Lots more on his website and blog at: http://artistdetective.com/arckon/
Subscribe to:
Posts (Atom)