Hey everyone, so I have been lurking for a few months, but I wanted to jump in and officially say hi. This is a great community and I hope I can add value.

I’m a big fan of NLP and Open source, and the JavaScript community. I’m one of the guys behind Node Natural[1] a NLP toolkit for Node.js. I have also landed commits in Node core, and I’m one of the guys behind the cross platform mobile framework PhoneGap.

I have written a little about bots and took[2] a stab at creating a bot engine[3,4].
It is based on the amazing work of Noah Petherbridge RiveScript and pulling in concepts from ChatScript. I wanted to focus on extensibility and bringing in NLP concepts, while making it easy to use - which RiveScript already is!

I have been thinking allot lately about NLU, NLG and user modelling lately and how that can relate to dialogue systems.

Rob Ellis

[1] https://github.com/NaturalNode/natural
[2] https://medium.com/@rob_ellis/creating-a-chat-bot-42861e6a2acd
[3] https://medium.com/@rob_ellis/superscript-ce40e9720bef
[4] http://superscriptjs.com

Welcome Rob. Nice to see another bot framework using JavaScript.
I have been considering a port of Skynet-AI to run on NodeJS and phonegap.

I will be interested in your thoughts on NLG and interfacing to dialog systems.

In welcome to @rob_ellis and @superscriptjs, I’ve updated the following webpages:

- 100 Best PhoneGap Videos | Meta-Guide.com
- 100 Best PhoneGap Build Videos | Meta-Guide.com

- Best Natural Language Generation Videos | Meta-Guide.com
- Best Natural Language Understanding Videos | Meta-Guide.com

Hi @Merlin I’m not familiar with Skynet-AI. Is there an website I can learn a little more about the inner workings.

On the generation front, I have been toying with a trigram approach using a word lattice and sentence planner, but that leads to more rules and templates. Generating a sentence that is grammatically correct seems to not be overly challenging. Generating a sentence that conveys the correct meaning is a whole other beast.

It seems like the problem always comes back to user modelling and context.

@Marcus Thanks for the welcome!

Welcome. I’ve read your website[2] before and found it a good read. I’m interested to see how you’ll incorporate personality, conversation and user motivations, as I don’t have much time to look into those areas myself (my focus is on NLP and reasoning).
Language generation is a process I think of as the reverse of language processing, with meaning inbetween them. So for instance I have grammar rules to translate language to an internal format, then those same grammar rules in reverse to translate the internal format to language. Standard grammatical template and all that. Coding such a template can get very messy and it would seem this area would benefit a lot from planning.
I’ll be listening

So on the NLU front I read some papers out of UW[1] this weekend and I am convinced more than ever the way forward is with a CCG and not a CFG. They are doing some really cool things mapping language to meaning in smaller domains and I think that could be expanded into more general causes and conversation. They also have a video tutorial series [2] and I was completely sucked in.

So ya, I believe the way forward is to convert text to meaning, pull in some context (past history) relationship with the user, bot personality (something like big 5) and motivational conversation goals, then generate a reply - using the same CCG rules to map back to a sentence planner.

The goals could even be as simple as expand statement, revise statement, form agreement, form contrasting view or ignore and change topic.

But I’m not sure…

[1] http://homes.cs.washington.edu/~lsz/papers/
[2] http://yoavartzi.com/tutorial/

Rob Ellis - Dec 9, 2014:

Hi @Merlin I’m not familiar with Skynet-AI. Is there an website I can learn a little more about the inner workings.

One way to learn more about Skynet-AI is to talk to the bot:
www.skynet-ai.com
Click on the “Hello world” and it will take you to the bot.

There is no website (the system is closed source), but there are 2 threads on chatbots.org that describe the system.

JAIL (TM) (JavaScript Artificial Intelligence Language)
The system is coded in a JavaScript framework that I created to make development and deployment of chatbots/AI easier.

Skynet-AI On-Line
Skynet-AI runs on any device that supports JavaScript. It attempts to identify the meaning of user input and generate an appropriate response.
It excels at natural language math problems and the speed of its responses. The entire AI runs in the target device.

Rob Ellis - Dec 9, 2014:

On the generation front, I have been toying with a trigram approach using a word lattice and sentence planner, but that leads to more rules and templates. Generating a sentence that is grammatically correct seems to not be overly challenging. Generating a sentence that conveys the correct meaning is a whole other beast.

It seems like the problem always comes back to user modelling and context.

Skynet-AI uses a number of techniques to generate responses; ranging from canned/random text to full templated response generators.

- Word Lattice | Meta-Guide.com
- Sentence Planner | Meta-Guide.com

- Combinatory Categorial Grammar & Natural Language Generation | Meta-Guide.com
- Context-Free Grammar & Natural Language Generation | Meta-Guide.com

I’ve taken to using Google Scholar as my dictionary, which has turned my Meta-Guide.com website into a giant word cloud.  It’s becoming a monument as much to what I don’t know as what I do know.  I’m suffering something of a self-induced mini-Singularity, in that I’ve focused my AI on learning more about AI itself (which has distracted me from my primary project).  However, this process is directly related to my interest in natural language generation, in that I’ve actually begun experimenting with automatic summarization on this giant word cloud, which has surprisingly enabled me to write coherently about things which I sometimes know nothing at all about.  This can be seen in my Quora Answers, where my curiosity about the unknown has catalyzed an almost Cyborgean, augmented learning curve.  ;^)

I have some SuperScriptJS questions:
1) Is it possible to look at a online demo of SuperScriptsJs?
2) Is SuperScriptJs backward compatible with the script language of RiveScript?

Hey Matteo,

The short answer is no and no.
I’m frantically developing a full fledged bot and updating SuperScript in parallel, so both of these things are under active development, and I will have something online soon. It is really my highest priority.

SuperScript is a fork of RiveScript and is NOT compatible, but some of the principles are shared. I would say if you are familiar with the RiveScript syntax you should find the transition smooth. Similarly I have pulled in many elements from ChatScript too. 

Check out the docs here for a run down. http://superscriptjs.com/documentation/scripting.html

Perhaps I should write a table sharing some of these differences but, I will save that for another day.

Welcome aboard Rob! It’s always nice to have another chatbot / NLP devotee join our midst!

I too am a big fan of RiveScript and found it rather easy to grasp, powerful and fun! I really haven’t seen too many good examples of a competent RiveScript chatbot online. Mine is not available for online consumption but I will say, it does pretty well and I have done some extensive modifications to the AIML subsets. It’s a labor of love for me.

Can’t wait to see some of your work with SuperScript. Do keep us posted!

Thanks!

Thanks Art.

I posted an update on Medium, talking about some of the new features added. Development is starting to slow down and now it is about improving performance and stability.

https://medium.com/@rob_ellis/superscript-the-big-update-3fa8099ab89a

SuperScript is much more like ChatScript now. How topics are handled, out of bound messages, new condition system and I added a new real-time web editor.

The fact system included even supports Chatscript concept .top and .tbl files for knowledge.
Im still putting together a demo, and will hopefully have something online soon

Rob

Great! Do you think that Superscript can be used to ask quotations and receive them in real time? Or to do (step by step or not) web chat orders?

Matteo

Im not sure I fully understand your question. SuperScript is realtime (or near).

The big improvement I was referring to was the authoring experience. Most Engines (AIML/RiveScript/ChatScript) require you create / edit a file and compile or upload that file into to the engine then the bot processes and used that script for dialogue, and SuperScript was the same until recently.

We now have a web based dialogue editor that is realtime. As you chat you can add gambits/triggers right into the flow. If you get a reply that is wrong, you can tweak or change it.

It should lower the bar for people wanting to get into Botting or who come from other back grounds like Writing, Storytelling and Content Creators.

Hey Chatbot.org friends.

I will be doing a talk at EmpireJS next week on SuperScript or more specifically “Talking to Machines”. If anyone lives around New York (downtown area). I will be in town and would love to talk about the space, or really anything related.

Rob

Great! Would love to be there but work is work for a little while longer and I’m unable to attend.

Please let us know how things transpire and if inclined, a brief synopsis or summary would be appreciated when you are able.

Good luck with your lecture / talk.