Sunday, December 31, 2017

Word of Tooth

Wifey is thinking of starting an Instagram account, so I named it preemptively.

Saturday, December 30, 2017

A tree organization principle

If you are trying to organize a subject into a set of topics and subtopics  - to form a VAR tree, consider what the narratives are that you will be forming with the nodes of the tree, and follow the principle that: if two nodes are connected by a narrative, they should be siblings rather than parent-child related. For example:

PATIENT
    CASE
        MOUTH
        HISTORY
        PRODUCT
SCHEDULE
BILLING

Here we make 'PATIENT' and 'SCHEDULE' siblings because it is natural to talk about where the patient is in the schedule. At the same time, all the dental details are localized to the children of 'CASE'.

Wednesday, December 27, 2017

Geometry - more verbal than visual

Exaggerating in order to make a point: We tend to think of Geometry as the science of visual data, and think of geometric objects as things we can see. I believe that is not correct and the Geometry is the description of what is visible.
In my example of a triangle, I acknowledge we can perceive sides, corners, angles, and such. We may also perceive size, a bit like 'area'. These words name things that are visual perceptions but naming their combination as "triangle" is an added step; and so is saying something that relates these named perceptions to each other. We do not perceive the triangle but rather it is the "invisible web of words and phrases" that make up our understanding of it.

Saturday, December 16, 2017

Me and my rock piles

(picture by Gail Coolidge)

Wednesday, December 13, 2017

Visual Chatbots / Visual Conversations / Visual Conversational Agents

With the help of Python's Tkinter I have a program with a main event loop, that iterates each time some text is entered on a form. The program updates a Narwhal class with the text, and then displays a graphic of that class's internal data. So the program sends the text in, which modifies the class, then the program displays the class's contents. It also responds with output text.
For example:
It is a fun little concept to play with - language controlled VI in its cheapest form. It's on GitHub.

Narwhal reader architecture is not great

I admit I spent much of a year refactoring different reading capabilities with a sequence of "reader" objects - NWReader, NWNReader, NWSReader... It does not really matter because the even higher level classes use simple APIs to that layer and deal with their own data and structure. At the same time the lower level Narwhal classes are evolving. But the crappy middle layer isn't changing much.
Also, to be honest the RecordSlotEvents is a replacement for original core reading methods and keeping the event sequence is equivalent to having a vault. But the RecordSlotEvents is stuck looking at NARs of order <= 1 [waiting for a golden algorithm] and the original core reader vaults aren't perfect either.
So we limp along. It would be fun to get a programming language genius like the author of Python to come clean up the API.

Friday, December 8, 2017

Joking about old UI concepts

They take a subject which must originally be conceived linguistically, convert it into a tabular format concept, and then ask a user to abandon their linguistic constructs in favor of menus, radio buttons, etc. that connect to the table values. Now that we can access the direct linguistic description it is a no-brainer that we should also abandon the tabular format.
But whatever the data format, it is a crime to ignore how people want to think about something and, I am telling you, that can only be about how they speak about the something. (I might avoid this extreme position but think it might be a practical assumption.)

Thursday, December 7, 2017

I am excited about TChat's

Pieces of my NLU world are coming together. Especially around this idea of a TChat that supports the '+' operator and the SetData() method. The TChat is supposed to be a minimal "text in/text out" entity that uses a topic specific tree and a set of topic specific narrative patterns. The TChat extracts information from the text and formulates a response. The '+' operator allows creating more complex TChats from simpler ones. It is more than likely that binding the input and the output within the same object is not quite right; but never mind. What is important is this whole business of "what is the data?". If you can answer that you have a start.

The TChat construct gets close to the heart of an important truth about geometric objects (or for that matter any data type that can be diagrammed or formulated into DB schema) - which is that a geometric object is comprised of a few visual perceptions and a lot of language constructs. For example Euclid's 'triangle' relies on intuitions of points, lines, and angles - which are glued together with words. What does a TChat have to do with this? It is about constructing more complex objects from simpler ones using '+', in a way that mirrors something of what builds understanding of the geometric object. There would be a tree of words related to triangles, and a collection of things that are said using the vocabulary of this tree.

The exciting corollary is that to design a language understanding system about a geometric object is simply a matter of writing down a definition for the geometric object. Quite literally, the language understanding [software] has the same organization as the geometric object because that object is linguistic organized in the first place! This is a huge simplification in the software design process.

So there. When you get harmony between your implementation of a topic and the natural way of thinking about it (sometimes called "business logic") you are on the roll.

But that is not all. Over the last couple days, a colleague named Serge Gomert from Belgium has been showing me what he has learned about NLU with Alexa and Microsoft LUIS. I just took a closer look and darned if Microsoft isn't headed in the right direction. But the good news is that they are still way behind Narwhal. And the last excitement is this: I can get a Javascript widget from Microsoft, designed to embed in a web page, and I am up and running with their voice-to-text accessible to me - where I can do my own NLU and, in particular, do TChats. Those Microsoft suckers may have given me a platform to leapfrog them.

Makes me want to quit my job and focus on creating a TChat wizard.

Parenthetically, the reason I believe LUIS is way behind Narwhal is for reasons big and small. The first is that their 'intent' construct seems to require MxN  rather than M+N variations when there are M choices for one part and N choices for a second part of a two part intent.  Further, although they have a score, it is not signed - which to me is another way of saying that they do not have proper support for sentiment. Still further, they do not support narratives like relation, event, sequence, or becoming. Still further, they do not allow any form of narrative nesting. So they are at the bottom of the ladder. I am three rungs up, waiting for someone smart to figure out the "Golden Algorithm".

So isn't there a commercial opportunity there? I mean if I could insert myself between Microsoft and the public, I might be able to afford a bigger retirement home but, more importantly, be able to continue feeling useful for a few more moments.

Wednesday, December 6, 2017

Oh My! Microsoft publishes Goodness of Fit score for matching text to intent

Just looked at their Luis.
By gosh , that is a goodness of fit score they are reporting. Seems the concept of an 'intent' is fast creeping up on that of a narrative. Maybe they'll get sophisticated, maybe they will beat me at my own game. I'll go to bed not knowing.

Tuesday, December 5, 2017

More about augmented reality and words

What I was saying previously about the "invisible web of words and ideas" surrounding a virtual object, may be a (I want to say) deep connection between the words and the actual geometric thoughts - arranged each in the same kind of structure.
An example of what I mean: working on the different parts of the data that define options for an abutment design - (base shape and width, margin depths, core height and thickness) each has its own linguistic world and each has its own specific words and phrases that people use to describe the option alternatives. I am ending by building a "chatbot" for each lowest level option, and building up a more complex chatbots following exactly the same [hierarchical] structure as the parts make up the whole of the abutment. The proposition then is: to design the abutment language software in the same form and arrangement as the parts of the abutment.

In this view, the geometric shape is a linguistic composite, it is no more than what we can say about it. [I am not sure this is right, but that is the extreme version of the idea.]

Sunday, December 3, 2017

Exhortation to Dental AR/VR

It is not just about flying around a dental design in 3D. It is about doing it with another person.

Virtual and augmented realities are about collaboration as much as personal visualization. For that you need to understand how, as it hovers in mid-air, the abutment it is surrounded by an invisible web of words and ideas. Those words and ideas need to be virtualized and augmented for an AR/VR concept to work. This is the rational for considering a conversational agency as part of the AR/VR product concept.