Thinking Solutions' Blog

Word Sense Disambiguation (Finding the meaning of words) - Reintroduction

johnball@thinkingsolutions.co — Sun, 29 Jan 2012 12:40:00 +1100

For the first post in 2012, it is opportune to look again at WSD. Word-sense disambiguation is the name for the challenge of choosing the correct meaning of a word, its word-sense, in a sentence. To understand the context of a sentence, each of the words must be aligned with the possible meanings in the sentence. See the definition at word-sense disambiguation for a more detailed description of the problem.

We solve the problem by identifying the set of valid senses that apply to a particular sentence. Sentences contain massive amounts of information which apply readily to this problem.

There is another video we will do on this topic that will be released in the next few days. The point is that for computers to understand us, a WSD solution is essential. Like Patom theory, the solution comes from the set of possible answers intersected together to create, yes, linkset intersection.

I look forward to seeing you again soon for the next explanation of this.

Happy to see the video and other phrases

johnball@thinkingsolutions.co — Tue, 13 Dec 2011 13:45:00 +1100

"I am happy to see the video" contains an adjective linked to a verb phrase. What does it mean?

Today's video looks at the structure of such sentences as these to see how Project Turing deals with it.

Errors in language

johnball@thinkingsolutions.co — Sun, 11 Dec 2011 13:11:00 +1100

If all we needed to do to understand a language was to match the words to phrases and sentences, we would be in good shape.

Unfortunately, languages don't work that way. If U NO WOT I um, er mean 2say 2u. Languages operate on many levels and effective machines need to address this aspect.

By storing, matching and using patterns - list and sets only - we are able to get through these problems.

More ambiguity

johnball@thinkingsolutions.co — Sat, 10 Dec 2011 12:00:00 +1100

Today's video shows the demonstration program breaking down the meanings of more examples.

Until next time we meet, good day to you!

Ambiguity - language's core element

johnball@thinkingsolutions.co — Fri, 09 Dec 2011 12:49:00 +1100

Ambiguity in language is a given. You name it and there will be more than one way to interpret it. As I like to say, if you can't think of 100 ways to say something, you probably aren't trying.

For computer programmers and linguists this has resulted in a need for new ways for machines to understand and a corresponding lack of progress.

This video shows how language ambiguity is handled within the Project Turing demonstration program.

The examples are continued in the more detailed video due out soon.

Embedded sentences in phrases within sentences

johnball@thinkingsolutions.co — Thu, 08 Dec 2011 13:21:00 +1100

Human languages use embeddings - sentences within phrases, for example. 'The boy who I like slept' and 'the girl who the boy likes ate' are two examples.

If you are interested in these examples and a brilliant description of them, the author Steven Pinker from Harvard University has written some wonderful books I would recommend including The Language Instinct.

Today's video explores some of these in detail.

Note that the new SAMPLE button on the web is called "Being", instead of "NounClauses" as it was at the time the video was made. We will continue to make changes to the web site and in time will re-record these videos to suit. Please accept our apology for any confusion caused in the interim!

For embedded sentences/clauses within phrases, we are still working through the best way to display some of this information. Any suggestions would be most welcome.

Intelligent Dictionary Summary - What it needs to do

johnball@thinkingsolutions.co — Tue, 06 Dec 2011 08:33:00 +1100

An "Intelligent Dictionary" needs to do more than just look for the parts of speech within a sentence. The meaning of a sentence depends upon the use of its words, or so said Ludwig Wittgenstein. What does that even mean?

Let's analyse a sentence or two to help explore this further.

See you next time,

John

Holiday updates - work plan

johnball@thinkingsolutions.co — Tue, 06 Dec 2011 08:28:00 +1100

What have we been doing? It's been a while between updates!

At Thinking Solutions we've just finished a major program change to move all our patterns to external files. This will allow future products to allow user access to the data, while removing the need for knowledge of the pattern-matching program itself. It's been a big investment for no visible external change, but one we felt was necessary.

As the holiday season begins, we are going to upload and discuss a number of new videos to explain the capabilities of the software.

Until next time, enjoy the start to the year-end break!

John Ball

Review of capability

johnball@thinkingsolutions.co — Thu, 29 Sep 2011 17:23:00 +1100

Our posts this month will focus on some of the technical capabilities of the Project Turing software. As the software matches patterns, the variety found in English are worth looking at in more detail. Today's video shows an example from the development web site, dev.thinkingsolutions.co, in which one of this month's examples is showcased.

The purpose of language comprehension software is to match the words used within their logical phrases and also match the correct word sense, or meaning, within the phrases. While the current focus around the world seems to be to look for statistical matches within large texts to determine meaning statistically, our approach is to enter patterns into the software and allow pattern matching to be done independently by the software.

We think with our movement back to "hand-coded" patterns – an approach last seen seriously about 20 years ago – the future looks bright. By taking this approach, and considering a future in which experience populates individual dictionaries automatically, we envisage machines working with people that converse, learn and interact in natural ways. But that is getting ahead of ourselves; for the moment we are focused on creating an intelligent dictionary that provides the framework for future machines by finalizing a system that works to comprehend human language.

Have a look at today's video to see an online example being dissected.

What is a word?

johnball@thinkingsolutions.co — Wed, 31 Aug 2011 16:25:00 +1100

Human languages allow us to very concisely and accurately talk about things. Although at some levels talk involves ambiguity (many meanings are possible for the words and phrases used) our ability to discuss complex ideas quickly stems from the fact that talk conveys meaning, not mere words.

Our prototype that resulted from project Turing demonstrates how simple patterns can quickly result in very complex meanings. The hierarchy is the key here – by matching patterns in a hierarchy, higher level patterns rapidly take on meaning which as a single level would be complex.

The video today shows how a word can take on many shapes. Whether it is a list of symbols, like in a number, or of characters as in English words or unbroken strings of characters as in Japanese sentences, meaning comes from the same elements of a sentence. Unfortunately for the programmers of the past trying to match these patterns, the pieces that fit into a sentence are quite varied, as varied as the words that make up a language.

The application of Patom theory does away with a number of issues that plagued programmers in the past. The computer finds the atoms that make up a word, based on its current use, and then seeks to apply it to the current sentence.

To try to make this clearer, a word is something that describes a particular idea. So the number “40” represents the concept of forty things. The phrase “John, Beth and Sue” is a word that describes 3 people. And the phrase “ruthere” which reads “are you there” is a question, a full sentence when you expand it out.

We think that the approach of bundling phrases into words on the fly and ensuring it fits into a sentence is a good way to comprehend real languages - as we want to do.