Alice Artificial Linguistic Internet Computer Entity ALICE is

  • Slides: 18
Download presentation
Alice Artificial Linguistic Internet Computer Entity

Alice Artificial Linguistic Internet Computer Entity

 • ALICE is super ELIZA!

• ALICE is super ELIZA!

The Loebner Prize

The Loebner Prize

Alice and the Bots • • Richard Wallace ALICE – a conversational agent Multiple

Alice and the Bots • • Richard Wallace ALICE – a conversational agent Multiple clients and wrappers Supported by A. L. I. C. E. Artificial Intelligence Foundation (www. alicebot. org) • Many versions: Program. D, Program. N, etc.

Alice Structure Front end Main Loop Graph. Mapper AIML Pre-processors: Substitution Normalization Attached Processors

Alice Structure Front end Main Loop Graph. Mapper AIML Pre-processors: Substitution Normalization Attached Processors

Attached Processors • • Prolog Scripting engines Escape to System Open. CYC

Attached Processors • • Prolog Scripting engines Escape to System Open. CYC

AIML <category> <pattern> A BOOK </pattern> <template> Should not be judged by its cover.

AIML <category> <pattern> A BOOK </pattern> <template> Should not be judged by its cover. </template> </category>

Graph. Master AIML X Input -> Response • • • Needed – an efficient

Graph. Master AIML X Input -> Response • • • Needed – an efficient algorithm Matching Order must be defined Depth first search Patterns may be ordered by some bots Graph. Master works word by word

Graph. Master Structure • • AIML patterns are transformed into a tree Nodes of

Graph. Master Structure • • AIML patterns are transformed into a tree Nodes of tree are words to match Input transformed into a path Recursive (<srai> tags)

Matching Order • Constrains on order: – Wildcards: _ and * – Last Bot

Matching Order • Constrains on order: – Wildcards: _ and * – Last Bot output: THAT – Bot Topic: TOPIC • Order: – Match _ first – Match word – Match *

Path • Input string becomes: • input <that> THAT <topic> TOPIC

Path • Input string becomes: • input <that> THAT <topic> TOPIC

Brain Image • Alice – ~ 24, 000 Categories

Brain Image • Alice – ~ 24, 000 Categories

Why does ALICE work? • Combinatorics – Assuming 10 word choices per word –

Why does ALICE work? • Combinatorics – Assuming 10 word choices per word – 20 word sentences -> 1020 sentences • Alice testing – 1800 starting word for 95% coverage – Second word choice is ~2

Digital Buddies • Enter the marketer! • Active. Buddy Inc. – Buddy. Script SDK:

Digital Buddies • Enter the marketer! • Active. Buddy Inc. – Buddy. Script SDK: – NL Support – Rules for Knowledge Domain – Dialog support with Query capability – Collects a user profile

Where Next? • Does Alice think? • How well does Alice work as a

Where Next? • Does Alice think? • How well does Alice work as a front-end?