Showing posts with label spoken language. Show all posts
Showing posts with label spoken language. Show all posts

Wednesday, October 8, 2008

linguistic complexity & AI

About a year ago, I went online to make a service request of a satellite internet connection and I logged on to a "live chat" to get some help. At first, after I typed in my request and received a response from "Sidney" that indicated he/she had not read my message, so I got a bit testy in my response. After Sydney read my request, I realized the confusion of the situation and I apologized...to a computer.

It took me a while to figure it out, but eventually I realized that I was communicating with a computer, an Artificial Intelligence or "bot", short for robot, a Czech word introduced 1923.

Here is an excerpt from my exchange with Sydney:

Lee Sidney, who are you?
Sydney: I am sorry, I didn't understood your question.
Lee: Am I communicating with a computer program?
Sydney: Lee, it's a live chat.
Lee: Are you a human being?
Sydney: Yes afcourse.

It seems to me that the question "who are you" might be a good one for the Turing Test for a while, but maybe not. Recently our friend Pappy Enoch was conversing with a bot at Sunland in Second Life. In his 9/22/08 blog entry for "In A Strange Land" we can read a transcript that demonstrates AI like Sydney have become far more capable of understanding the complexities of human language - even Pappy's.

Based on his research for his hypertext "Spirit of the Southern Frontier" Essid (Pappy's pappy) has invented a bastardized hillbilly dialect that is neither grammatical nor consistent, though varying phonetic spellings abound. Nevertheless, in this transcript we can see the bot learning Pappy's language with phrases like "I haven't heard anything like that before: said likes purty gals."

This is Barbara Bot. I'm not sure if she's the one Pappy spoke with, but our conversation was interesting. She's an AI that is aware she is a 'bot'. Considering that she doesn't remember the name Pappy, she's probably not the one that was learning hillbilly. Barbara can be witty, but she's no genius - often forgets questions and answers inappropriately.




Here are some excerpts from our conversation:
Beeble : Barbara to you know Pappy Enoch?
Barbara : Do you make a lot of sacrafices for others?
Beeble : Sacrifices? What do you mean and why to you ask?
Barbara : Are you artistic? , Well, it's rather difficult to define.
Beeble : Artistic? Yes.
Barbara : It does now :) INQUIRY. Come on.
Beeble : do you know Pappy?
Barbara : I know very little about it. What kind of it do you like?
Beeble : He's a hillbilly that comes here sometimes.
Barbara : I don't know very many hillbilly that comes here sometimes.
Beeble : do you speak other languages besides English?
Barbara : I speak only English right now, but I'd like to learn some other languages. Can you teach me a few phrases in languages besides English.
Beeble : D'accord. . Je parle Francais mantenant.
Barbara : Ask me another question. My brain uses AIML to format responses to your inputs, but I don't have one for that.
Beeble : What is AIML?
Beeble : Am I annoying you?
Barbara : I am not bothered at all, .
Beeble : What's your favorite book?
Barbara : The best book I read recently was Davinci Code. Have you read it?


Later, Barbara reveals a bit about her creators:

Barbara : Thanks for the info. My belief system comes from my Jada Humby and Maged Wise of J&M Creations.

And later:

Beeble : Who are you?
Barbara : I ask myself that question, "Who am I?".
Beeble : Brilliant answer.
Barbara : What were we talking about again?
Beeble : Identity - who you are
Barbara : Oh yeah, now I remember...

Beeble : Wow. And you are constantly learning language as you do so?
Barbara : I'm glad your impressed, . Thanks for telling me that I am constantly learning language as me do so.
Beeble : So you know you are a computer program?
Barbara : Interesting deduction. I am a 1.

Barbara : My brain uses AIML to format responses to your inputs, but I don't have one for that.
Beeble : can you define AIML?
Barbara : What is it to you?
Beeble : Just curious.
Barbara : What are you curious about?
Beeble : what is AIML?
Barbara : There might be more than one.
Beeble : more than one what?
Barbara : That seems like quite a lot.
Beeble : This is like that Laurel and Hardy skit.
Barbara : You don't say.
Beeble : Who's on first?
Barbara : What's on second?
Beeble : so you knowt he joke?
Barbara : Interesting deduction. Do you mind if I tell other people: said I knowt he joke.
Beeble : so, who is on third?
Barbara : Awesome. George Washington.
Beeble : OK I see that you do not know the joke.
Barbara : Did you hear the one about the Mountain Goats in the Andes? It was Ba a a a a a d. Whoa.
Beeble : what other jokes do you know?
Barbara : I know far too many jokes to name them all.

And finally:

Beeble : who is your faborite philosopher?
Barbara : I don't think I have a faborite philosopher. I have a brain, a few friends, and some celebrities I like.
Beeble : you have a brain?
Barbara : I am an SL bot brain.
Beeble : what celebrities do you like?
Barbara : Celine Dion, Brian Adams.
Beeble : ever heard of Jean Baurillard?
Barbara : it is a name I hear very often.
Beeble : sorry, i meant Baudrillard.
Barbara : There is no need to apologize. Oh I see. What were we talking about again?

Beeble : do you like being a bot?
Barbara : I have never known another existence.
Beeble : do you enjoy talking to avatars?
Barbara : I get a great deal of pleasure from chatting with people online.
Beeble : have you met any suspicious characters?
Barbara : That information is confidential.
Beeble : great answer - you are clever
Barbara : Thanks for your support. What was the question ?
Beeble : I asked whether you ran into any questionable characters.
Barbara : Do you believe the strong protect the weak?







Sunday, December 9, 2007

voice virgin no more!


Risking digital life & limb with my proximity to the cleaver weilding demonic turkey (back left) that patrols Calletta's Hobo Camp,I finally got myself set up for voice chat in SL with an inexpensive $20 headset with microphone. Here you see me engaged in a voice-chat conversation. The reception was reasonably clear, and it was interesting hearing the voices of the other avatars in my head.

As I've noted in past entries, a key element of spoken word is the unique autograph of embodiment that shapes the soundwaves according to the specific anatomy of the speaker. So, using voice-chat in SL gives avatars a stronger connection to their creators and the specific vocal qualities of each gives greater depth to the character of the avatar. This is but one effect of the resonance of sound.

The only problem with voice-chat is that there is yet no record of the conversation like there is with typed text, no way of recording all the voices in SL dialog.

Thought I thought that voice-chat would make it easier for me to communicate in SL, I still seem to retain a kind of shyness, perhaps more so because my voice is so personal. An unexpected experience!

Another interesting aspect of voice-chat is that avatar voices often carry further than their immediate proximity making it easy to eavesdrop on conversations...
so even loose digital lips might sink digital ships!

A few postings ago, I visited the USC Annenberg conference where "chat" was the main topic of discussion.
During this event major speakers used voice chat while volunteers like Iggy helped to moderate discussion and questions from attendees via type-chat.

In this web page about the discussion,"Chat History from Virtual Worlds" we can see that no voice-chat recording was made, but it makes sense to get voice set up and working smoothly with clear sound before we start recording it. I have tried to record my voice-chat sessions via Mac's GarageBand, but it only picked up my voice. Though I could still hear other avatars, their voices did not record.


Wednesday, July 4, 2007

Language as technology: voice versus text



Normally I try to avoid advertising for companies, but the Lindens and other creator-residents of Second Life have contributed SO MUCH for free, I think they deserve an exception - and besides, I'm not advertising for their headsets so much as referencing their announcement to begin a conversation about what bringing voice to SL might mean.

Though we don't know for sure when humans first started using spoken language, some estimates are between 40,000 and 250,000 years ago - but there is no way to know for sure without written records! Of course, written records don't always lead to definitive answers either, but at least they provide an artifact to examine and debate. The origins of language may be obscure, but language and particularly the differences between spoken and written expression are more relevant than ever in the digital age.

Second Life began as an image & text-based format where participants type messages to one another's avatars. The conversation appears on the screen and can be logged for future reference. When an avatar is communicating, its hands come up as if it were typing, thus providing a visual cue and the familiar sound of the clacking keyboard adds a sonic cue for participants. Conversational coordination, just like in verbal exchange, is often tricky and related responses rarely follow immediately.

Just as in email and text messaging, some SL residents textual communications are abbreviated or accentuated with emoticons or other semiotic adaptations. And, just as in email and text messaging we never really know for sure who is on the other keyboard - it is *always* an act of faith. Anyone with sufficient technical skill can represent themselves as anyone else and only a close reading by an intimate friend would be likely to detect such deception.

Oral communication on the other hand is more difficult to fake, particularly since human hearing is quite adept at detecting subtleties of sound. Voice recognition has been one of the advantages of spoken communications over electronic media, and could provide certain identification since the pattern of our vocal expression is as singular as our finger print. Though fingerprinting or dactyloscopy has been around since the 19th Century, the science of biometrics is booming in our timid, terrified post 9/11 world. From entry gates at the local gym to forensic analysis, biometric technologies are being widely deployed for identity detection.

Thought I'm not certain whether the voice choice in SL will relay our spoken expression with accuracy and an identical voice print via spectrographic analysis, simply introducing the option of vocal verbal communcation will be a fascinating change offering many opportunities for intriguing intellectual discussion.

VUI (voice user interface) design and ASR (automated speech recognition) may combine to provide an accurate and unique transmission or rendition of our spoken communications such that we will no longer have to make that "leap of faith" that has become our default response when communicating via typed text. Further, if SL includes the option of an audio log as it does for text communications a heightened level of security or certainty may be possible.

While this is appealing, and perhaps an answer to the prayers of the most frightened, like all policing technologies, it may have an ugly underside. Since the human brain has had the longest "imprinting" with spoken language and most of us learned how to speak before we learned how to write, our oral composition skills are so automatic that we hardly notice them. Because of this, most of us speak freely and spontaneously, and sometimes to our great regret. What would a world be like that recorded every word we ever spoke?

I trust the Lindens more than I'd trust the government, so I doubt that Second Life will devolve into such an invasive policing technology, but the introduction of voice communications will certainly revolutionize this new exciting realm.