Friday, March 24, 2006

出發前一天

剛好 佈置來讓大家說關於Push的故事的房間
因為很多道具不齊 所以搞的我四處奔波
剛好 要set up一台server讓大學部的學生作小實驗的任務
就是必須要我先做好前製的部分
得在出發前搞定讓他們能好好利用春假的時間
剛好我的Society of Mind偏偏要在昨天被退件
讓我得在這個莫名其妙的時間起床寫功課
剛好Motorola跟LG的人都要在這兩天出現 不見面傳遞經驗不行

學東西壓 學東西
想要可以不斷 再多學一點 再多學一點
學怎樣在沒有人能告訴你怎麼進步 往哪裡進步的狀況下
想辦法藉助人家的幫忙 然後把自己往前推
同時帶別人也往前走
然而莫非定律就是要發生 其實也不出預期
平常貪心慣了的人
一到要出國就是會碰到這種狀況
剛好 也不是那麼令人感到意外

不管怎樣
想到這次要去泰國 不免小小的高興
雖然搞不好到頭來有可能會發現這整件事根本就是個錯誤
起碼是向外踏出的一步
也是我們自己創造出來的 無中生有的一步
若要說學習已經到了這種階段: 化創造為學習
其實我們老早老早就在作這種事了
只不過 從結果看來
我們一步又一步在往前推進
到不同的年齡 在做不一樣scale 不一樣深度的事情
這 很是值得高興的

興奮地
再次出發吧

延伸閱讀

Friday, March 17, 2006

Representing Time in Stories

"ConceptNet doesn't take into account the concept of time", Glorianna always says so. I think it means that, unfortunately, the commonsene knowledge that we have currently does not encode enough time information at all. It seems to be wierd to most of us who are familiar with ConceptNet, since ConceptNet DOES encode causality inside. When it's cloudy, then it might rains later. When you play sports, you may get sweat. When a party begins, people may dance. All these are casual relations. What does she mean by not taking into account the concept of time?

I've been puzzled by this question for a while, and finally we came to this idea in the conversation with my friend ppjoey today.Consider the causal relations in ConceptNet ("I swim"-> "I feel hungry") and ("I feel hungry"->"I eat"). Both these relations are represented as state transformation. That is, from one state to another,just like this figure:
But if we think about it more carefully, these phenomena really don't happen discretely. It's not that I swim at an instant, and at the next instant I feel hungry. If we use this kind of simply representation of the casusal relationship in life, how are we gonna represent the things happening in the following story?

I was working in my office this morning. It was almost noon, and I gradually got hungry. I decided to have lunch, so I put on my jacket, and locked my door. As I am walking out of the office, Henry called me, and I stopped and talked to him. After a small conversation, I turned again and walked out of the office. I met my friend standing in front of the elevator, so we decided to eat together and we walked together toward the cafeteria. He was talking on the cellphone as we walked. After we arrived the cafeteria, we ordered the food, and sat down with some other friends who were already eating there. We ate and talked, and they left after they finished the lunch. Later, we finished ours and left too.

The state"feeling hungry" lasts for a very long period in this story, while "walking to the cafeteria" lasts for a shorter one, and "my friend talking on the phone" an even shorter one, and so on, so forth. It is hard to use the simple way of describing causality above to represent, for further reasoning, everything happening in this story. For example, it's hard to explain, in my friend's case, why the event of buying the food comes after the event of talking on the cellphone.
And, if I forgot my wallet in my office and had to go back to take it, it's gonna be hard to use collaboratively the commonsense knowledge of "feeling hungry"->"eating" and "forgeting to bring something with me"->"going back and get it" without confusing. Yes, of course we can use stack to push goals in our mind and pop them to carry them out. But since each event comes with a different sense of time, it makes no sense to use a single type of stack and push them altogether.

So there're three things I think to be lacked in the current representation: 1) the time span of each state, 2) the time span of the transition between two states, and 3) whether two states can occur concurrently. If these three elements can be incorporated, it would be much easier to solve the representation problem of arbitrary stories. For again the example of "I swim"-> "I feel hungry" and "I feel hungry"->"I eat", one could get a clearer picture when looking at this figure:
The idea in this figure is that, it is a gradual, continuous process for one to feel hungry as he/she swim, and the hunger fades out gradually too as eating more and more. The state of feeling hungry can take place concurrently with either swiming or eating, but eating and swimming have to be done seperately.


延伸閱讀

Wednesday, March 15, 2006

Society of Mind Writing Assignment

第一次寫Society of Mind的作業,發現大概不會有任何其他的課的作業會像這樣,兩頁的英文字花老半天,絞盡腦汁才寫出來。There's no absolute right or wrong with the argument in this essay, but just clear or unspecific. All opinions welcome.

SOM_Essay1.doc

延伸閱讀

Experiencing Collection-Based Media Stories

The problem of collection-based story browsing/viewing/interaction is that the immersion is very poor, as opposed to singly completed videos. The immersion is poor because the video and audio is segmented during the playback, so stories cannot really continue in a fluent way. Ever since stories came into our lives, the ones that are compelling and intriguing to the audience always come with formats that provide great immersion, be them textual fictions, spoken words, live body motion or drama, films, games/virtual reality environment, etc. But, considering the stage at which we are standing currently, jumping through one another clip (or photo) on the web using the media collection is so much like a search (or at best, a conversation) interaction. The whole experience is cut into independent episodes filled with pauses in between, so there's no fluent flow, and there's no immersive expeirence.

I think the way that "textable movie" uses to merge all the video clips together has eliminate this problem. The question follows thus becomes that, if we are applying the same strategy as textable movie, how we're gonna design the interaction mechanism such that the narrative enviornment could take viewers' input at any time and respond by using semantically-related clips based on commonsense reasoning and building a fluent, undisruptive narration? Given limited resource from the viewers, how is the system gonna figure out what makes interesting (or at least, reasonable) stories, and how to use all those clips to tell the stories? How should the clips be chosen, sequenced, and bridged?

There has to be a way, a good way, to present, and for users to interact with, the stories within the collections. Otherwise, the emotions, desires, and all sorts of feelings will not be able to be conveyed in a way where the viewers, as investigators, can explore the story world and create the narration collaboratively with the media creator.

延伸閱讀

Thursday, March 02, 2006

Push in memorial

The most powerful channel for sharing experience is people's words, is people's laghter, is people's silence, is people's hugs, is people's tears. The most powerful channel for sharing experience is the atmosphere created by a group of people. I'm not sure whether this is the right way to put it (or whether you will agree with me), but I disagree with that the combination of text, video, audio, and maybe even physical objects is always good enough for conveying experiences that somebody once had. To me, however, the gathering of people, or in particular, the gathering of the characters in the story, is something I think to be a very powerful channel for conveying experience. For the case of re-experiencing the stories about myself and my friends, looking at a video we took years ago or listening to a song we recorded at home does not necessarily reminds me the experience better than gathering all my buddies and heading to the beach together. For the case of getting to understand others' shared experience, on the other hand, an efficient and effective way would be being shared by the whole group of characters as well.

I was just thinking about Push's death, thinking about the scene in Bartos this afternoon. I guess almost all the people at the lab at that time were there. Everyone was stunned. Everyone was silent. Everyone was moving slowly, slightly having any expression on his/her face, and, the whole atmosphere was just a single word - grief. In my eyes, I saw everyone with their brains blanked out, but at the same time thinking about numerous things, thinking about numerous little stories they have with Push. Maybe because it's what I was doing: I first knew Push when I came for interview last winter. During those two weeks, he was the one I talked to the most, because he was so willing to help. I saw people hugging each other. I was people slowly walked into the elevators and went back to their own offieces. And, if I felt something strong, I believe it was because there were all these people. It doesn't matter whether they talked or talked not, or whether they use any other possible modalities they could use to share their experience. The gathering is powerful enough, I think.

So I'm thinking about making a website for the people in the lab to share stories they have about/with Push. The input can be text, video, audio, drawings, pictures, and maybe still other format. (I don't want people to be limited with their own ways of telling stories, so it's better if I can make it as flexible as possible.) The stories can become a collection of the Media Lab's "Push's Story" (Which is somewhat similar to the photo mosaic work I've been doing as a collection of people's memories: http://web.media.mit.edu/~jteng/project/cm.html) Since what I've been working on is commonsense computing, I think it's easier for me apply what I can do right now in this assignment: to relate people's stories. By analyzing the annotated files they use to tell the stories, the system will be able to show the user all the related stories after they provide their own ones. The difference between this idea and what I've done with Glorianna last semester is that, I think it is really gonna be a collaborative sharing activity joined by many many people, where I can get to see/hear many other people's stories with different styles and channels upon inputting my own one. So the interaction would not feel like an one-to-one interaction. The tools I'm gonna use are stuffs I've already been familiar with: ConceptNet?, WordNet?, Montylingua, etc. I think I will shoot some video as my own input too.

I don't know actually whether it is a good idea to do my assignments using this sad event. But it is something with impact to us, and therefore needs to be documented somehow. (Just like Katrina or the 921 earthquake we had in Taiwan) And, I consider maybe it is good too if we can really do something memorial.


延伸閱讀