TECH + DESIGN / NYC

ITP Blog

RWET - #6 :: Generating Conversation With Word Vectors

>>>>>>>>>>Github<<<<<<<<<<

 
Screen Shot 2018-04-14 at 10.15.36 PM.png
 

I treat this week's assignment as a practice for my final project: Generating peer conversations. Last week I created a weird "drug dealer" Utsav. I found the peer conversation is fascinating because the most frequently happened conversations between me and my friends in New York are that we comfort each other and talking shit about living a stressful life. Comforting my friend is a way to adjust my expectation for life and work. It gets me to think about the bright side and endless opportunities. This week I want to use word vector approach to generate a peer conversation between voice assistant and me.

Voice assistant like Siri, Alexa, Google Assistant, which are all programmed to be efficient and take command, which put them in a subservient position. They are the least one to have a peer conversation. It would be entertaining if we treat them as a depressed friend and comfort them.

Instead of using the built-in machine learning voice interaction, I want to make the voice assistant simply reading my pre-generated script based on a movie script. I choose the script from 500 days of summer. I realize a lot of emotion comfort actually happens when your friends break up. 

Borrow from the simple structure I had last week, scenes between my friend Utsav and I behind the Tisch building. I simply create a conversation as the base:

Google Assistant: I'm depressed.
Me: I will be there.

The answer from me is a simple google search "how to talk with your depressed friend".

The notebook I made:

1. Pull word vectors from a movie script - 500 days of summer.
2. Finding similar sentences with the conversation.
3. Mix it up by random choice

UPDATE:

One of the feedback I received from Ayal is that audience may want to see the arc of conversation. So I delete the ending sentences and let it generate according to the timeline of the movie.

 

The emotional support is a proof that we treat the others like the peer of us. And for people who provide the comfort to the other, they also archive the happiness of helping a friend out.

I'm still exploring the process of making meaningful and interesting conversations more than one round.