Week 1 - NYC!

This is my first time in New York, and I am so excited! I'm living on the 18th floor of East Campus, and the view from our suite lounge is spectacular. Below is a picture my suitemate took of a beautiful rainbow outside our window. Last weekend I took the subway for the first time and went to Times Square and Chinatown.


Everyone in the Speech Lab is very nice, and the gender balance is amazing. I'm working with Ph.D. student Bob Coyne, the founder of WordsEye. I read seven papers on NLP for Twitter, met with the graduate student who was working on my project last year, and started brainstorming templates for using information from tweets to generate 3D scenes.

Week 2 - Settling In

I worked with Bob this week on identifying patterns in tweets, such as emotions (i.e. "jealous of X"), icons (i.e. "birthday" => gift box), light verbs (i.e. "make X" => map instrument/action), and events (i.e. "basketball game"). Then I wrote a Python script to normalize and filter tweets for emotion words, since emotion patterns were extremely common. My greatest challenge by far was reading, understanding, and completely rewriting the code written by the graduate student before me because there were packages that needed reinstalling, a couple bugs, and very little documentation.

I've fallen into a routine by now. I have to avoid eating out because I'm on a gluten-free diet (and had a couple allergic reactions last week), so I cook dinner with my vegetarian suitemate every night, buy groceries on the weekends, and pack lunches. My favorite meal so far was quinoa tabouli and sweet potato patties with avocado (yum). After I leave the lab, I exercise at the fitness center next door. I then spend my evenings reading fun, easy novels (which I never get around to during the school year)!

Over the weekend I went with my suitemate to see the movie Prometheus (which I don't really recommend), got a NY Public Library card, and went with a Barnard student who works in my lab to Central Park for the first time!

Week 3 - Exciting Work

My mentor, Professor Julia Hirschberg, was back this week after spending a couple weeks traveling to and from conferences, such as the NAACL. She met with every project group in the lab, checking in with us and giving us ideas and suggestions. She also came to our weekly lab lunch on Thursday and had many interesting topics to talk about.

This week I have made significant progress on my project. I have been working on extracting information from tweets that can be used to generate a 3D scene. As of now, my Python script can not only identify emotion words in a tweet, but can even extract the topic of the emotion using the part of speech tags of the words in the tweet. Since this is a challenging task that many research papers have been written about, I hope to improve this algorithm in the coming weeks. Using emotion words from the Inquiry and Word Count (LIWC) dictionary, I assign each emotion word to a specific emotion category and include negation (i.e. "don't like"). My script also filters tweets for nouns and verbs that can be depicted by WordsEye, countries that can be represented by flags, gendered words (i.e. "he/she"), and the author's name (which I use to assign his/her gender). Finally, I filter out any tweets with more than 50% unknown words because we are only dealing with tweets written in English for now.

Week 4 - Heating Up

So far, the weather has been very nice, but today the temperature reached 100 degrees. I've been swimming more than usual lately to try to cool off. Thank goodness our lab and dorms are air conditioned!

This week I focused on extracting verbs that WordsEye could represent graphically, as well as the subject and object of the verb, prepositional phrases, WordsEye attributes such as colors or textures, and emotions of the subject and object. Finding the correct subject and object of verbs is quite a challenge, but I wrote a simple algorithm that mostly works. Bob's suggestion for separating verbs into active and passive tenses was helpful, since the order of the subject and object is switched for passive verbs. My script's output is now written to a file in a format that will work with Bob's interface.

Last weekend I went with my suitemate to the Met, which was amazing. We were there for four hours and feel like we didn’t even see a quarter of it all. The roof garden's gigantic sculpture that you can walk through and is full of disorienting mirrors was the most fun for both of us. The Egyptian exhibit was cool, especially with the imported temple and mummies. There were interesting European paintings and fun modern art, old musical instruments, and beautiful stained glass and statues. Oh, and we got in for free with our Columbia IDs! Meanwhile, my roommate has been meeting celebrities, including Nick Jonas and Justin Bieber.

Week 5 - Halfway Already!

It is so hard to believe I am already halfway done. Over the weekend I visited the Natural History Museum with two of my suitemates. We first watched a cool video on the Big Bang theory, where it felt like the whole floor was the screen. Then we walked through the cosmic pathway, which gave us a good feel for how many years humans have been on earth compared to all the time that’s passed since the Big Bang. I also enjoyed the ocean life and dinosaur exhibits.

I made lots of progress again on my research this week. I have been working on identifying the subject and object of emotions expressed in tweets. This was a complicated process that involved dividing emotion words into direct and implied categories, and implementing different algorithms for each. For example, a direct emotion word such as "happy" directly describes the emotion of the subject. However, an implied emotion word such as "yay" indicates that the tweet's author is happy. In addition, I discovered that for some emotional verbs, the subject experiences the emotion (i.e. "He is happy"), whereas for others, the object experiences the emotion (i.e. "That gladdened him").

Today I volunteered as a test subject for an entrainment experiment another student in my lab is working on. It was interesting to hear the contrast between speech that is similar to mine and speech that is different from mine.

As an extremely early birthday present, my parents sent me a camera, so now I can finally take pictures of my NY experience! Check out a couple pictures of the view from my suite below.

view1 view3

Week 6 - The Best Week Yet!

This week I finally checked off my list most of the NY sightseeing I've been wanting to do since I arrived here. My relatives from MA visited me on the weekend and took me to see The Lion King on Broadway, which was absolutely phenomenal. Below are pictures I took of Times Square before the show and the NY Public Library afterwards.

times_square nyp_library

We saw Bryant Park, the Empire State Building, and the Chrysler Building. The next day, we walked through Central Park (picture of the lake shown below), ate at a seafood restaurant in Grand Central Station, walked through the NY Public Library, and visited the World Trade Center Memorial (also shown below).

central_park wtc

To finish off a wonderful weekend, I went with two girls from my lab to get gelato in Alphabet City. During the week, my lab went on a picnic to Riverside Park and saw Grant's Tomb. On the Fourth of July, I went with my suitemate and her friends to Coney Island to observe the famous hot dog eating contest (fascinating, but disgusting). The winner, Joey Chestnut, ate 68 hotdogs in ten minutes.


In the evening, we watched fireworks on the Hudson River, but what was even more amazing was that we could see 15-20 different displays, and lightning, at the same time from our suite lounge. Finally, I attended a Fourth of July special of the NY Philharmonic last night. The West Point jazz band, all dressed up in uniform, played many patriotic songs with the orchestra.

In terms of research, I have determined how to use the Twitter Python API to post tweets to the WordsEye Twitter account I created. Bob helped me categorize the list of icons I made so that I can more easily expand the list.

Week 7 - Almost There

Bob and I are much closer now to completing the first pass at generating images from tweets and tweeting back the image. We decided to focus on tweets that post a photo and express an emotion, since these are simple, but will seem more complex with the added photo. I mostly tested and improved my code this week, but I am now starting to translate the output into sentences that are more compatible with WordsEye.

Last weekend, I went with my suitemate to the Guggenheim to see contemporary art, which is shown in the picture with the spiral architecture. Then I ate the most delicious ice cream ever at the Shake Shack: chocolate custard with peanut sauce, chocolate toffee, and sprinkles. Afterwards, we saw the movie Moonrise Kingdom, which I really liked.

guggenheim shake_shack

On Sunday, my advisor hosted a brunch at her apartment for everyone in our lab. She provided so much delicious, gluten-free food! Yesterday, during lunch, a visiting student from Austria presented his work on error detection in speech recognition using machine learning.

Week 8 - Text Generation

This week I have been exploring text generation while working on outputting sentences that WordsEye will understand. I also learned how to use Python to upload photos to Twitpic and to record traffic analytics on a Twitpic image (i.e. how many times the image was viewed).

Last weekend, I went with my suitemate to see the Phantom of the Opera on Broadway, and it was spectacular! Then we ate at Risotteria, which specializes in gluten-free food and served us the best gluten-free breadsticks I've ever eaten. We also visited Sprinkles, where I got a gluten-free red velvet cupcake. Finally, we checked out Dylan's Candy Bar, a three-story candy heaven. The next day, I went to the New York Botanical Garden in the Bronx with my other suitemate and her friend. It was beautiful and made you feel as if you weren't in the city.

phantom sprinkles

The day before Julia left for another conference, our lab took a picture together, shown below. (I'm the one in yellow standing next to Sarah Ita, the other DREU student working with Julia this summer.)


Week 9 - First Pass Almost Complete

We have finally started connecting all the pieces of my project together. I am running WordsEye from within my Python script, but we're encountering lots of errors right now. Hopefully before long I will be able to run my script and watch it search for tweets with photos, process the tweets, extract verbs and emotions, run WordsEye to generate a 3D scene, upload the WordsEye scene to Twitpic, and post a tweet with the Twitpic link to the WordsEye Twitter account. I've also been using a program called FaceGen to create 3D models of politicians and celebrities who are likely to be mentioned on Twitter (i.e. Justin Bieber).

Over the weekend, I went with my suitemate to a farmer's market in Union Square and to get a deliciously rich dark hot chocolate at Max Brenner's. They even sell chocolate syringes. Then we ate tons of good food at the Smorgasborg in Williamsburg, Brooklyn. I'm eating a gluten-free lobster roll in the picture below!


Since we finished lunch earlier than expected, we decided to go to the top of the Rock, which had the most amazing view! You can see the Empire State Building in the picture below.


On Sunday, I went with a couple girls in my lab get French gelato near Union Square. We ate in Washington Square Park, which is beautiful.


Week 10 - Finished!

It's hard to believe tomorrow is my last day in New York. I have had an incredible once in a lifetime experience this summer that I will never forget. My advisor, Bob, the lab, and my suitemates were all wonderful. I cannot thank them enough. And I absolutely loved my research project.

Last weekend I went to a lovely Korean restaurant and ate delicious gluten-free kale pancakes and "zen noodles". I also got more gelato with Laura, a grad student in my lab.


Today my lab went to see the highline and Chelsea Market! This is a picture of me and Sarah Ita, the other DREU student in our lab.