Compete with a neural network

In this little game you get a chance to compete with a neural network in restoring missing commas, periods and question marks in text. The network learned to restore punctuation by reading about 40 million words of text during its 15 hour training.

This neural network was developed in the Institute of Cybernetics at Tallinn University of Technology and your participation helps our research. So please do your best when playing! Also, we encourage and appreciate sharing this game with as many people as possible if you find it interesting or entertaining.

Can you beat a neural network?

Do your best to get commas, periods and question marks right in the following text.

So far you have won {{wins}} times and lost {{losses}} times.

1. Click on words to change punctuation:

{{w}}{{vocab[myPuncts[$index]]}} text continues...

2. Click "Evaluate me!" to compare your punctuation to the neural network and the original:

Processing...

In this game, you compete with a bidirectional recurrent neural network. Your performance is logged and helps us analyze how our model compares to humans.

The source code of the model is available here and you can read more about it here.

We used roughly first 80% of lines from the Europarl v7 monolingual English corpus as trainging data, next 10% as development data and last 10% as test data. This game takes random samples from the test set. The training set size was about 40 million words and the corpus was obtained from the IWSLT 2012 TED task web page.

E-mail me at ottokar.tilk@phon.ioc.ee if you have any further questions.

You win!

You lose!

It's a tie!

{{myScore | number: 1}} vs. {{nnScore | number: 1}}

You:

{{w}}{{vocab[myPuncts[$index]]}}

Neural network:

{{w}}{{vocab[nnPuncts[$index]]}}

Expected:

{{w}}{{vocab[correctPuncts[$index]]}}

Share