README.md

MTweet

A collection of 13,150 tweets on Machine Translation (MT). All tweets

• were created between 1 Jan 2015 and 31 Jul 2017, and
• contain the string ‘machine translation’ and/or ‘machine translated’.

We provide id (id), timestamp (created_at), full text (text), and polarity (see below) for each tweet.

Sentiment Analysis

All tweets come with polarity labels, where

• 1 denotes positive tweets,
• 2 denotes neutral tweets, and
• 3 denotes negative tweets.

Labels in the automatic column were assigned by an automatic sentiment classifier (Baziotis et al., 2017) trained on 50,000 tweets not related to MT (i.e., the SemEval 2017 training data; see Rosenthal et al., 2017).

Labels in the human_A and human_B columns were assigned by two independent human judges. Human judgements are available for 150 tweets in this collection.

For details on data collection, filtering, and classification, please have a look at our paper.

Reference

If you use any of the data provided here, please reference the following paper:

@inproceedings{mtweet,
  Author = {L\"{a}ubli, Samuel and Orrego-Carmona, David},
  Title = {When Google Translate is better than some human colleagues, those people are no longer colleagues},
  Year = {2017},
  Booktitle = {Proceedings of the 39th Conference on Translating and the Computer},
  Address = {London, UK},
  Month = {November}}

Important Note

While we share this data directly for you to use without downloading full texts from Twitter, it is your responsibility to maintain Twitter's Terms of Service and to delete tweets that are no longer public.