UMASS BOOK TRANSLATION DETECTION DATASET

Purpose:

These datasets were created to evaluate the effectiveness of the translation detection frameworks. The 2K dataset was created by Krstovski and Smith (2011), and the list of translation pairs was updated for use in Yalniz and Manmatha (2012). The 30-book dataset was created for use in Yalniz and Manmatha (2012). If you decide to make use of the dataset in your own work, please read the copyright notice first and also cite the papers listed below. This dataset is for research purposes only.

- I. Zeki Yalniz and R. Manmatha, Finding Translations in Scanned Book Collections. Proceedings of SIGIR’12.

- Kriste Krstovski and David A. Smith, A Minimally Supervised Approach for Detecting and Ranking Document Translation Pairs. Proceedings of the 6^th Workshop on Statistical Machine Translation, 2011.

IMPORTANT NOTICE:

According to the Project Gutenberg and Internet Archive websites, the books are out of copyright in the United States. This may not be the situation in a particular country so you are advised to check this and follow the law of your country. If you just want to read the book, you are better off looking at their websites where they have much nicer interfaces for doing this. We do not know the specifics of the OCR and preprocessing used.

THIS DATA IS PROVIDED BY THE UNIVERSITY OF MASSACHUSETTS AND OTHER CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS DATA, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Dataset characteristics:

The dataset consists of two book collections. They include raw text and corresponding metadata for each book. It does not include any page image. All the books are downloaded from the Internet Archive website (www.archive.org). Project Gutenberg and the Internet Archive disclaimers in the front or rear are removed in order to avoid false matches.

- The 30-book dataset: 30 books (16 English, 14 German)

- The 2K dataset: 2000 books (1000 English, 1000 German)

Lists of translation pairs are also included for both datasets. Each line in the ground truth file specifies a book translation pair where the first column indicates the filename of the book in English.

- 30books.truth

- 2K.truth

How to obtain the dataset?

Please contact downloads[at]ciir[dot]cs[dot]umass[dot]edu. A download link will be provided.

Last updated: June 1, 2012