Western World: Localizing a Mobile Game

For our final project in Software & Games Localization this semester, my group decided to localize a mobile game called Western World, available in the App store and the Google Play store, into Japanese, French and German. Aaron Szucs, the creator of the game, kindly made the source code available for download on GitHub.

The game is a dressed-up version of rock-paper-scissors in a wild west setting, available in single player and multiplayer versions. The amount of text to be localized is moderate to high, with the most words appearing in the shop and the instructions. When we analyzed the source code, we found that the game was not at all prepped for localization; user-facing text is found in many different parts of the code, not just scripts, and most of the words the player sees are hard-coded as images. This meant that internationalizing by wrapping strings and externalizing the text would not be enough, so I took responsibility for creating images with corresponding Japanese, English and German text via desktop publishing (DTP) in Photoshop.

Read more

Advanced Computer Assisted Translation 2019: Portfolio

What is Microsoft Translator Hub?

Our big project this semester was using Microsoft Translator Hub to build our own machine translation system to translate museum guides from English to French. The Hub uses statistical machine translation to produce target documents similar in style and tone to the sample documents provided. (Microsoft released a preview neural machine translation option this month.)

How Does It Work?

A lot of data is required to train the machine. We uploaded pairs of documents that have been published in English and French versions. The Hub can handle pdf files, but the segmentation can be off. In later rounds we often converted the pdfs to .txt files to check alignment and delete information that was different between languages, for example, if the guided tour in English meets in the atrium at 3 while the tour in French meets at coat check at 1:30. Finding and cleaning appropriate data was very time-consuming. The Hub suggests a minimum of 10,000 parallel sentences of training data, 2,000 of tuning data and an additional 2,000 of testing data. The testing data is what the machine produces after a round of training. It displays the source sentence, the parallel target sentence provided in the data, and the target sentence produced by the machine. The machine compares the parallel target sentence provided to the one it produced to produce a BLEU score.

 

Read more

Translation Management Systems 2019: Portfolio of Work

Group Translation Project in Worldserver

To gain experience using translation management systems, my group and I translated an article in the San Francisco Chronicle about Stephen Hawking into three languages: French, Chinese and Japanese. We set up the project so that we would all be using the same workflow and reference files. Our proposal, deliverables and conclusions are available below.

Proposal

Deliverables

Presentation

Read more

How to Choose and Implement a TMS

If you’re part of a company that does a lot of translation and transcreation and are considering picking out your first translation management system (TMS),there are a lot of factors to consider. These include how much you want to automate your process, process management, how centralized you want your system to be and who should have access to it, integration with your current content management system and machine translation, access to linguistic resources (TMs, glossaries), options for vendor management, and of course analytics and reporting capabilities so you can track the status of ongoing projects and see what’s working and what can be improved.

Read more

Machines – Translation – Metrics – Quality – Art

Until fairly recently, evaluating translation has been like evaluating art: most people agree that some art is better than other art, but there are few hard-and-fast criteria for objectively evaluating quality. This is also true for machine translation- how can we consistently and objectively evaluate machine translation output across various levels of quality and style? It’s not enough to say that one is simply “better” or “worse”, because that doesn’t give us any useful information about how to improve quality.

Here are some examples from art. On the left is a mural of two hands on the side of a building in Berlin. On the right is a set of two photographs of a hand hanging in the Münchner Stadtmuseum. Which one of these two images is better art? Why? You likely have your preference, but what are the determining factors? Is black and white better than color? Is photography better than paint? Is a small print better than a large mural?

Berlin          Münchner Stadtmuseum

Read more

How to Write Better Short Papers in 6 Easy Steps

In undergrad, I minored in political science. My senior year I had a political theory class, and every week the professor expected a two-page critical response to the assigned reading. It was dense stuff, from Plato to Emma Goldman, and a lot of my classmates were tearing their hair out every week trying to complete the reading and then write an intelligent response to it. I didn’t want to spend more time on this assignment than I had to, so I came up with a system, which I’m sharing with you.

Iris’s 6-Step Guide to Writing Short Papers

  1. While reading, write down (or copy and paste into a document) quotes that jump out at you for any reason. Include page numbers.
  2. When you’re done with your reading, look through the quotes you’ve written down.
  3. Identify the general theme among the quotes you’ve selected. Read more