Welcome to Karupora!
Can you help us translate Rashōmon into English?
Rashōmon (羅生門) is a short story by Ryūnosuke Akutagawa (芥川龍之介) based on tales from the Konjaku Monogatarishū (今昔物語集). The story was first published in 1915 in Teikoku Bungaku (帝国文学). The story recounts the encounter between a servant and an old woman in the dilapidated Rashōmon, the southern gate of the then-ruined city of Kyoto, where unclaimed corpses were sometimes dumped. Akira Kurosawa's film Rashomon (1950) takes only its name and some of the material for the frame scenes, such as the theft of a kimono and the discussion of the moral ambiguity of thieving to survive, from this story.
What is Karupora?
Although there are over 1 billion bilingual and polyglot individuals around the world, their talents are largely not utilized in translation work because the level of sophistication required for the translation process demand high level of understanding in lexical and grammatical nuances of two languages. Therefore, translation has largely remained in the domain of professional translators. Thus, translation of a large literary work can be costly and time consuming for those who requested the work.
Karupora aims to tap into the talents of millions of bilingual and polyglot individuals around the world and transform the existing paradigm in the domain of translation. Using the collective intelligence of the crowd, Karupora will reduce the time and the financial cost of translation while demonstrating the crowd's capability of producing viable, quality translation.
While many popular international literary works have been translated into English and other widely spoken languages, much of the world's literatures have never been translated into another language. Either due to lack of financial interest or due to lack of bilingual talents in certain rare languages, consumption of these literatures remain confined to audiences who speak and read that language natively.
Karupora will provide a platform to translate literary works that have never been exposed to the international audiences. In addition, Karupora will serve to help preserve rare languages that are in steady state of decline in usage by building a large volume of translation work in them.
Despite the progress that have been made in machine translation, a truly viable machine translation system remain elusive. Moreover, in order to train machine translation algorithms, computer scientists must collect large volume of parallel corpora (texts) that is translated and matched line by line. For some language pairs such as English and French, large volume of parallel corpora already exists. However, large volume of parallel corpora for other language pairs are still hard to come by for computer scientists because building parallel corpora can be time consuming and difficult as the process of translation itself.
Your translations contributed through Karupora will be used to build parallel corpora in multiple languages. Today, Karupora only exhibits one Japanese short story, Rashōmon, in order to demonstrate Karupora's potential. Our ultimate goal is to expand our offering to hundreds of literary works in multiple languages including Chinese, Arabic, French, etc. as well as those in rare languages such as Navajo. Your contributions will enable computer scientists to build powerful machine translation systems, which will further reduce the cost of translation in the future.
How does it work?
Karupora is a crowdsourcing game that brings together thousands, and hopefully millions, of individuals with diverse language skills and proficiency levels to collaborate together in the translation process. Translating an entire literary work can be a time consuming and difficult process for one individual; this is why Karupora asks you to contribute one line of translation at a time, at your convenience.
Karupora will serve you a randomly selected sentence from the literary work that you are translating as well as existing versions of translation. Your task is to vote on either one of the existing versions of translation or provide your own version.
After you vote, Karupora will randomly select another user who previously translated the same sentence. If it determines that the randomly selected user also voted on the same version of translation, then you will receive a point greater than or equal to 1 according to its probabilistic algorithm. If not, your score will be 0.
You may receive a higher score if the version that you voted is less popular (had less votes in the past) than the others; on the other hand, the probability of matching with another user who voted for the same version will be lower. If you select the version that is more popular, your score may be lower if matched, but your probability of being matched will be greater. Therefore, pay attention to the number of past votes in the choices. Nevertheless, your best bet is to select a version that you find to be the most accurate translation.
You may choose to submit your own version of translation if you are not satisfied by other versions available. In such case, you will be awarded 1 point. However, for the extra effort of making your own translation, Karupora will reward you additional points equivalent to the points received by another user if he or she votes for your version. Thus, for the additional effort up front, you can wait and see your score go up as others vote for your version.
Why does it work?
A bottleneck in large scale crowdsourcing platforms has been the cost of verifying user contributions by the platform managers. Moreover, crowdsourcing platforms that fail to correctly align the incentive of the contributors with quality of their work yield suboptimal contributions. Worse, malicious contributors ("trolls") can intentionally cloud the platforms and ruin the system for other honest contributors.
Karupora utilizes the latest algorithmic mechanism, Bayesian Peer Prediction Mechanism, designed by students and faculty from Harvard University. This mechanism ensures that honest answers by a user will maximize his or her score; therefore, a user's best course of action is to report his or her true belief about the correct version of translation.
If you are interested in the technical details of the Dirichlet Process Mechanism, feel free to look at this page that discusses history of peer prediction mechanisms and proof of truthful (or strictly-proper) property of Dirichlet Process Mechanism: Keeping the Crowd Truthful: Mechanism Powering Karupora's Translation Game