Neural Machine Translation (NMT) represents one of the most advanced forms of machine translation available today. Powered by neural networks, it has revolutionized how we understand and utilize machine translations.

However, like all technological advancements, it is not without its challenges. One of the core issues lies in the fact that while NMT can generate impressively fluent translations, those translations do not always accurately convey the original content's intent. To address this concern, NTT has pioneered a new "post-editing" technology.

NTT's neural machine translation technology has a number of potential commercial uses, which include:

The essence of NMT lies in its foundation—neural networks. These intricate systems have greatly improved the accuracy of machine translations, pushing the boundaries of what we previously believed possible. Nevertheless, there is a lingering concern, which is that neural networks, despite their advancements, cannot guarantee a complete absence of errors. This limitation becomes particularly noticeable in specialized fields like medicine and patents, where even the smallest translation error can have significant repercussions.

That's where NTT's "post-editing" comes in. Recognizing the occasional misalignments between the original and translated content, NTT has developed a cooperative solution involving both humans and machines. Once NMT completes its translation, human experts then step in, identifying and fixing any discrepancies. To streamline this process, NTT has crafted a method to precisely align words from the source text with their counterparts in the translated version. This strategic alignment effectively pinpoints areas where the machine translation might have deviated from the intended meaning.

A key objective for NTT is user-friendliness. The company aims to evolve the post-editing mechanism to be as intuitive and user-friendly as today's spell checkers. To that end, they have tested their approach carefully. Utilizing the "Quality Estimation Task datasets in WMT-2020," NTT evaluated 8,000 sets—comprising original sentences, their machine translations, and the subsequent human post-edited versions. To aid in error identification, each word within these sets was categorized as either well-translated or poorly-translated. Furthermore, a subset of these, approximately 1,000 sets, underwent manual word-level alignment. The data served as a foundation, with the majority being used for training purposes and a fraction reserved for testing the technology's accuracy.

By combining the power of machine translation with the expertise of human editors, NTT's post-editing technology addresses the limitations of NMT and offers a vast array of potential applications across industries. NTT's vision for the user interface is clear: human post-editors refining machine-translated content, supported by precise word alignments and quality tags. Bridging the gap between source and target texts, and ensuring that human editors can seamlessly perfect translated content.

NTT—Innovating the Future of Communication

Daniel O'Connor

Daniel O'Connor joined the NTT Group in 1999 when he began work as the Public Relations Manager of NTT Europe. While in London, he liaised with the local press, created the company's intranet site, wrote technical copy for industry magazines and managed exhibition stands from initial design to finished displays.

Later seconded to the headquarters of NTT Communications in Tokyo, he contributed to the company's first-ever winning of global telecoms awards and the digitalisation of internal company information exchange.

Since 2015 Daniel has created content for the Group's Global Leadership Institute, the One NTT Network and is currently working with NTT R&D teams to grow public understanding of the cutting-edge research undertaken by the NTT Group.