Samsung Electronics’ Global Research & Development (R&D) Centers are persevering with to trailblaze of their analysis within the subject of synthetic intelligence (AI). Following the granting of a number of international AI awards and business recognition to Samsung researchers across the globe, researchers in Poland and China just lately received a set of extremely prestigious international AI challenges.
Spearheading Speech Translation Research
Samsung R&D Institute Poland and Samsung R&D Institute China-Beijing competed with a number of the world’s prime universities and analysis labs to win first place in two separate challenges on the International Workshop on Spoken Language Translation (IWSLT), one of many world’s longest-running workshops on automated language translation. This yr, IWSLT joined the Association for Computational Linguistics convention (ACL), a premier convention within the subject of computational linguistics, to cowl a broad spectrum of analysis areas which are involved with computational approaches to pure language.
For the Offline Speech Translation activity, which assesses the interpretation of TED talks from English to German, Samsung R&D Institute Poland received first place for the second time with its personal analysis capabilities in audio to textual content translation. The conferral of this award marks the fourth consecutive yr that groups from Samsung R&D Institute Poland have taken first prize in IWSLT challenges, together with earlier years’ textual content translation duties.
This yr’s Offline Speech Translation activity allowed members to submit programs primarily based on both the standard speech translation pipeline system composed of an automated speech recognition (ASR) and a machine translation (MT) or an End-to-End (E2E) system. Samsung R&D Institute Poland’s system relies on a single encoder-decoder deep neural community – an E2E system – able to each English and German texts.
In computational linguistics, E2E programs are harnessed to unravel the widespread drawback of error accumulation, whereby, in a conventional pipeline, an error within the speech recognition section can result in a nonsensical translation. However, analysis from over the previous three years has proven that conventional programs have always been outperforming E2E speech translation programs. The Samsung crew’s system not solely positioned first within the E2E class, but additionally outscored all conventional pipeline system entrants, a exceptional achievement that places Samsung R&D Institute Poland on the forefront of speech translation analysis.
The crew from Samsung R&D Institute Poland Team who participated on this yr’s IWSLT challenges
Innovative Approaches within the Field of Computational Linguistics AI
Samsung R&D Institute China-Beijing took half in a second problem, the Open Domain Translation activity evaluating Japanese to Chinese translation functionality, in the end taking first place. The fundamental targets of this activity have been the promotion of analysis into translation between Asian language, the exploitation of noisy parallel internet corpora for machine translation and the considerate dealing with of information provenance.
Samsung R&D Institute China-Beijing submitted a system primarily based on Transformer mannequin structure and adopted the relative place consideration. The crew centered on bettering the Transform baseline system with elaborate knowledge preprocessing and managed to attain important enhancements. The crew additionally tried shared and unique phrase embedding and in contrast totally different granularity of tokens, approaching the method at a sub-word stage, together with Byte Pair Encoding (BPE) and Sentence Piece. Large-scale again translation on monolingual corpus was used to enhance the Neural Machine Translation (NMT) efficiency.
Members of the crew from Samsung R&D Institute China-Beijing Team who participated on this yr’s IWSLT challenges
Achievements in AI Audio Signal Interpretation
In addition to their first-place end within the IWSLT…
