Google I/O 2022 — the corporate’s annual developer convention — is underway, hours after the keynote occasion was held on Wednesday. In the course of the occasion, Google introduced a variety of updates and enhancements to its services and products, together with Google Maps, Android, Google Workspace, Google Meet, YouTube, and the Google Assistant. In the course of the keynote occasion, the corporate additionally introduced that Google Translate was including help for twenty-four new languages, supported by machine studying. Nevertheless, customers have identified that the corporate’s slide on the occasion had a number of errors for various languages.
In the course of the occasion, a display behind Google CEOlisted among the languages added to Google Translate because of the corporate’s breakthrough machine studying (ML) know-how known as Zero-Shot Machine Translation. The function permits the ML mannequin to study a brand new language with out ever seeing a direct translation of it, in line with the corporate. In the meantime, Twitter person Sam Ettinger (@DHammarskjold), identified that languages that weren’t Latin- or Cyrillic-based have been incorrectly written on the display. A screengrab was additionally shared by person Rami Ismail (@tha_rami) highlighting one of many errors.
I made a useful annotation for everybody. Each single one which’s not Latin- or Cyrillic-based is unsuitable (a minimum of a bit bit).
I struggled a bit with displaying the Bengali script, however I made it work higher than did this “Google” startup, no matter that’s
— Sam Ettinger (@DHammarskjold) Could 11, 2022
In a picture with annotations of the incorrectly written slides, Ettinger revealed that the phrase “Sudanese dialect” has the phrases disconnected, whereas the structure is left-to-right. Saraiki, however, a language spoken in components of Pakistan has disconnected letters and written as left to proper on the slide. The phrase ‘Marwari’ can be incorrectly displayed on the display, with the nukta and the final vowel moved to the correct, whereas the phrase for “Lhasa Tibetan” has the vowel and subjoined h/okay letters moved far to the correct, in line with Ettinger.
In line with a response by one other person @rdorje_aabrug to Ettinger’s put up, the visible glitch for one of many languages may have been brought about because of the Noto Sans Tibetan font on Google Docs that causes shifted consonant stacks and vowels when Docs creates a picture from the typeface for a PDF, or a slide throughout a presentation. The person states that an oblique repair for the difficulty displaying the Tibetan font incorrectly is to make use of a brand new font known as Jomolhari as an alternative of the default Google Noto Sans Tibetan font.
Sadly, it seems that Google’s keynote on the occasion doesn’t present the identical slide that was current through the livestream of the occasion, that was shared by customers on Twitter. As an alternative, an up to date display is displayed with some adjustments. Nevertheless, IHNS was in a position to verify that the slide shared by customers was certainly current through the livestream of the keynote, and never the one that’s at the moment accessible to customers on Google’s YouTube channel.