Multilingual presentations: translating and narrating slides

With a presentation, the slide text comes to mind first. But there are further layers that are easily forgotten: labels in charts and diagrams, text in images and screenshots, speaker notes, the narration, captions, file names and metadata such as title and description. Only when all layers are right is a presentation truly multilingual.

The best translation starts with the design of the original slides. Text should be real text, not an image. Graphics should be built with text layers, ideally as vector graphics whose labels can be replaced. Text boxes need spare room, because translations are often longer than the original; languages such as Chinese or Japanese, on the other hand, often need less space but larger type for good legibility.

Avoid puns, idioms and cultural allusions that do not carry over. What is witty in one language is often incomprehensible in another.

Language models now translate presentations quickly and well. They usually grasp the context of a slide and choose suitable terms. Sources of error remain, however: technical terms, internal names, product names, legal wording and number formats.

A glossary of fixed terms helps the machine as much as the people who check the result. Have at least the key messages and technical terms reviewed by someone who knows both language and subject. For presentations with legal effect, such as statutory briefings, this review is indispensable. It is also important that text that has been reviewed once is not overwritten in the next translation run.

Narrated presentations need a separate audio track per language. With synthetic voices, it is generated from the translated narration without booking speakers. Make sure the voice suits the language and pronounces technical terms correctly. If an avatar appears as the speaker, it should also speak other languages with its own assigned voice.

Producing many language versions costs computing time. An economical approach is to pre-produce only the first slides in each language and the rest only when someone actually uses that language.

Languages such as Arabic and Hebrew are written from right to left. On slides this means right-aligned text, reading from the right, and mirrored arrows and timelines. Mixed content, such as Arabic text with Latin abbreviations, needs particular care.

Localisation goes beyond language: date formats, number formats, currencies, examples and images should suit the target audience. An example from German employment law is of little help to employees in another country.

For live talks to an audience with mixed languages there are further options: slides in a shared working language with captions or interpretation of what is said, bilingual slides with a main and second language, or slides provided in several languages to follow on one’s own device. Bilingual slides fill up quickly; they suit titles and key messages better than detailed content.

Before a translated presentation is used, someone should look through it completely in the target language, ideally someone from the target audience. Look not only at language but also at cut-off text, overlapping elements, wrongly mirrored graphics and pronunciation errors in the narration. A short list of the most common errors per language helps next time.

In Universe the author chooses “full”, “beginning” or “off” for each language; in “beginning” mode the first slides are ready as video and the rest is produced while the learner watches. Narration, captions and interface are available in fifteen languages, including Arabic and Hebrew with a mirrored interface. Explainer graphics are created as text rather than pixels, so their labels are translated in one pass with the slide text. A translation can be created as a separate version of the course; the original remains unchanged.