"And then, Client and Vendor fused in a long, passionate kiss - Charms and tricks to get out of the friend zone" - Jose Palomares (Director, Localization | Coupa Software)

Do you ever feel like your client-vendor relationship is stuck in the friend zone? You know, where everything is pleasant but about as exciting as waiting for a PO to arrive? Alright! This talk aims to turn your cordial handshake into a high-five with sparks, and that boring QBR with charts no one believes into something to talk about at conferences. It's time to spice up that client-vendor romance. Roarrr.

The speaker will talk from his own experience in building partnerships that last for years, the ones you bring with you when you change jobs, not because you don't know better, but because it doesn't get any better. You'll hear about the art of 'Empathy espionage' – getting into your counterpart's head (in a non-creepy way) and discovering their deepest desires. You'll also hear about 'Soft-touch small gestures', ways to massage the relationship and create low effort, high impact touchpoints that will add so much value that no one will notice that big red bold KPI blinking like DEFCON 2 on the slide.

If you're ready to take your client-vendor relationships from 'meh' to 'marvelous', join us for a session that's part strategy, part stand-up comedy, and zero regrets.

"Death of TM - or. Long Live the TM?" - Olga Beregoyava (VP of AI | Smartling)

For years our industry has been relying on TMs as the ultimate source of truth, adhering to the source-target one-to-one correlation format. It made a lot of sense as we got the perfect alignment for previously translated content matching, and also developed a perfect corpus for training several generations of Machine Translation engines. And here comes the question: do we even have the source as we have known it, and do the advancements in Generative AI support the idea of source equal target translation units. Our statement is: the TMX format that has been serving us for years must now give way to non-parallel data, and the accompanying metadata and database structure will as well be changing to “content block in one language mapping to a content block in another”, with no segment-based correlation. The value of this parallel data for future reuse is much more about fine-tuning LLMs on the most representative - and much smaller - subsets of the translated data, and using translations for few-shot prompting and RAG. So, shall we say goodbye to the TMX-based TM and embrace the new generation of multilingual corpora?

Host organization: Globalization and Localization Association