Microtalks


Track: Inside Track | IN9 | MICRO TALK |
Friday, June 7, 2024, 3:45pm – 4:30pm
Held in: Room 6
Presenters:
Sharon Biggar - Happy Scribe 
Ben Cornelius - Cornelius Communications 
Maria Emilia Maisterrena - Keylingo 
Ady Namaran Coulibaly - Bolingo Consult 
Marina Pantcheva - RWS
Host: Gary Lefman

In this session, we will offer several quick and dynamic presentations covering topics and ideas too interesting to ignore!

Synergy in the AI Era: Evolving Organizational Structures — Maria Emilia Maisterrena (Keylingo)
In alignment with the ever-evolving technological landscape, particularly led by AI, organizations must develop resilient organizational structures, teams with adaptable skill sets, and dynamic work methodologies. Leveraging insights from Keylingo’s organizational experience, this microtalk underscores the transition from traditional project managers to dynamic roles such as directors, relationship managers, and production and solutions leads. Discover how forward-thinking companies are redefining their workforce, creating synergy between hybrid job positions and the transformative potential of AI without compromising human expertise and employment. For example, production and solutions leads, emphasizing the need for adaptable, mixed teams. The objective of the presentation is to discover how forward-thinking companies are redefining their workforce, creating synergy between hybrid job positions, and embracing the transformative potential of AI without compromising human expertise and employment.
Takeaways:  AI Encourages Skill Evolution, Organizational Adaptation

Using AI and IPaaS to Power Scaled Global Releases Ben Cornelius (Cornelius Communications)
This presentation delves into the future of global releases, transforming the way global content is produced, delivered, and supported. Key features include a fully automated process powered by AI and IPaaS that harnesses LLM and an internal information database, making knowledge available to developers, writers, and translation teams via Slack. All are powered through the translation workflow. Product content becomes a valuable resource for training and support teams. The resulting multilingual bot benefits internal developers, support teams, and end-users, delivering a unified experience across product, help, and support UIs. This self-updating cycle ensures a seamless global content deployment and user support experience with each release.
Takeaways: How existing and emergent technologies can be combined to scale up Global releases. Not pie in the sky, but real application that is available now.

Aligning LLM and Human Assessments of Translation Quality Marina Pantcheva (RWS)
One of the most useful applications of large language models (LLMs) lies in their ability to assess the linguistic quality of translations (LQA). In this talk, the presenter will describe their research findings from extensive explorations of LLM-powered LQA. She will discuss effective prompting strategies to improve response accuracy and present methods to increase the alignment between the assessments of LLMs and those of human reviewers. The talk centers around what factors influence the performance of LLMs in quality evaluation. It addresses the risks and advantages LLM brings, highlighting specific behaviors that must be considered for their effective implementation.
Takeaways: The attendees will learn (1) what prompt strategies increase the success of LLM-powered LQA evaluations; (2) what risks are associated with LLM-powered LQA; (3) where LLMs do really well in LQA tasks and why.

African Languages Online: A Case Study of Twi and Fante — Ady Namaran Coulibaly (Bolingo Consult)
Most African languages are considered “low resource”, indicating that these languages lack enough documentation online and offline. This presentation will focus on lessons learned from an initiative to boost the presence of African languages online. Specifically, it will shed light on the first ever news and information website in two language varieties of the Akan language, Twi and Fante, spoken by the Ashanti ethnic group in Ghana.
Takeaways: Several African languages are more spoken than they are written. Content production and localization for low-resource languages contribute to enhancing the digital presence of these languages and also boost confidence in these languages.

The Importance of Not Being Frank — Discovering and Solving Speech-to-text AI Biases — Sharon Biggar (Happy Scribe)
Automated speech recognition (ASR) technology revolutionizes audiovisual projects, enhancing transcription precision. Despite benefits, inherent biases impact accuracy. This session explores biases, from dialectal variations to socio-cultural influences, and their repercussions. Proactive solutions involving rigorous data preprocessing and model training are advocated, aiming to detect, measure, and mitigate biases. Diversified dataset collection and algorithmic fine-tuning are highlighted for equitable speech recognition. An open forum invites collaboration among industry experts to develop ethical ASR solutions, promoting fairness, inclusivity, and reliability.
Takeaways: Identifying biases in ASR: Dialectal and socio-cultural biases persist, affecting accuracy; Implementing mitigation Strategy: A holistic approach includes rigorous data preprocessing and diversified datasets; Fostering collaborative Development: Experts collaborate for fair and reliable ASR solutions