Lucitext is a small research and development studio. We build natural language processing, speech and large language model systems with care for the languages and the people they serve, especially less-resourced languages.
Each project starts from the language and the people behind the data. Then we choose the simplest method that holds up in real use, and we say so when a smaller tool would serve you better.
Text analytics, classification, sentiment and emotion analysis, information extraction, search and summarisation, in English, Turkish, French and languages with far fewer resources.
Delivered: a suite of more than thirty text tasks, from intent detection to emotion recognition; extraction for a task platform, maintained over four years; multilingual libraries written from scratch where nothing existed.
Recognising emotion in speech, preparing and annotating speech data, and adapting multilingual speech models to build voices for languages that have few.
Delivered: an emotion-aware Turkish speech model, from collecting and labelling the data to production.
Retrieval-based assistants, structured extraction from documents, training and evaluation data, and honest testing of how well a model serves its users, including in low-resource languages.
Delivered: assistants on local models that answer from an organisation's own documents, with no data leaving its servers.
Recommendation, ranking and forecasting, from automated model selection to monitoring and retraining in production.
Delivered: a multi-tenant ranking and prioritisation service for a software product that monitors and retrains itself automatically.
Alongside client work, we do research in computational linguistics, asking what today's language technology misses, and for whom.
How general-purpose models treat languages with little data, and what thoughtful adaptation can change.
Recognition and synthesis for languages with little recorded speech, and the patient data work that makes them possible.
How feeling shows up in what we say and how we say it, across languages.
Lucitext is an independent research and development studio in Istanbul, founded by Dicle Öztürk in 2020. We work remotely with teams in Europe and beyond, on projects that need both research depth and working software.
Our working languages are English, French and Turkish, and we have built language technology for many more, Arabic among them. Projects usually begin with a short, well-defined first phase, so both sides can see whether the fit is right.
For consulting, research collaborations and projects, small or large, you can contact us. A few lines about your language, your data and what you hope to do are a good start. We are happy to begin with a short call.
info@lucitext.ioIstanbul, Turkey