Prepare
Multilingual texts are translated and normalized through one reproducible workflow.
RESEARCH PROJECT · MULTILINGUAL NLP
Can personality
survive translation?
This project asks whether personality signals carried by text can be measured consistently when language changes—and how we can tell when the measurement is trustworthy.
ENFAESFR
SECSR5D
THE IDEA
One construct. Several languages. A fair comparison.
Text carries more than information: it can feel sincere, exciting, competent, sophisticated, or rugged. Comparing those signals across languages is difficult because translation may preserve the literal message while subtly changing its tone. The project turns that concern into a measurable research problem.
THE METHOD
A validation-aware pipeline, not a black box.
Validate
A human-authored benchmark tests whether the translated text preserves the personality signal needed by the model.
Model
A RoBERTa-based classifier represents each text across five established personality dimensions.
Compare
Five-dimensional profiles are aggregated and compared with cosine similarity.
1,680human-authored benchmark sentences
38 / 42traits exceeded 0.90 average similarity
THE CONTRIBUTION
Making multilingual comparison more credible—and more useful.
The main contribution is a scalable workflow that treats translation quality as something to test, rather than assume. It combines multilingual processing, transformer-based classification, and an interpretable similarity measure so researchers can compare personality profiles across languages while retaining a clear validation trail.
The broader idea is simple: if an analytical system crosses languages, validation should cross them too.
FROM RESEARCH TO PRACTICE
Is the personality you communicate the one people perceive?
Across languages and markets, small differences in tone can become meaningful gaps between intended identity and audience perception. This framework can help brand owners:
- compare intended and perceived personality profiles
- identify differences across languages and audiences
- detect communication drift or inconsistent positioning
- evaluate whether localisation preserves the intended tone
- turn multilingual text into evidence for communication decisions
Exploring a similar question?
Discuss your brand ↗If you are responsible for a brand and want to understand how its personality is expressed or perceived across languages, I am open to selected advisory and collaborative projects.