Whitepaper

Data science in localisation: practical applications for multilingual content

How natural language processing, sentiment analysis and incrementality testing make multilingual content more effective, and why human linguists remain essential to getting it right. With original data from Locaria’s own accounts.

Read the whitepaper

Cover artwork for "Data Science and its Practical Applications in Multilingual Content", showing a spiral tunnel of words in many languages

+13%

More SEO sessions from a winning title-tag syntax, at 99.4% significance

650K+

Organic keyword positions analysed across six European languages

39.9%

Of searches used three-word keywords, the most common length

1.1%

Of searches used a single word

Three practical applications

Smarter keyword management

Techniques such as n-grams, stemming and lemmatisation expand and clean multilingual keyword lists, search query reports and negative keyword lists at scale.

Long-tail search insight

People rarely search in single words, and each language behaves differently. German searchers often drop articles: ‘Sneaker schwarz Damen’ is searched far more than ‘Der Sneaker’.

Incrementality testing

Controlled tests across matched product groups show which content changes actually cause performance gains, rather than relying on correlation.

Why people still matter

Data scales language work. People make it accurate.

Machine translation, keyword filtering and sentiment analysis all depend on training data, and English is far better served than any other language. Translating content into English before analysing it compounds errors, first in the translation and then in the analysis.

The report concludes that a human-in-the-loop process, using machine learning to pre-process and filter content with expert input throughout, is the only way to achieve almost 100% content accuracy.

Inside the title-tag test

Locaria’s data science team tested new page-title syntaxes across four matched groups of product pages, each with its own control group, and compared the results with a forecast over several weeks. Two syntaxes drove significantly more traffic. The winner delivered 13% more sessions than the old syntax, at 99.4% significance. The same method extends to cross-channel questions, such as how pausing paid search affects organic results.

Published in 2022, before generative AI went mainstream, the report’s view of AI’s creative limits reflects that moment. Its testing and data principles remain current.

Read the whitepaper

Explore our multilingual SEO and language technology.