The language you speak may shape how a conversation unfolds. Languages that squeeze more meaning into each word are spoken faster, and their conversations tend to circle a narrower set of ideas in more depth.
Pedro Aceves of Johns Hopkins University and James A. Evans of the University of Chicago and the Santa Fe Institute reported the findings in Nature Human Behaviour, published online on February 16, 2024.
Measuring how much a word carries
The authors started with the observation that languages carve up meaning differently. In English, “play” covers soccer, board games and the violin. Spanish uses one verb for games and another for instruments, so the English word carries more conceptual information.
To measure this at scale, the researchers used 18 parallel translation corpora containing the same text in English and other languages, totalling 14 billion tokens across 998 languages from 101 language families. They compressed each translation with the Huffman coding algorithm and compared its size with the English version. Languages needing fewer bits for the same content count as denser.
Density varied widely between languages and was consistent across subjects as different as religion, banking, medicine and film. Denser languages also had denser semantic spaces in word embedding models: a one unit increase in information density went with a 1.02 unit increase in conceptual density.
Faster speech, tighter topics
To test speed, the team timed audio recordings of the Bible in 265 languages, the same message in each. Denser languages took less time to say it.
Next they analysed more than 5,800 real conversations in 14 languages, using embeddings to measure how much conceptual ground each one covered. After accounting for length and number of turns, conversations in denser languages stayed in a narrower area and drifted less from their opening topic. A one standard deviation increase in density went with a change of about two-thirds of a standard deviation in conversational breadth.
The same pattern appeared in Wikipedia. Among more than 90,000 articles written collaboratively in 140 languages, those in denser languages covered narrower conceptual terrain.
What it does not show
The authors stress that the design is observational and cannot prove that language density causes these conversational patterns. They suggest laboratory experiments across smaller sets of languages as a next step.
They also noticed that the most widely spoken languages sit close to the middle of the density range, which they read as a hint that large modern languages may settle near an optimal level of density.
Study Details:
- Title: Human languages with greater information density have higher communication speed but lower conversation breadth
- Authors: Pedro Aceves, James A. Evans
- Journal: Nature Human Behaviour
- Publication Date: February 16, 2024
- DOI: 10.1038/s41562-024-01815-w