== Small word network ==

English words are connected by just three degrees of separation.
2 July 2002

PHILIP BALL

Word association can link just about any two common words in the English
language using an average of three steps, says a team of scientists in
Arizona.

The semantic links between English words make the thesaurus a 'small world',
much as the network of human social interactions connect us all by six degrees
of separation, find Adilson Motter and colleagues at Arizona State University
in Tempe1. The researchers expect languages other than English to have the
same properties, even if their syntactic structure is very different.

The researchers traced the links between 30,000 English words in an online
thesaurus. For example, the word 'actor' can be connected to 'universe'
through two intermediaries. The thesaurus lists 'character' as a synonym for
'actor'; 'character' is also equated with 'nature'; and 'nature' with
'universe'.

Moving from 'actor' to 'universe' in the network of words therefore takes
three steps. To the surprise of Motter and colleagues, they found that the
same was true of just about any randomly chosen pair of words in the
thesaurus. The English language, in other words, enjoys just three degrees of
separation.

Obviously this isn't true for all words - 'vanadium' and 'ambidextrous', say,
might be more difficult to connect. But the researchers included in their
network only 'root' words, which themselves warrant a primary entry in the
thesaurus. Many of the synonyms listed for a given root word are not
themselves root words - they don't appear elsewhere in the thesaurus. Root
words tend to be the most common or generic - 'big', say, rather than
'gargantuan'.

A small-world network isn't defined solely by the property any two nodes are
connected by a small number of steps. Nodes must also be highly clustered -
that is, if two nodes are both linked to a third, there must be a high
probability that the two are also directly linked. Friendship networks are
highly clustered - if you and I share a common friend, we probably know each
other too.

Motter's team show that the language network has a high degree of clustering.
For example, 'nature' is listed under the entries for both 'universe' and
'world', and 'world' is also listed under 'universe' (and vice versa).

The researchers think that the network structure of a language probably has
its origins in the nature of cognition and memory. It is not surprising that
language is highly clustered, as we remember things associatively - by
grouping similar concepts together. This makes it easier for us to retrieve
related words from our memory.

The small-world network also means that apparently quite different concepts,
such as 'actor' and 'universe', are closely linked by a short series of
semantic steps. This, say the researchers, makes it easier for us to carry out
mental searches when using language - we can get to our intended destination
quickly, regardless of our starting point. A database cross-referenced in this
way would be relatively easy to search computationally.

According to this argument, language has evolved its current structure because
it is easy to use. Words acquire multiple meanings because that makes the
network highly interconnected, so that the average path length between two
words is small. It remains to be seen whether searching our memories for a
particular word really entails wandering mentally along links in the network.
References

1. Motter, A. E., de Moura, A. P. S., Lai, Y.-C. & Dasgupta, P. Topology of
   the conceptual network of language. Physical Review E 65, 065102, (2002).


Nature News Service / Macmillan Magazines Ltd 2003