Earliest times

The first thing that has to happen in order for linguistics to be a remotely plausible enterprise is that language must be available in permanent form. By about 3000 B.C. this had happened for Egyptian hieroglyphics, as well as other written languages. In most cases the texts were incidental by-products of commerce and trade. Reliable recordings of music and speech had to wait until the late 19th century, or mid-twentieth century if high-quality reproduction is required. Data availability is a prerequisite for many forms of scientific endeavour: For example:

• Indonesia has undoubtedly had a rich literary tradition, but the climate makes it highly improbable that paper documents will survive for very long. This limits the potential for diachronic literary studies.
• Nobody really knows how Classical Greek was pronounced. While it is possible to make inferences from contemporary descriptions of the language, from the patterns found in poetry, and from the spellings of evidently onomatapoetic [check spelling] words, there are many areas in which doubt must remain.
• Little can be said about the acoustics and physiology of the European-trained castrato voice. There is one early recording of a singer from the Papal chapel, recorded in [whenever] when he was well past his prime. While this may be of some interest, one sample is no basis on which to draw any but the most cautious inferences about such singers. Early music specialists constantly face the problem of making intelligent inferences from many different grossly incommensurable sources, and must learn to live with the inevitable uncertainty. This problem of exploiting limited partial information arises again and again in data-intensive linguistics.

By around 1000 B.C there was a substantial body of authoritative texts which we now recognise as the Hebrew Bible. Note that this body of text is a more or less closed colection of texts imbued with particular authority, and that major social engineering would be necessary to add or subtract anything. Texts which are like this are usually termed ``canonical''. This is in marked contrast to a public library, where the contents are open-ended and constantly changing. An important question for data-intensive linguisticsis whether language, seen as the object of study, is more like a canon, more like a public library, or even more like the sort of chat which you randomly overhear on a bus.