Chapter 2: Sources of Error

As all scientists know, an experiment are only as good as its control of error. Every experiment has sources of error which have to be accounted for. A source of error is only fatal to the results when it is unknown. Compiling a wordlist is not an exact science. The larger the sample, the more reliable the categories, but, since the size of a list is theoretically unlimited it is always possible that this view will change when we include more words or move the focus from one geographical area to another. So far, as a rule of thumb, a word-list of 800 words is large enough to be stable and representative. It is large enough to withstand the removal of a few problem words. The edges of any list and its internal categories will always be slightly fuzzy, no matter how exact we are, but this can be regarded as a feature of language.

Nichols lists four obstacles to the comparative lexical method: affixal pronouns, monophonemic roots, lack of citation forms, and bound inalienable possession. We are, fortunately, not involved in comparative lexicology but prefixes are a hazard in any study that relies on the structure of words. Generally they show up in a dictionary, if the dictionary is large enough, and can be avoided, even if the language is unfamiliar. A solution is to avoid any affixing or head-marking languages.

A further cause of potential error is the tendency of languages to evolve by combining ‘suffixes and prefixes and fixes in the middle’, as one writer puts it. T. Anderson, Bread and Ashes, 81, Nichols notes that the Circassian languages of the west Caucasus have possessive prefixes which in certain circumstances might obscure the original root consonants Nichols 1999, 269 However all European languages use prefixes to some extent. If the dictionary is inadequate or the language unfamiliar, two limitations which often coincide, it can be difficult to distinguish between a prefix and the true root of a word. The only solution then is to become more familiar with the language or to check words against a good on-line dictionary.

Initial aspiration can lead to the loss of a initial consonant. The sample of aspirated words in the previous chapter (figs. 1.4-1.6) shows that when initial F or H is aspirated a second time it disappears. Double aspiration produces an eroded word with an exposed initial vowel. Thus, Lat. buro 'to burn' is now found as uro ‘to burn’, with intermediate stages marked by furo ‘to rage at’, furor ‘to steal (hunt)’ and furnus ‘oven’. Given these links it is safe to identify uro as a BR 'fire' word. However it seldom possible to be certain about the original consonant in eroded words and the safest solution is to omit words which begin with vowels.

Another problem with initial vowels is that they may be prothetic or euphonic. Such sounds were added in certain languages such as Greek and French to make pronunciation easier. They represent little more than an intake of breath which, with the advent of literacy, became a permanent feature. There are several ways of distinguishing between prothetic vowels and those exposed by double aspiration but none of them is foolproof. However prothetic vowels generally fail to cross language boundaries. This allows us to equate the following words.

Gr. amarusso ‘sparkle, twinkle, glance’ and Lat. coruscus ‘flashing, gleaming, glittering’
Gr. astraph ‘flash of lightning’ and Gr. strobos ‘whirling’
Lat. astrum, Gr. aster, and E. star
Fr. étincelle ‘spark’and E. tinseli
Gr. omicho ‘to urinate’ (also Gr. omichle ‘smoke, steam’) and Lat. mictio.

Once again any error can be avoided by omitting all words beginning with a vowel until their structure is understood.

It is more difficult to legislate for the loss of a medial consonant to aspiration. Such an event may be marked by a diphthong or double vowel sound, or by an aspirate such as H, F, V or W. Atholl retains a medial Th in English but in Gaelic is now pronounced ‘Aul’, like E. owl. Dougal retains a medial G in English but is now ‘Doo-ul’ in Gaelic. In Gaelic extra vowels are regularly added to satisfy spelling conventions. Of the thirty-three double vowels listed by Dwelly only fourteen have a double sound. However Gaelic spelling appears to be a reliable guide to the original root structure, retaining as it does many consonants which are not now pronounced. The problem for us is that medial aspiration changes the form of the root. The solution, as before, is to omit any word with a diphthong as its root is uncertain, but not all dictionaries provide pronunciation and not all double vowels are diphthongs: many are spelling conventions for a single sound. It is possible in many cases to identify the original form of such a word by checking the cognates listed in etymological dictionaries. The word-lists themselves offer a partial solution to this problem since they bring together related words, and a word which is listed in error will tend to stand out. Again, with a list of seven or eight hundred words, the numbers of diphthongs included in error is likely to be insignificant.

The changes produced by aspiration appear to have been part of the evolution of prehistoric language and are discussed in more detail elsewhere (page ref). This, and the limited number of words involved, suggests that we will not go far wrong if we stick to the principles of Radical Linguistics and list possibly eroded words according to their modern spelling. Their eventual removal will not have any significant effect on the structure of the list if the list is long enough.

The effect of medial aspiration on language is intriguing. A comparison of French and Latin suggests that the loss of a medial consonant has in the main affected literary words or recent technical terms borrowed into French. Latin itself was once an aspirating language – Lat. forma is cognate with Gr. morph ‘shape, form’ – and has its own lost consonants (see Chapter X), but it apparently stopped aspirating at some point and in consequence often retains a medial consonant which has droped out of French, much as we saw with Atholl and Dougal above. Thus we find Lat. credere and Fr. croire ‘to believe’, Lat. navicella and Fr. nacelle ‘skiff’, Lat. nativus ‘native’ and Fr. naïf ‘artless’, Lat. pullitra and Fr. poutre ‘beam’. These are all imported items and their introduction and change of form are documented. It is more difficult to understand the relationship between pedestrian words such as Lat. fodicare and Fr. fouiller ‘to dig’, and Lat. plicare and Fr. plier ‘to fold’. They are related but it is less obvious that the one derives from the other. However French (like Italian) tends to elide certain consonants which survive in Latin. We can therefore avoid this error by referring to Latin rather than to French or Italian.

When working with an unfamiliar language, when the only source is a basic word-list, more basic errors are liable to creep in. It is possible to mistake the meaning of bascic words such as bear v. bear; lead v. lead; score v. score. We can only hope for eventual enlightenment. Again, the total number of errors arising from such sources is likely to be a very small proportion of the total when the focus is on category and meaning rather than on quantity.

Silent letters such as those in E. balm, chalk, half, knife, know, stalk, and write also cause problems. However we can compare these words with E. balsam, Du. kalk, Du. half, Fr. canif, Sc. ken, Du. steel and Sc. vreet which have no silent letters. This sustains the general rule that when compiling a word-list we should list words according to their spelling. If our sample is large enough and it covers a sufficient variety of languages the inclusion of a few ‘false friends’ will have no significant effect.

In summary, none of these oddities are liable to cause serious distortion in a large sample. They are, after all, natural features of the languages of Europe, one of the many ways in which they have been shaped. If a few words slide from one category to another in a way that defies exact definition it is possible that a few words have been doing this for a very long time.

Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License