in father), and g has disappeared. Several other changes are equally well attested; so that we can say, not as a matter of conjecture but as a matter of historic fact, that most of the words in the modern Welsh dictionaries have undergone noteworthy changes in form. Still, the evidence of Welsh literature does not carry us quite as far back as we require to go. The oldest literary Welsh had already lost the endings indicating case and gender, which ancient Gaulish inscriptions show that the language originally possessed. These, however, have not disappeared without leaving traces. For example, while the dictionary form of the adjective meaning 'white' is gwyn, representing the Old British masculine windos and neuter windon, the Welsh for 'white river' is afon wen. The explanation of this is that the Old British was abonā windā; the final ā of the feminine adjective has had the effect of changing the i into e, and the fact that the feminine noun ended in a vowel has prevented the w of the following adjective from becoming gw, as it did when the word began a sentence or when the preceding word ended in a consonant. Much light on the prehistoric forms of Welsh words is afforded by comparison with the closely related Irish language, in which the nouns and adjectives often retain more distinct traces of the primitive declensions than the corresponding words in Welsh. Further, just as a rare Old French word, not surviving in modern French, may sometimes be interpreted by reference to the equivalent word in Italian or Spanish, it often happens that words contained in Ancient British place-names have been preserved (with the normal change of form) in Irish, although even the earliest known Welsh had already lost them.
The vocabulary, then, which has to be used for the interpretation of the British place-names of the Roman period, is not that of modern or even of early Welsh, but is a vocabulary reconstructed by means of a comparative study of the Celtic languages. It would, of course, be absurd to suppose
