Part 1, 'Introduction'. Begins with the chapter, 'General Background to Error Analysis'. A survey of the thesis is presented with a statement of the research objectives. Tbc chapter then goes on to demonstrate that error analysis was inspired by the generative linguistics movement of the 1960's which concentrated on the creative aspects of language acquisition. Generative linguistics was a reaction against the behaviourist-structuralist position. Error analysis itself was a reaction to the behaviourist contrastive analysis which held sway in the 1950's. To explain how error analysis first arose, an account of behaviourist psychology as pioneered by Skinner, amongst others, is provided, and its effect on second language learning is explored. Evidence is provided which contradicted behaviourist claims and supported Chomsky's mentalism which gradually superseded behaviourist theory in the field of second language acquisition. This eventually led to the idea of interlanguage being formulated by Selinker and a reappraisal of the significance of learners' errors by Corder. His ideas helped to lay the foundations of the error analysis movement which reached its height amongst EFL researchers during the 1970's and early 1980's. Corder rejected the idea that errors had only negative significance, being solely the result of the laziness of the learner or faulty teaching methods. Instead, he affirmed that errors were to be viewed in a positive light. This was so because learners progressed through processes of overgeneralization and regularization. Therefore, learners errors provided evidence, of their progress and could possibly reveal the 'internal syllabus' of second-language learners. Chapter 2 provides a survey of the language error literature from 1917 to 1995. Section I deals with written errors. Until the 196M the papers reviewed were mainly behaviourist in tenor. This was the case as their main purpose in identifying errors was the belief that they could be eradicated by the formulation of remedial exercises. From the 1950's, it was also believed that errors could be corrected by the devising of drills based upon the areas of perceived weakness. The language which received more attention than any other was English, both native and second language and as spoken or acquired by both children (hearing or deaf) and adults. Other languages surveyed include French, German, Russian, Spanish, Hebrew, and Irish. Section 2 of Chapter 2 gives a comprehensive survey of error analysis in Welsh. beginning in the 1980's. It was originally motivated by fears that the Welsh of primary level children who were fluent in the language (both native speakers and learners) was becoming progressively anglicized. This view was largely upheld by the research conducted and furthermore it was demonstrated that native-speaking children in Welsh-medium schools were assimilating to the interlanguage spoken by the L2 majority. Relatively little research had been conducted on the errors committed by adult learners. Gillian Evans of the Polytechnic of Wales had created a taxonomy of errors committed in the Adult Welsh 'O'Level but no research had been conducted on the errors of higher level learners. .I- Chapter 3 details the theory to be tested. It comprises a number of statistical hypotheses. These are as follows: (1) Welsh adult learners commit mainly local errors. (2) It is impossible to assign a cause to many of the errors. (3) Of the errors to which it is possible to assign a cause the majority are developmental. (4) There are more syntactic errors than lexical. (5) Women commit fewer errors than men. (6) Ile Welshness of the area where learners live does not affect the correctness of their written Welsh. The major hypothesis was that a longitudinal analysis of errors encompassing the Use of Welsh and Advanced Use of Welsh Written Tests would reveal a developmental pattern in the errors committed. Chapter 6 presents the results of the analysis. Approximately 135,000 words were analyzed, yielding some 14,000 lexical and grammatical errors. All the hypotheses were confirmed by these results. Ninety-six categories of error were created and the taxonomy was organized according to surface strategies as recommended by Dulay, Burt and Krashen. It was demonstrated that there was a general decrease in error frequency between Level 4 and Level 6, despite the hierarchy of written assignments. Chapter 7 comprises a detailed discussion of the results. It begins at the level of surface strategy and shows that the decrease in error frequency was evinced in each one of the surface strategies although the rate of decrease varied. It then goes on to discuss each error category in detail organized according to sub-samples (Use of Welsh, Advanced Level la, Advanced Level lb, Advanced Level 2). Firstly, Omissions of Major Constituents are discussed, followed by Omissions of Grammatical Morphemes. A new strategy is created to analyze the errors of Celtic languages, namely the Omission of Mutations, as they are syntactic errors but are neither major constituents nor grammatical morphemes. The chapter goes on to discuss the variation within the strategies of Double Marking and Simple Addition. However, Welsh departs from the universal pattern of error analysis in that no evidence was Found of Regularization Addition errors .Even so, examples were found of all three types of Misformations namely Regularization, Archi-forms and Alternating Forms. The last strategy to be discussed was the least prominent, namely Misordering, which included confusion of the normal and abnormal sentence order. The chapter concludes by revealing the rate of change in frequency of each error category. Fifty seven categories decreased between Use of Welsh and Advanced Level. Here the learners seem to be successfully internalizing the rules of the language items involved, with overgeneralization and language transfer being abandoned. This seems to represent the last stage of acquisition. Ile intermediate stage appears to comprise the seven error categories which exhibit no statistically significant rate change between the two examinations. These categories may well go on to decrease at the Post-A Level stage or fossilize. Lastly, 32 categories increased in frequency. These appear to represent the initial stage of acquisition, in which new language items (especially literary forms) have been introduced and overgeneralization has increased in the attempt to acquire the new rules. In this way, clues are given to the linguistic difficulties which learners experience during the Advanced Level course of study. It is hoped that the insights provided will assist those engaged in the formulation of a curriculum or syllabus for Level 5 and 6 learners Other practical applications of the research are discussed and the final chapter makes suggestions for further research.