The earliest, but scarce, evidence of cuneiform signs being used syllabically to write Akkadian words and proper nouns is at Fara and Tell Abu Salabikh between 2600 BC and 2500 BC. Between around 2350 BC and 1800 BC, there is an increase in the development and use of signs with syllabic values across Mesopotamia and Syria, but these syllabic values (together called 'syllabaries') are still very local in nature with significant and observable differences in sign usage and values between sites. Starting around 1800 BC, reforms to the system begin to be enforced that standardise these signs and their values, which essentially ends any major variability in the script within specific periods. This provides us with a period of almost 600 years, spanning the second half of the third millennium and early second millennium BC, during which there is a wealth of textual data documenting the first full adaptation of the cuneiform script to syllabically write Semitic words and proper nouns. This thesis investigates the attestations and usage of syllabic values to write Semitic lexemes in the cuneiform text corpora from Ebla, Mari, Nabada, Tuttul, Adab, Eshnunna, Kish, Tutub, Assur, and Gasur - with a particular focus on the Syrian sites - during the second half of the third millennium BC and early second millennium BC in order to answer the following two research questions: 1. Did each third millennium site in Mesopotamia and Syria have its own unique syllabary? 2. What were the primary factors that influenced the differences between the syllabaries? This research uses a series of three interdependent techniques to determine and understand the use and distribution of syllabic values within the cuneiform writing system during the second half of the third millennium BC and early second millennium BC. The results suggest that during this period cuneiform syllabaries are variable, and that variation can further inform us about the regional, temporal, and dialectical contexts in which they existed. The addition of this research to the wider literature on the early adaptation of cuneiform will enhance the field's understanding of how cuneiform syllabic values began to develop and emerge across the ancient Near East, and demonstrates how scientific and computational methods of analysis can be applied to research questions in humanities subjects.

Thesis. 1976. Ph.D.--Massachusetts Institute of Technology. Dept. of Foreign Literatures and Linguistics. / Microfiche copy available in Archives and Humanities. / Bibliography: leaves 211-218. / Ph.D.

Text in Afrikaans / Summaries in Afrikaans and English / In Afrikaans, soos in NederJands en Duits, word saamgestelde woorde aanmekaar geskryf. Nuwe woorde word dus voortdurend geskep deur woorde aanmekaar te haak Dit bemoeilik die proses van woordafkapping tydens teksprosessering, wat deesdae deur rekenaars gedoen word, aangesien die verwysingsbron gedurig verander. Daar bestaan verskeie afkappingsalgoritmes en tegnieke, maar die resultate is onbevredigend. Afrikaanse woorde met korrekte lettergreepverdeling is net die elektroniese weergawe van die handwoordeboek van die Afrikaanse Taal (HAT) onttrek. 'n Neutrale netwerk ( vorentoevoer-terugpropagering) is met sowat. 5 000 van hierdie woorde afgerig. Die neurale netwerk is verfyn deur 'n gcskikte afrigtingsalgoritme en oorfragfunksie vir die probleem asook die optimale aantal verborge lae en aantal neurone in elke laag te bepaal. Die neurale netwerk is met 5 000 nuwe woorde getoets en dit het 97,56% van moontlike posisies korrek as of geldige of ongeldige afkappingsposisies geklassifiseer. Verder is 510 woorde uit tydskrifartikels met die neurale netwerk getoets en 98,75% van moontlike posisies is korrek geklassifiseer. / In Afrikaans, like in Dutch and German, compound words are written as one word. New words are therefore created by simply joining words. Word hyphenation during typesetting by computer is a problem, because the source of reference changes all the time. Several algorithms and techniques for hyphenation exist, but results are not satisfactory. Afrikaans words with correct syllabification were extracted from the electronic version of the Handwoordeboek van die Afrikaans Taal (HAT). A neural network (feedforward backpropagation) was trained with about 5 000 of these words. The neural network was refined by heuristically finding a suitable training algorithm and transfer function for the problem as well as determining the optimal number of layers and number of neurons in each layer. The neural network was tested with 5 000 words not the training data. It classified 97,56% of possible points in these words correctly as either valid or invalid hyphenation points. Furthermore, 510 words from articles in a magazine were tested with the neural network and 98,75% of possible positions were classified correctly. / Computing / M.Sc. (Operasionele Navorsing)

Text in Afrikaans / Summaries in Afrikaans and English / In Afrikaans, soos in NederJands en Duits, word saamgestelde woorde aanmekaar geskryf. Nuwe woorde word dus voortdurend geskep deur woorde aanmekaar te haak Dit bemoeilik die proses van woordafkapping tydens teksprosessering, wat deesdae deur rekenaars gedoen word, aangesien die verwysingsbron gedurig verander. Daar bestaan verskeie afkappingsalgoritmes en tegnieke, maar die resultate is onbevredigend. Afrikaanse woorde met korrekte lettergreepverdeling is net die elektroniese weergawe van die handwoordeboek van die Afrikaanse Taal (HAT) onttrek. 'n Neutrale netwerk ( vorentoevoer-terugpropagering) is met sowat. 5 000 van hierdie woorde afgerig. Die neurale netwerk is verfyn deur 'n gcskikte afrigtingsalgoritme en oorfragfunksie vir die probleem asook die optimale aantal verborge lae en aantal neurone in elke laag te bepaal. Die neurale netwerk is met 5 000 nuwe woorde getoets en dit het 97,56% van moontlike posisies korrek as of geldige of ongeldige afkappingsposisies geklassifiseer. Verder is 510 woorde uit tydskrifartikels met die neurale netwerk getoets en 98,75% van moontlike posisies is korrek geklassifiseer. / In Afrikaans, like in Dutch and German, compound words are written as one word. New words are therefore created by simply joining words. Word hyphenation during typesetting by computer is a problem, because the source of reference changes all the time. Several algorithms and techniques for hyphenation exist, but results are not satisfactory. Afrikaans words with correct syllabification were extracted from the electronic version of the Handwoordeboek van die Afrikaans Taal (HAT). A neural network (feedforward backpropagation) was trained with about 5 000 of these words. The neural network was refined by heuristically finding a suitable training algorithm and transfer function for the problem as well as determining the optimal number of layers and number of neurons in each layer. The neural network was tested with 5 000 words not the training data. It classified 97,56% of possible points in these words correctly as either valid or invalid hyphenation points. Furthermore, 510 words from articles in a magazine were tested with the neural network and 98,75% of possible positions were classified correctly. / Computing / M.Sc. (Operasionele Navorsing)

Thesis (MA)--Stellenbosch University, 2004. / ENGLISH ABSTRACT: This study intends to describe and explain syllable structure processes in Northern Sotho. It deals with phonological processes such as vowel deletion, semivocalization and semivowel insertion. The major aim of these processes is to restore the preferred ICVI syllable structure which has been violated by morphological processes such as passive, diminutive, the construction of absolute pronouns, etc. Two phonological models are applied with the intention to determine the one that presents the most credible explanation for the phenomenon at hand. The two models employed are, respectively, the Transformational (TG) and Feature Geometry (FG) models. It appears that Feature Geometry model yields better results in the description of syllable structure processes in Northern Sotho. / AFRIKAANSE OPSOMMING: Hierdie studie beskryf en verklaar sillabestruktuur prosesse in Noord-Sotho. Die tersaaklike fonologiese prosesse is vokaaldelesie, semivokalisasie en semivokaalinvoeging. Die doel van hierdie prosesse is om "n bepaalde voorkeursillabestruktuur IKVI te herstel wat versteur word deur morfologiese prosesse met die vorming van die passief, diminutief, die konstruksie van absolute voornaamwoorde ensovoorts. Twee fonologiese modelle word geïmplementeer ten einde te bepaal welke model die mees geloofwaardige verklarings vir die betrokke verskynsels kan bied. Die Transformasioneel-Generatiewe (TG) en Kenmerk Geometriese(KG) modelle word respektiewelik toegepas. Dit skyn asof die Kenmerk Geometriese model beter resultate lewer in die beskrywing van sillabestruktuurprosesse in Noord-Sotho.

由於普通話的語音系統與粤方言的相去甚遠，粤方言人士學習普通話語音時，來自粤方言的語音干擾對學習產生障礙，發音因而有所偏差。本研究應用多元概化理論的統計方法，分析由4位國家級普通話測試員對37名大學本科生朗讀1218個普通話帶調音節的評分數據，從而描繪香港學生朗讀普通話音節的能力面貌。 / 本研究在音節朗讀能力分析上應用概化理論，能同時估計影響分數變異的多個誤差來源，以及估計各側面的變異量及其佔總變異量的比例。通過本研究可以進一步了解來自題目、評分者的測量誤差對分數概化精確性的影響。而且，本研究採用的數據是對聲母、韻母、聲調的分項評分，把音節分解為三個變量來分析，能彌補整體性評分的不足，從而確定朗讀普通話音節時的偏誤所在。本研究對朗讀能力進行分解，首先是在音節的層面，對音節之下的子能力（聲母、韻母、聲調）進行解構分析。然後，則分別在聲母和韻母的層面探究兩者的子能力。研究發現，聲母、韻母、聲調之間的關聯程度很高，但韻母最能區分被試的朗讀能力，聲母卻最弱。各種韻母之間的協方差明顯比各種聲母之間的為大，表明各類韻母背後所隱含的共同能力相對較強。研究結果也顯示，被試能力和題目之間的交互作用是相對主要的誤差來源。 / 除了解構音節朗讀能力外，本研究還針對音節測試的實際情況進行應用研究。通過題目數量、評分者人數以及測量設計的改變，驗證不同測試條件下的概化係數等各種技術指標。結果顯示，各種改變條件的測試方案皆具相當高的測量信度，證明在實際的測試要求下，音節朗讀測試能測出被試的普通話語音能力。 / 本研究基於大量客觀數據而概括出的學習難點和規律，為普通話教與學提供客觀參照，更具針對性地教好、學好音節發音。同時，也有助本地測試機構，制定更符合粤方言者能力特質的音節題目。評分者效應的誤差則為評分者培訓提供訊息，有助提升評分的準確性，建設穩定的評分者團隊。 / As the phonological system of Putonghua is vastly different from that of Cantonese, the phonological interference from Cantonese poses critical learning obstacles to Cantonese speakers in their learning of Putonghua, which leads to various pronunciation errors. In this study, the statistical method Multivariate Generalizability Theory was used to analyze the rating data collected from 4 National Putonghua examiners' rating on the pronunciation of 1218 Putonghua syllables by 37 undergraduates at a university in Hong Kong. The purpose of this study was to examine the ability profiles and psychometric characteristics of Hong Kong students in the pronunciation of Putonghua syllables. / The application of generalizability theory on the study of the pronunciation competency of Putonghua syllables made it possible to analyze and estimate the multiple sources of errors which affected the variation of scores as well as the variance of each facet and its proportion against total variance. Through this study, it was possible to further understand the influence of measurement errors of test items and raters on the generalization of their accuracy. In the study, Putonghua syllables were divided into three variables (components: initials, finals and tones) and analytical scoring was applied in the analyses that could specify the location of errors in pronunciation, thus partly remedying the deficiency of holistic scoring. Deconstruction analysis of pronunciation competence was carried out in a 2-level hierarchy. First, the deconstruction took place at syllables in which initials, finals and tones were the sub-skills. Second, the initials and finals were also deconstructed. Research findings showed that correlations among initials, finals and tones were relatively high and that finals were the strongest in differentiating pronunciation competence while initials were the weakest. The covariances among various finals were apparently higher than those of initials. This implies that the mutual hidden competence linked to various types of finals was relatively strong. Findings also revealed that the interaction between subject’s competence and test items was the main source of errors. / In addition to the deconstruction analysis of pronunciation competence, applied research basing on the authentic situations of Putonghua tests was also conducted. Verification of technical specifications, including generalizability coefficients, was carried out under different testing conditions through changing test design and the number of test items and raters. The results showed that various testing solutions with the change in testing conditions possess high degree of measurement reliability, suggesting that syllable reading test could be used to examine pronunciation competence under authentic testing conditions. / In this study, learning difficulties and their patterns, which are derived from abundant unprejudiced data, provide objective reference to the teaching and learning of Putonghua syllable pronunciation with clear targets. These findings are also useful to local testing institutions in stipulating syllable reading tests which better suit the competence characteristics of Cantonese speakers. Furthermore, data and study results on rater errors offer reference for rater training which helps to promote the accuracy of rating, thus strengthening the quality of the rating team. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / Detailed summary in vernacular field only. / 陳志良. / "2013年7月". / "2013 nian 7 yue". / Thesis (Ed.D.)--Chinese University of Hong Kong, 2013. / Includes bibliographical references (leaves 119-127). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstract in Chinese and English. / Chen Zhiliang. / Chapter 第一章 --- 緒論 --- p.1 / Chapter 1.1 --- 研究背景 --- p.1 / Chapter 1.2 --- 研究的目的、問題、預設與假設、內容、意義 --- p.2 / Chapter 1.2.1 --- 研究目的 --- p.2 / Chapter 1.2.2 --- 研究問題 --- p.3 / Chapter 1.2.2.1 --- 音節的理論探討 --- p.3 / Chapter 1.2.2.2 --- 應用研究探討 --- p.4 / Chapter 1.2.3 --- 研究預設與假設 --- p.5 / Chapter 1.2.4 --- 研究內容 --- p.6 / Chapter 1.2.5 --- 研究意義 --- p.7 / Chapter 第二章 --- 文獻綜述 --- p.9 / Chapter 2.1 --- 普通話在香港的地位 --- p.9 / Chapter 2.2 --- 普通話音節與漢語拼音 --- p.12 / Chapter 2.3 --- 普通話音節分析 --- p.15 / Chapter 2.4 --- 方言地區的普通話音節教學與學習 --- p.17 / Chapter 2.5 --- 普通話音節評分方法 --- p.22 / Chapter 2.6 --- 概化理論在普通話測試上的應用 --- p.27 / Chapter 第三章 --- 研究方法 --- p.33 / Chapter 3.1 --- 數據收集程序 --- p.33 / Chapter 3.1.1 --- 被試 --- p.33 / Chapter 3.1.2 --- 題目 --- p.33 / Chapter 3.1.3 --- 評分者 --- p.35 / Chapter 3.1.4 --- 評分方法 --- p.35 / Chapter 3.2 --- 能力分解研究 --- p.36 / Chapter 3.2.1 --- 子研究一：音節研究 --- p.36 / Chapter 3.2.2 --- 子研究二：聲母研究 --- p.38 / Chapter 3.2.3 --- 子研究三：韻母研究 --- p.39 / Chapter 3.3 --- 應用研究 --- p.39 / Chapter 3.3.1 --- 子研究一：組卷啟示 --- p.40 / Chapter 3.3.2 --- 子研究二：考試設計探索 --- p.40 / Chapter 3.3.2.1 --- G研究 --- p.40 / Chapter 3.3.2.2 --- D₁研究：改變題目數量 --- p.40 / Chapter 3.3.2.3 --- D₂研究：改變評分者人數 --- p.41 / Chapter 3.3.2.4 --- D₃研究：（P：R）× I嵌套設計（不同的評分者評不同的被試） --- p.41 / Chapter 3.3.2.5 --- D₄研究：（I：P）× R嵌套設計（不同的被試朗讀不同的音節） --- p.42 / Chapter 第四章 --- 能力分解研究 --- p.44 / Chapter 4.1 --- 子研究一：音節研究（三個因子模型） --- p.44 / Chapter 4.1.1 --- G研究 --- p.44 / Chapter 4.1.1.1 --- 測量目標p --- p.45 / Chapter 4.1.1.2 --- 測量側面i --- p.45 / Chapter 4.1.1.3 --- 測量側面r --- p.46 / Chapter 4.1.1.4 --- 各效應之間的比較 --- p.46 / Chapter 4.1.1.5 --- 交互效應 --- p.47 / Chapter 4.1.2 --- 默認D研究 --- p.48 / Chapter 4.1.2.1 --- 概化全域 --- p.48 / Chapter 4.1.2.2 --- 默認D研究中的方差與協方差分量的估計 --- p.48 / Chapter 4.1.2.3 --- 默認D研究中各效應在三個變量上的概化係數等指標 --- p.49 / Chapter 4.1.2.4 --- 全域合成分數的研究 --- p.51 / Chapter 4.1.2.5 --- 各變量對整體測試的貢獻比例 --- p.52 / Chapter 4.1.3 --- D₁研究：改變評分者人數 --- p.52 / Chapter 4.1.3.1 --- 模式設計 --- p.52 / Chapter 4.1.3.2 --- 研究發現 --- p.53 / Chapter 4.1.4 --- 小結 --- p.54 / Chapter 4.2 --- 子研究二：聲母研究（七個因子模型） --- p.56 / Chapter 4.2.1 --- G研究 --- p.56 / Chapter 4.2.1.1 --- 測量目標p --- p.56 / Chapter 4.2.1.2 --- 測量側面i --- p.59 / Chapter 4.2.1.3 --- 測量側面r --- p.60 / Chapter 4.2.1.4 --- 各效應之間的比較 --- p.60 / Chapter 4.2.1.5 --- 交互效應 --- p.61 / Chapter 4.2.2 --- 默認D研究 --- p.62 / Chapter 4.2.2.1 --- 概化全域 --- p.62 / Chapter 4.2.2.2 --- 默認D研究中的方差與協方差分量的估計 --- p.62 / Chapter 4.2.2.3 --- 默認D研究中各效應在三個變量上的概化係數等指標 --- p.62 / Chapter 4.2.3 --- D₂研究：改變評分者人數 --- p.65 / Chapter 4.2.3.1 --- 模式設計 --- p.65 / Chapter 4.2.3.2 --- 研究發現 --- p.65 / Chapter 4.2.4 --- 小結 --- p.67 / Chapter 4.3 --- 子研究三：韻母研究（四個因子模型） --- p.69 / Chapter 4.3.1 --- G研究 --- p.69 / Chapter 4.3.1.1 --- 測量目標p --- p.70 / Chapter 4.3.1.2 --- 測量側面i --- p.71 / Chapter 4.3.1.3 --- 測量側面r --- p.71 / Chapter 4.3.1.4 --- 各效應之間的比較 --- p.72 / Chapter 4.3.1.5 --- 交互效應 --- p.72 / Chapter 4.3.2 --- 默認D研究 --- p.73 / Chapter 4.3.2.1 --- 概化全域 --- p.73 / Chapter 4.3.2.2 --- 默認D研究中的方差與協方差分量的估計 --- p.73 / Chapter 4.3.2.3 --- 默認D研究中各效應在三個變量上的概化係數等指標 --- p.74 / Chapter 4.3.3 --- D₃研究：改變評分者人數 --- p.76 / Chapter 4.3.3.1 --- 模式設計 --- p.76 / Chapter 4.3.3.2 --- 研究發現 --- p.76 / Chapter 4.3.4 --- 小結 --- p.78 / Chapter 4.4 --- 綜合討論 --- p.80 / Chapter 第五章 --- 應用研究 --- p.82 / Chapter 5.1 --- 描述統計 --- p.82 / Chapter 5.1.1 --- 平均分和標準差 --- p.82 / Chapter 5.1.2 --- 被試表現分析 --- p.85 / Chapter 5.2 --- 子研究一：組卷啟示 --- p.87 / Chapter 5.2.1 --- 難度控制組卷 --- p.87 / Chapter 5.2.2 --- 組卷策略 --- p.88 / Chapter 5.2.3 --- 借助計算機程序 --- p.90 / Chapter 5.2.4 --- 組卷方式 --- p.90 / Chapter 5.2.4.1 --- 教學導向組卷 --- p.90 / Chapter 5.2.4.2 --- 測試導向組卷 --- p.91 / Chapter 5.2.4.3 --- 教學導向與測試導向結合 --- p.92 / Chapter 5.2.5 --- 小結 --- p.92 / Chapter 5.3 --- 子研究二：考試設計探索 --- p.94 / Chapter 5.3.1 --- G研究 --- p.94 / Chapter 5.3.2 --- 默認D研究 --- p.95 / Chapter 5.3.3 --- D₁研究：改變題目數量 --- p.96 / Chapter 5.3.3.1 --- 取樣模式 --- p.96 / Chapter 5.3.3.2 --- 研究發現 --- p.96 / Chapter 5.3.4 --- D₂研究：改變評分者人數 --- p.97 / Chapter 5.3.4.1 --- 取樣模式 --- p.98 / Chapter 5.3.4.2 --- 研究發現 --- p.98 / Chapter 5.3.5 --- D₃研究：（P：R）× I嵌套設計（不同的評分者評不同的被試） --- p.99 / Chapter 5.3.5.1 --- 研究設計 --- p.99 / Chapter 5.3.5.2 --- 研究發現 --- p.101 / Chapter 5.3.6 --- D₄研究：（I：P）× R嵌套設計（不同的被試朗讀不同的音節） --- p.101 / Chapter 5.3.6.1 --- 研究設計 --- p.101 / Chapter 5.3.6.2 --- 研究發現 --- p.102 / Chapter 5.3.7 --- 其他嵌套設計 --- p.103 / Chapter 5.3.8 --- 小結 --- p.105 / Chapter 第六章 --- 總結及建議 --- p.107 / Chapter 6.1 --- 結語 --- p.107 / Chapter 6.2 --- 主要結論 --- p.108 / Chapter 6.2.1 --- 音節研究 --- p.108 / Chapter 6.2.2 --- 聲母研究 --- p.109 / Chapter 6.2.3 --- 韻母研究 --- p.110 / Chapter 6.2.4 --- 評分者因素 --- p.111 / Chapter 6.2.5 --- 組卷啟示 --- p.111 / Chapter 6.2.6 --- 考試設計探索 --- p.112 / Chapter 6.3 --- 研究的不足 --- p.113 / Chapter 6.3.1 --- 被試的同質性 --- p.113 / Chapter 6.3.2 --- 題目側面的細化 --- p.114 / Chapter 6.3.3 --- 選題拼卷的聲韻覆蓋 --- p.114 / Chapter 6.4 --- 主要建議 --- p.115 / Chapter 6.4.1 --- 分項評分 --- p.115 / Chapter 6.4.2 --- 測試設計 --- p.116 / Chapter 6.4.3 --- 評分者人數與題目數量之間的關係 --- p.116 / Chapter 6.4.4 --- 音節教學和評分者培訓 --- p.117 / Chapter 6.4.5 --- 擴大研究對象 --- p.117 / Chapter 6.4.6 --- 擴大題目側面 --- p.118 / Chapter 6.4.7 --- 借助計算機程序 --- p.118 / 參考文獻 --- p.119 / Chapter 附錄 1 --- 音節與漢字對照表 --- p.128 / Chapter 附錄 2 --- 1218 個音節 --- p.136 / Chapter 附錄 3 --- 按聲母排列的音節難度表 --- p.144 / Chapter 附錄 4 --- 以聲母分類隨機選取的 100 音節試卷 --- p.156 / Chapter 附錄 5 --- 音節研究 mGENOVA程序 --- p.158

abstract: Learning to read in English is difficult for adult English language learners due to their diverse background, their level of experience with literacy in their first language, and their reason and desire for wanting to learn to read in English. Teachers of adult language learners must consider the educational and language experiences of adults enrolled in English as a Second Language (ESL) classes in order to provide adequate learning opportunities for a diverse student body. Promoting learning opportunities for adult Arabic speakers was an area of interest for me when I first began teaching adult English language learners six years ago. The purpose of my action research study was to provide the adult Arabic speakers in my classroom with strategies they could use in order to read accurately in English. Current research used to guide my study focused on the difficulties Arabic speakers have with the orthographic features of the English language. As I conducted various cycles of action research in an ESL reading class, I developed an intervention to support adult Arabic speakers gain an understanding of the sound spelling system of the English language inclusive of instructional strategies to support accurate word reading. Data was collected to identify the individuals experience in learning to read. I included a pre and post miscue analysis to help identify the common error patterns of the participants of my study. Over an eight-week period, I followed a constructivist approach and facilitated word sorts to help students identify common sound spellings found in the English language. Instructional strategies were included to help the participants decode multisyllabic words by bringing awareness to the syllable types found in the English language. The findings of my study revealed that Arabic speakers benefited from an intervention focused on the sound spellings and syllabication of the English language. / Dissertation/Thesis / Doctoral Dissertation Educational Leadership and Policy Studies 2020

Text in Afrikaans / Die doel van hierdie studie was om te bepaal tot watter mate ’n suiwer patroongebaseerde benadering tot woordafbreking bevredigende resultate lewer. Die masjienleertegnieke kunsmatige neurale netwerke, beslissingsbome en die TEX-algoritme is ondersoek aangesien dit met letterpatrone uit woordelyste afgerig kan word om lettergreep- en saamgesteldewoordverdeling te doen. ’n Leksikon van Afrikaanse woorde is uit ’n korpus van elektroniese teks genereer. Om lyste vir lettergreep- en saamgesteldewoordverdeling te kry, is woorde in die leksikon in lettergrepe verdeel en saamgestelde woorde is in hul samestellende dele verdeel. Uit elkeen van hierdie lyste van ±183 000 woorde is ±10 000 woorde as toetsdata gereserveer terwyl die res as afrigtingsdata gebruik is. ’n Rekursiewe algoritme is vir saamgesteldewoordverdeling ontwikkel. In hierdie algoritme word alle ooreenstemmende woorde uit ’n verwysingslys (die leksikon) onttrek deur stringpassing van die begin en einde van woorde af. Verdelingspunte word dan op grond van woordlengte uit die samestelling van begin- en eindwoorde bepaal. Die algoritme is uitgebrei deur die tekortkominge van hierdie basiese prosedure aan te spreek. Neurale netwerke en beslissingsbome is afgerig en variasies van beide tegnieke is ondersoek om die optimale modelle te kry. Patrone vir die TEX-algoritme is met die OPatGen-program gegenereer. Tydens toetsing het die TEX-algoritme die beste op beide lettergreep- en saamgesteldewoordverdeling presteer met 99,56% en 99,12% akkuraatheid, respektiewelik. Dit kan dus vir woordafbreking gebruik word met min risiko vir afbrekingsfoute in gedrukte teks. Die neurale netwerk met 98,82% en 98,42% akkuraatheid op lettergreep- en saamgesteldewoordverdeling, respektiewelik, is ook bruikbaar vir lettergreepverdeling, maar dis meer riskant. Ons het bevind dat beslissingsbome te riskant is om vir lettergreepverdeling en veral vir woordverdeling te gebruik, met 97,91% en 90,71% akkuraatheid, respektiewelik. ’n Gekombineerde algoritme is ontwerp waarin saamgesteldewoordverdeling eers met die TEXalgoritme gedoen word, waarna die resultate van lettergreepverdeling deur beide die TEXalgoritme en die neurale netwerk gekombineer word. Die algoritme het 1,3% minder foute as die TEX-algoritme gemaak. ’n Toets op gepubliseerde Afrikaanse teks het getoon dat die risiko vir woordafbrekingsfoute in teks met gemiddeld tien woorde per re¨el ±0,02% is. / The aim of this study was to determine the level of success achievable with a purely pattern based approach to hyphenation in Afrikaans. The machine learning techniques artificial neural networks, decision trees and the TEX algorithm were investigated since they can be trained with patterns of letters from word lists for syllabification and decompounding. A lexicon of Afrikaans words was extracted from a corpus of electronic text. To obtain lists for syllabification and decompounding, words in the lexicon were respectively syllabified and compound words were decomposed. From each list of ±183 000 words, ±10 000 words were reserved as testing data and the rest was used as training data. A recursive algorithm for decompounding was developed. In this algorithm all words corresponding with a reference list (the lexicon) are extracted by string fitting from beginning and end of words. Splitting points are then determined based on the length of reassembled words. The algorithm was expanded by addressing shortcomings of this basic procedure. Artificial neural networks and decision trees were trained and variations of both were examined to find optimal syllabification and decompounding models. Patterns for the TEX algorithm were generated by using the program OPatGen. Testing showed that the TEX algorithm performed best on both syllabification and decompounding tasks with 99,56% and 99,12% accuracy, respectively. It can therefore be used for hyphenation in Afrikaans with little risk of hyphenation errors in printed text. The performance of the artificial neural network was lower, but still acceptable, with 98,82% and 98,42% accuracy for syllabification and decompounding, respectively. The decision tree with accuracy of 97,91% on syllabification and 90,71% on decompounding was found to be too risky to use for either of the tasks A combined algorithm was developed where words are first decompounded by using the TEX algorithm before syllabifying them with both the TEX algoritm and the neural network and combining the results. This algoritm reduced the number of errors made by the TEX algorithm by 1,3% but missed more hyphens. Testing the algorithm on Afrikaans publications showed the risk for hyphenation errors to be ±0,02% for text assumed to have an average of ten words per line. / Decision Sciences / D. Phil. (Operational Research)

Text in Afrikaans / Die doel van hierdie studie was om te bepaal tot watter mate ’n suiwer patroongebaseerde benadering tot woordafbreking bevredigende resultate lewer. Die masjienleertegnieke kunsmatige neurale netwerke, beslissingsbome en die TEX-algoritme is ondersoek aangesien dit met letterpatrone uit woordelyste afgerig kan word om lettergreep- en saamgesteldewoordverdeling te doen. ’n Leksikon van Afrikaanse woorde is uit ’n korpus van elektroniese teks genereer. Om lyste vir lettergreep- en saamgesteldewoordverdeling te kry, is woorde in die leksikon in lettergrepe verdeel en saamgestelde woorde is in hul samestellende dele verdeel. Uit elkeen van hierdie lyste van ±183 000 woorde is ±10 000 woorde as toetsdata gereserveer terwyl die res as afrigtingsdata gebruik is. ’n Rekursiewe algoritme is vir saamgesteldewoordverdeling ontwikkel. In hierdie algoritme word alle ooreenstemmende woorde uit ’n verwysingslys (die leksikon) onttrek deur stringpassing van die begin en einde van woorde af. Verdelingspunte word dan op grond van woordlengte uit die samestelling van begin- en eindwoorde bepaal. Die algoritme is uitgebrei deur die tekortkominge van hierdie basiese prosedure aan te spreek. Neurale netwerke en beslissingsbome is afgerig en variasies van beide tegnieke is ondersoek om die optimale modelle te kry. Patrone vir die TEX-algoritme is met die OPatGen-program gegenereer. Tydens toetsing het die TEX-algoritme die beste op beide lettergreep- en saamgesteldewoordverdeling presteer met 99,56% en 99,12% akkuraatheid, respektiewelik. Dit kan dus vir woordafbreking gebruik word met min risiko vir afbrekingsfoute in gedrukte teks. Die neurale netwerk met 98,82% en 98,42% akkuraatheid op lettergreep- en saamgesteldewoordverdeling, respektiewelik, is ook bruikbaar vir lettergreepverdeling, maar dis meer riskant. Ons het bevind dat beslissingsbome te riskant is om vir lettergreepverdeling en veral vir woordverdeling te gebruik, met 97,91% en 90,71% akkuraatheid, respektiewelik. ’n Gekombineerde algoritme is ontwerp waarin saamgesteldewoordverdeling eers met die TEXalgoritme gedoen word, waarna die resultate van lettergreepverdeling deur beide die TEXalgoritme en die neurale netwerk gekombineer word. Die algoritme het 1,3% minder foute as die TEX-algoritme gemaak. ’n Toets op gepubliseerde Afrikaanse teks het getoon dat die risiko vir woordafbrekingsfoute in teks met gemiddeld tien woorde per re¨el ±0,02% is. / The aim of this study was to determine the level of success achievable with a purely pattern based approach to hyphenation in Afrikaans. The machine learning techniques artificial neural networks, decision trees and the TEX algorithm were investigated since they can be trained with patterns of letters from word lists for syllabification and decompounding. A lexicon of Afrikaans words was extracted from a corpus of electronic text. To obtain lists for syllabification and decompounding, words in the lexicon were respectively syllabified and compound words were decomposed. From each list of ±183 000 words, ±10 000 words were reserved as testing data and the rest was used as training data. A recursive algorithm for decompounding was developed. In this algorithm all words corresponding with a reference list (the lexicon) are extracted by string fitting from beginning and end of words. Splitting points are then determined based on the length of reassembled words. The algorithm was expanded by addressing shortcomings of this basic procedure. Artificial neural networks and decision trees were trained and variations of both were examined to find optimal syllabification and decompounding models. Patterns for the TEX algorithm were generated by using the program OPatGen. Testing showed that the TEX algorithm performed best on both syllabification and decompounding tasks with 99,56% and 99,12% accuracy, respectively. It can therefore be used for hyphenation in Afrikaans with little risk of hyphenation errors in printed text. The performance of the artificial neural network was lower, but still acceptable, with 98,82% and 98,42% accuracy for syllabification and decompounding, respectively. The decision tree with accuracy of 97,91% on syllabification and 90,71% on decompounding was found to be too risky to use for either of the tasks A combined algorithm was developed where words are first decompounded by using the TEX algorithm before syllabifying them with both the TEX algoritm and the neural network and combining the results. This algoritm reduced the number of errors made by the TEX algorithm by 1,3% but missed more hyphens. Testing the algorithm on Afrikaans publications showed the risk for hyphenation errors to be ±0,02% for text assumed to have an average of ten words per line. / Decision Sciences / D. Phil. (Operational Research)