Washington University in St. Louis Washington University Open Scholarship All Theses and Dissertations (ETDs) 1-1-2011 Speling "Successful" Sucesfuly: Statistical Learning in Spelling Siti Syuhada Binte Faizal Follow this and additional works at:http://openscholarship.wustl.edu/etd Recommended Citation Binte Faizal, Siti Syuhada, "Speling "Successful" Sucesfuly: Statistical Learning in Spelling" (2011).All Theses and Dissertations (ETDs). 542. http://openscholarship.wustl.edu/etd/542 This Thesis is brought to you for free and open access by Washington University Open Scholarship. It has been accepted for inclusion in All Theses and Dissertations (ETDs) by an authorized administrator of Washington University Open Scholarship. For more information, please contact [email protected]. WASHINGTON UNIVERSITY Department of Psychology Speling “Successful” Sucesfuly: Statistical Learning in Spelling by Siti Syuhada Binte Faizal A thesis presented to the Graduate School of Arts and Sciences of Washington University in partial fulfillment of the requirements for the degree of Master of Arts December 2011 Saint Louis, Missouri Abstract Many spelling errors in English are doubling errors, as when people are stumped by the double ‹l› in ‹trellis›. In Study 1, we tabulated statistical patterns with regards to doubling in English. In Study 2, we collected behavioral data to see if people were sensitive to these statistical patterns in doubling and to explore other factors that might influence doubling such as context, individual differences (language background and spelling ability), and task. We gave two nonword spelling tasks to US college students (N=68) and bilingual Singaporean college students from an English-based education system but with diverse language backgrounds: Mandarin (N=54), Malay (N=44), or Tamil (N=42). In the choice task, participants heard a nonword and chose between two spelling options, e.g. dremmib/dremib. In the free task, they wrote down its best spelling. We found a vowel length effect (more doubling after short vowels than long vowels) that was moderated by spelling ability (better spellers were more influenced by vowel length) and language background. Americans had the largest vowel length effect and Tamil Singaporeans had none, as they possibly associated consonant doubling with the lengthening of doubled consonants in Tamil instead of the preceding vowel. The Mandarin group spelled nonwords least accurately, and greater knowledge of pinyin, a phoneme-based writing system, was associated with higher nonword spelling accuracy. These and other findings reflect how linguistic factors and language background moderate the role of statistical learning and context in spelling. Keywords: spelling, context, statistical learning, language background, doubling ii Acknowledgements I would like to thank the following for all of their help and advice throughout the entire process: My advisor, Rebecca Treiman; Members of my Master’s committee, Brett Kessler and Dave Balota; My programming and Python consultant, Mohammed Ansari; Nicole Rosales, for assisting in the data collection in the US; Members of the Reading and Language lab; My mum, for being the best mother in the world and my pillar of strength, for accompanying me during the experimental sessions, and for constantly including me in her love and prayers; and lastly, God, for all these wonderful things could only have happened with His grace and blessings. iii Table of Contents Abstract ............................................................................................................................... ii Acknowledgements ............................................................................................................ iii Table of Contents ............................................................................................................... iv List of Tables ..................................................................................................................... vi List of Figures ................................................................................................................... vii Speling ―Successful‖ Sucesfuly: Statistical Learning in Spelling ...................................... 1 Doubling of Medial Consonants in English .................................................................... 5 Patterns in the Doubling of Medial Consonants.............................................................. 7 Influence of Language Background on Spelling ........................................................... 11 Overview of Studies 1 and 2 ......................................................................................... 18 Study 1 .............................................................................................................................. 20 Method .......................................................................................................................... 20 Results ........................................................................................................................... 21 Discussion ..................................................................................................................... 23 Study 2 .............................................................................................................................. 24 Method .......................................................................................................................... 25 Results ........................................................................................................................... 30 Discussion ..................................................................................................................... 37 General Discussion ........................................................................................................... 44 References ......................................................................................................................... 48 Appendix A ....................................................................................................................... 79 iv Appendix B ....................................................................................................................... 80 Appendix C ....................................................................................................................... 82 Appendix D ....................................................................................................................... 84 Appendix E ....................................................................................................................... 87 v List of Tables Table 1. Statistical Patterns Pertaining to Proportion of Doubling of Medial Consonants in Disyllables (N = 8790) ................................................................................................. 61 Table 2. Statistical Patterns Pertaining to Proportion of Doubling of Medial Consonants in Disyllables (N = 8790) ................................................................................................. 62 Table 3. Mean Individual Characteristics across Groups in Study 2 ............................... 63 Table 4. Bigram Frequency for the CC Bigrams in Study 2 ............................................ 64 Table 5. Proportion of Doubling of Medial Consonants as a Function of Vowel Length, Number of Letters in First Vowel Spelling, Task, and Language Group ........................ 65 Table 6. Nonword Spelling Accuracy across Language Groups and Spelling Ability .... 66 Table 7. Nonword Spelling Accuracy across Spelling Ability and Pinyin Accuracy (for Mandarin Group) ............................................................................................................. 67 vi List of Figures Figure 1. Proportion of doubling of medial consonants for vowel length across spelling ability (for all groups). ...................................................................................................... 68 Figure 2. Proportion of doubling of medial consonants for the interaction between spelling ability and vowel length across tasks. ................................................................. 69 Figure 3. Proportion of doubling of medial consonants for vowel length across language groups. ............................................................................................................................... 70 Figure 4. Proportion of doubling of medial consonants for bigram frequency across vowel length. ..................................................................................................................... 71 Figure 5. Proportion of doubling of medial consonants for bigram frequency across tasks. ........................................................................................................................................... 72 Figure 6. Proportion of doubling of medial consonants for the interaction between bigram frequency and vowel length across tasks. ......................................................................... 73 Figure 7. Proportion of doubling of medial consonants in tasks across spelling ability. . 74 Figure 8. Nonword spelling accuracy across vowels for Americans and Singaporeans. . 75 Figure 9. Proportion of doubling of medial consonants for vowel length across pinyin accuracy (for Mandarin group). ........................................................................................ 76 Figure 10. Proportion of doubling of medial consonants for the interaction between vowel length and pinyin accuracy across tasks (for Mandarin group). ............................ 77 Figure 11. Proportion of doubling of medial consonants for bigram frequency across pinyin accuracy (for Mandarin group). ............................................................................. 78 vii Speling ―Successful‖ Sucesfuly: Statistical Learning in Spelling Spelling is an important skill. In many cultures, people need to be able to spell their names correctly in application forms, their addresses, or the things they need in a shopping list. Spelling errors could result in disastrous consequences, such as the wrong medication being given to a patient or a misunderstanding of the writer’s intention. Even people who use a spell-checker may still make spelling errors. A case in point is spelling errors that are related to doubling. After all, the spell-checker cannot tell if the writer intends to spell /kænən/ as ‹cannon› or ‹canon›, both of which are plausible spellings but with different meanings. How then do people spell? Statistical learning has been widely used as a theory to explain language acquisition and learning (e.g. Saffran, 2002; Saffran, 2003) and, more pertinently, spelling (e.g. Treiman & Kessler, 2006; Kessler, 2009; Treiman et al., 2007; Pacton, Fayol, & Perruchet, 2005; Pollo, Kessler, & Treiman, 2009; Kemp & Bryant, 2003; Deacon, Conrad, & Pacton, 2008). According to Kessler (2009), the statistical learning perspective postulates that ―complex sound-spelling correspondences are learned the same way many other patterns are learned in life: by observing and internalizing the relative frequency with which objects and events occur and co-occur‖ (p. 20). Through repeated exposure to text, people gradually pick up sound-spelling correspondences, without necessarily formulating any conscious accounts of them, making an implicit numerical analysis where orthographic patterns are essentially observations (or computations) that some sound-spelling correspondences are more frequent than others (Kessler, 2009). Indeed, people seem to be sensitive to statistical patterns in the language 1 when spelling and such sensitivity begins quite early in life. For example, Pollo et al. (2009) found that even the prephonological spellings of children as young as 4-years-old are not random strings of letters but reflected a sensitivity to the frequencies of letters and bigrams in the children’s language. Following this theory of statistical learning, models of spelling consider the role of learning of statistical patterns or correspondences from exposure (e.g. Barry & Seymour, 1988; Kreiner & Gough, 1990; Kreiner, 1992). Barry and Seymour (1988) postulated that there exists a set of probabilistic sound-to-spelling mappings that relate phonemes to weighted lists of alternative spelling patterns, and the mapping that ends up being selected for spelling production is the more common one. These sound-to-spelling mappings are ordered by a continuum of sound-to-spelling probabilities, i.e., the frequency with which spelling patterns represent phonemes in words. Thus, the frequency of a spelling pattern can range from high, as with the correspondence between /æ/ and ‹a› (99% of the time), to middle, as with the correspondence between /i/ and ‹ea› (40% of the time), to low, as with the correspondence between /i/ and ‹ie› (6.7% of the time). The probability with which people pick a particular letter or letter group when spelling should be related to its overall frequency. Indeed, Barry and Seymour (1988) found that, when asked to spell nonwords, spellers produced more common spelling patterns of vowels than rare spellings. Furthermore, the probability of the least typical correspondence in a word should predict spelling difficulty. Kreiner and Gough (1990) found evidence to support this—college students found it more difficult to spell /ə/, which has many alternative spellings with similar probabilities, than other vowels. 2