README FILE FOR SUPPLEMENTARY DATA TO ALBIN (2015) Aaron Albin (usagi5886@hotmail.com) 8/25/2015 --------------------------------------- DESCRIPTION OF DISSERTATION The files in the attached dataset accompany the following dissertation (cf. footnote 15 on page 164): Albin, A. (2015). Typologizing native language influence on intonation in a second language: Three transfer phenomena in Japanese EFL learners. (Doctoral dissertation). Indiana University, Bloomington. The dissertation is freely available in the IUScholarWorks repository at http://dx.doi.org/10.5967/K8JW8BSC In the empirical component of the dissertation, three kinds of cross-linguistic transfer relating to intonation are examined in Japanese learners of English as a Foreign Language (EFL). a) Research Question 1 investigates how often Japanese EFL learners produced a peak suggestive of an L1 Japanese phrasal H- on an L2 English word with late primary stress (e.g. 'ambidextrous') at the beginning of a sentence. b) Research Question 2 ascertains whether the pauses in long (10+ word) sentences are preceded by a low F0 target in Japanese EFL learners' productions more often than native speakers due to transfer of L1 Japanese phrasal L%, which is associated with a minor phrase (as opposed to a 'major phrase' for English L-). c) Research Question 3 examines Japanese EFL learners' productions of the utterance-final rise in yes-no questions whose final word ends with one or more unstressed syllables (e.g. 'photograph') - in particular, how often the rise begins within the final syllable (due to transfer of the alignment of L% from L1 Japanese). For each research question, fourteen sentences were selected from a large L2 speech corpus, and productions of each sentence by approximately 20-30 learners and 11 native speakers were analyzed. The corpus used for this purpose is the English Speech Database Read by Japanese Students (ERJ), which contains recordings of Japanese EFL learners reading aloud word and sentence materials (plus baseline data from native speakers of American English). Further information on this corpus can be found at http://research.nii.ac.jp/src/en/UME-ERJ.html --------------------------------------- DESCRIPTION OF DATASET The ERJ itself consists only of the soundfiles themselves (plus accentedness/intelligibility judgments from 2-5 native American English raters). For the purpose of the conducting analyses for the dissertation described above, between September 2014 and February 2015, several kinds of annotations were created to accompany these soundfiles. Three such annotations are included in the attached dataset: a) F0 Ranges: The F0 floor and F0 ceiling values used in the process of extracting F0 out of the soundfiles - more specifically, in the process of creating Pitch objects using PraatR (http://www.aaronalbin.com/praatr/) b) Segmentation: Timestamps of the divisions between segments/words (depending on the research question) in the crucial region of each utterance, along with the corresponding labels (either in IPA or normal orthography). c) Stylizations: A schematized representation for each soundfile's F0 contour, produced using the package 'intonation' (https://github.com/usagi5886/intonation) for the R programming language (https://www.r-project.org/). All of these are stored as tab-delimited tables in plain text files. This format facilitates analyzing the data with a scripting language like R or Python. However, the contents of these files can also easily be copied-and-pasted into a spreadsheet program like Microsoft Excel for analysis. --------------------------------------- FILE LIST Each of the above three kinds of annotation is available in separate files for each of the three research questions. Thus, with three kinds of data for each of the three research questions, there are 9 total files, structured in a 3x3. Accordingly, the list of files in the attached dataset (along with each's file size) is as follows: (The last file is a copy of this README.) ./Albin2015_ResearchQuestion1_F0Ranges.txt 15 KB ./Albin2015_ResearchQuestion1_Segmentation.txt 167 KB ./Albin2015_ResearchQuestion1_Stylizations.txt 163 KB ./Albin2015_ResearchQuestion2_F0Ranges.txt 15 KB ./Albin2015_ResearchQuestion2_Segmentation.txt 444 KB ./Albin2015_ResearchQuestion2_Stylizations.txt 675 KB ./Albin2015_ResearchQuestion3_F0Ranges.txt 16 KB ./Albin2015_ResearchQuestion3_Segmentation.txt 117 KB ./Albin2015_ResearchQuestion3_Stylizations.txt 123 KB ./README.txt 1 KB These files have been zipped into a single archive using 7-Zip File Manager (http://www.7-zip.org/). --------------------------------------- CONTENTS OF FILES Inside these files, the first column is always 'Path', indicating which ERJ soundfile the annotations are intended to accompany in the form of a file path on the ERJ DVD. An example of one such file path is 'JE/HIR/F05/S1_050.wav'. All file paths point to sub-directories within the 'wav' folder on the ERJ DVD, hence this points to './wav/JE/HIR/F05/S1_050.wav'. These file paths provide a common point of reference tying together the three different kinds of annotation for any given soundfile. The number of soundfiles analyzed for each research question is as follows: a) Research Question 1: 481 files, as broken down in Table 5.4 and the accompanying discussion (p.139-140) b) Research Question 2: 517 files, as broken down in Table 5.5 and the accompanying discussion (p.142-143) c) Research Question 3: 501 files, as broken down in Table 5.6 and the accompanying discussion (p.145-146) Note that these counts all exclude the empty dummy files referenced in footnotes 3, 5, and 7 of Chapter 5. Besides 'Path', the remaining columns depend on the kind of annotation: a) F0 Ranges: Floor, Ceiling The bottom and top of the F0 range b) Segmentation: Label, Time_ms The label for a given interval and the timestamp (in milliseconds) of the division *after* that interval Note: The first row for a given soundfile always has an 'NA' label since that represents only a division (i.e. the word/segment onset) without any associated label. c) Stylizations: nthTransition, Index, F0, Time_ms, Threshold, Gradience, MAD, PercentVoiced See Chapter 3, especially section 3.5 (p.89ff) of Albin (2015) for details, but in brief: - Index: the index for this vertex (i.e. which frame in the Pitch object corresponds to this vertex) - F0: The F0 value (in Hertz) for this vertex - Time_ms: The timestamp (in millseconds) of this vertex - Threshold: The threshold value of the F0 transition leading up to this vertex - Gradience: The gradience value of the F0 transition leading up to this vertex - MAD: The intensity-weighted median absolute deviation, describing the goodness of fit of the stylization to the raw data (with smaller values indicating better fit) - PercentVoiced: A number between 0 (0%) and 1 (100%) indicating the proportion of the frames in the transition leading up to this vertex that are voiced --------------------------------------- ANTICIPATED USES OF DATASET Performing the analyses in Albin (2015) require three components: - The attached dataset - A copy of the ERJ, cf. registration form at http://research.nii.ac.jp/src/en/register.html - The R functions distributed at https://github.com/usagi5886/intonation By assembling these three components, it is theoretically possible to reproduce all of the analyses in the dissertation. As such, making the attached dataset publicly available helps takes an important step toward scientific replicability (e.g. by checking the validity of the segmentations and stylizations). Even without a copy of the ERJ, however, it is possible to perform various ad-hoc analyses to better understand the empiricial results presented in Albin (2015). The set of stylizations can also provide a testing ground for future expansions of the methodological framework outlined in Chapter 3. --------------------------------------- COPYRIGHT AND LICENSING INFORMATION All annotations contained in the attached dataset were created by Aaron Albin (usagi5886@hotmail.com). The contents of this dataset are licensed under an Open Data Commons Attribution License (ODC-BY). For details, see http://opendefinition.org/licenses/odc-by/ The ERJ itself was compiled through a collaborative effort between Nobuaki Minematsu and colleagues. The copyright holder for the ERJ is Seiichi Nakagawa.