II. EXPOSICIÓN DE NECESIDADES
3- Necesidades materiales a) Edificios
For a study on L2 intonation, the "94 sentences with various intonation patterns"
component of the corpus would seem a natural choice. This component consisted of the sentence sets in the following table, labeled (under "Type") with direct translations from the corpus's documentation. (The parenthesized number in the last column indicates contextualizing sentences, e.g. where learners were recorded reading Is it John who writes poetry? before the target sentence No, it is Bill who writes poetry.
Type Example Number
Phrase-level stress take care of themselves 20
Chunking-related intonation differences I don't know(,) Miss Brown. 16
Intonation when enumerating (rising or flat)
Cauliflower, broccoli,
cabbage, sprouts, and onions. 4 Classification
by tone
Rising tone When I came, he greeted me warmly. 2
Falling-rising tone I should go. [But I don't think I will.] 2
Rising-falling tone Who knows? [= Nobody knows.] 6
Flat tone He drank, he stole, he was soon despised. 1
Classification by sentence pattern
Declarative Legumes are a good source of vitamins. (2+)3
Yes/no interrogative Are legumes a good source of vitamins? (1+)5
Alternative interrogative Is this elevator going up or down? 4
Wh interrogative How long have you been waiting? (3+)6
Tag question I haven't seen you before, have I? 3
Exclamative Isn't it wonderful weather! 2
Differences based on
old vs. new information No, it is Bill who writes poetry. (7+)7
Total: (13+)81
Table 5.2: Structure of materials in the ERJ's 'Sentence, Intonation' component
While such materials are attractive for tapping learners' knowledge of intonation, the metalinguistic annotations that accompanied these sentences are problematic for the present
study's purposes. In particular, these sentences were accompanied by arrows indicating where one's pitch should rise or fall, which were shown during practice as well as the actual recording. While, as described above, such annotations were well-suited to the corpus's original purposes, they are inappropriate for a study on naturalistic interlanguage development. The core problem is that, with such annotations, it is impossible to know exactly how much of a given utterance produced by a learner reflect their implicit phonological knowledge and how much was explicit knowledge 'learned' on the spot from rehearsing with the metalinguistic annotations on the page (and is therefore merely an artifactual byproduct of the elicitation process itself). Since the goal of the present study is to understand each learner's phonological knowledge, it is of greatest interest what an L2 learner would do in normal circumstances, i.e. without any such
metalinguistic annotations.
Due to these considerations, another part of the overall corpus needed to be selected. Since similar remarks apply to the "121 sentences with various rhythm patterns" (namely the use of rhythm marks), and since rhythm is intertwined with intonation in various ways, the entire [Sentences, Suprasegmental] component of the corpus was thus rendered unusable. Moreover, the [Words, Segmental] and [Words, Suprasegmental] components to the corpus are a poor fit because the focus largely on individual words whereas the present study is concerned with utterance-level intonation. These criteria left only the [Sentences, Segmental] component of the corpus, which was ultimately adopted for as the dataset for the present study.
Recall from above that this subset consists of 592 sentences with three sub-components: 460 phonemically balanced sentences adapted from the TIMIT corpus, 32 sentences with phonotactic sequences that are difficult for Japanese EFL learners, and 100 sentences designed
by actual English teachers in Japan for testing pronunciation.2 Since the present study is concerned with intonation, not segmentals, the three-way contrast between these different sub- types of sentence is not relevant for the present study. As such, throughout the rest of the present dissertation, the distinction between these three subgroups is collapsed (such that all three are treated together as a single 592-sentence whole).
Since this component of the corpus was focused on segmentals, during the practice phase, in addition to the normal English orthography for each sentence, learners saw the ASCII
phoneme codes modified from TIMIT, segmented into words by square brackets. (An example of this sort of representation can be seen in last line of (1) above.) However, Minematsu et al. (2002a:3) reasoned that "reading sentences with referring to phonemic symbols is expected to induce unnatural pronunciation"; more specifically, "[w]ith the phonemic symbols for each word, some speakers may not read a sentence but a sequence of isolated words." As such, no phonetic symbols were shown during the actual recording session, i.e., every sentence was elicited with only its normal representation in English orthography.
This particular method of elicitation is not particularly problematic for the purposes of the present study. Given the markedly foreign appearance of the TIMIT transcriptions, many learners may not have even looked closely at the phonetic symbols when practicing the stimuli, and even the learners who did may not have gained much from it (especially since the
transcription was unavailable during the actual recording itself). Even if a learner was influenced by the transcriptions, since they contain segmental information (plus stress levels), this should
2 Two of the 100 "sentences" designed by English teachers for testing pronunciation are technically sequences of two sentences (produced by the same learner in sequence). These are the following: (1) "Where do you live? I live in the woods." and (2) "I counted the dogs. There were ten of them."
mostly influence their production of segmentals. In fact, since the learner's explicit attention was drawn to something other than prosody during the recording of these sentences, learners'
production of these sentences can be thought of as tapping each learner's 'unguarded' implicit knowledge of prosody (i.e. relatively uninfluenced by conscious metalinguistic awareness). In this sense, these stimuli can actually be thought of as ideal for a study of intonation.
What is crucial for the purposes of the present study is that, for these sentences, learners were never given any metalinguistic annotations regarding intonation (e.g. curved arrows pointing up or down). Of course, the learners saw such annotations when producing their piece of the "94 sentences with various intonation patterns" component of the corpus (both during the practice as well as the actual recording). However, it is unlikely that this kind of brief exposure to arrows printed on the page would have a sufficiently deep impact on the learners as to
generalize to the other sentences elsewhere in a given student's list, especially with their attention drawn to segmentals. Thus, the method of elicitation for the selected set of 592 sentences is at least compatible with - if not ideal for - the purposes of the present study.
Recall from above that the words were grouped into 5 lists and the sentences were
grouped into 8 lists, and each learner was randomly assigned one word list plus one sentence list. Each of the eight sentence lists contained different (non-overlapping) subsets of the selected 592 sentences, ranging in size from 57 to 78 sentences each. The number of learners reading each of the eight lists ranged from 22 to 28 learners. (No individual learner read more than one list.) This means there are 22-28 data points from Japanese EFL learners for any given sentence. The full details about the number of sentences and the number of learners for each list are provided below.
Number of sentences Number of learners List 1 76 26 List 2 76 25 List 3 76 22 List 4 76 23 List 5 78 26 List 6 78 27 List 7 78 25 List 8 57 28 Total: 592 202
Table 5.3: Number of sentences and learners for each of the eight lists
The structure for the native speakers was much simpler. Since the entire set of materials were split into two equal-sized halves (list 'X' and list 'Y') for these speakers, each of the two lists had half (296) of the 592 sentences. The number of speakers also matches that of the overall corpus: 9 native speakers reading list X, 9 reading list Y, and 2 reading both. This means that, for any given sentence (from either list X or list Y), there are always exactly 11 native-speaker productions to serve as the control.