Sara’s Study — Early Results

Read this:

“jʊər ʤʌst ə ʧaɪld”

The sentence above is a simple English sentence written in the International Phonetic Alphabet (IPA). I am using the IPA to conduct a study (ipa.sara.ai) in computer-aided learning through immersive reading of an unfamiliar orthography. Simply put, I’m asking adults to re-enact the role of emergent reader by learning to read again. IPA is foreign enough to allow you the opportunity to learn to read again without learning a new language.

There is a dispute over the best way to teach children to read. Just a few days ago the New York Times published an opinion piece on this dispute. This dispute goes back generations and was particularly heated in the late ’80s early ’90s in what is known as the Reading Wars. The two sides are illustrated simply by the two images below, which are screen captures of query interactions pulled from my study.

On team-phonics we have those who insist that research evidence supports a phonemic recoding approach to instruction. Here, ‘ jʊər ’ is taught by deconstructing it into alphabetic code, j–ʊ–ə–r, and each character is recoded into a phoneme. Enough practice and recoding the alphabetic code from letters to sounds becomes second nature (so the theory goes) and voilà — children can read.

On team-whole-language we have those who believe in supported or aided reading for meaning, de-emphasizing or ignoring instruction in phonemic recoding. In the current study, I can support learners by pairing the word coded in an unfamiliar orthography (“ʤʌst”) with the same word coded in a familiar orthography (“just”).

The current study can directly compare the results of the two. In theory, if team-phonics is right then the first option for learning to read jʊər should out-perform the second option for learning to read ʤʌst. I’m putting users through each method, side-by-side, in a 6,600-word assignment. I don’t have enough data to make this comparison at the moment, but I suspect team-phonics will win out, especially because the students already have developed phonemic awareness and are familiar with the process of sounding-out words.

But this battle isn’t the current study’s focus. There is a true middle way — not both, balanced or blended, but a fundamentally different approach. This middle way is the subject of this study. It is illustrated in this screen capture of a third interaction method.

This query response, shown above, begins with a matching task rather than a reveal. The user is prompted to match ʧaɪld in the heading to one of the three options in blue. Like the second option for learning to read ʤʌst, the standard orthography for ʧaɪld (“child”) will be revealed after the matching task is completed correctly. This may seem like a pointless hurdle, but it actually serves an important purpose. This cognitive work is actually more difficult than it appears, but it engages the brain in a similar manner as the deconstruction of jʊər into the characters j–ʊ–ə–r.

In 1200 of these word-match queries by different study participants, the user chose the wrong match 3 out of every 22 tries. As an example, the choices offered for ‘ baʊ ’ which did result in a mismatch in 4 out of 15 attempts are:

boʊ
baʊ
daʊ

This cognitive work, engaging attention to each character of the lexigram, brings much if not all of the cognitive benefit of the more complex task of phonemic recoding to the simpler practice of aided reading, or at least that’s the theory I’m testing in this study. The results detailed below speak for themselves.

User X, represented by the zig-zagging yellow line in the chart below, was assigned the second query method, the one used above for learning to read ʤʌst. The y-axis is the percentage of words on a particular screen that the user has already begun to become familiar with — this familiarity threshold is set at 1–4 previous encounters with the word, depending on the nature of the encounter(s). For User X, as the words become more and more familiar, the reading pace (words per minute, on the x-axis) remains relatively unchanged.

For Users 1–4, the assigned query method was the third, word-match option. The red trend line shows the exponential gains in reading pace as the words become more familiar.

In total, Users 1–4 encountered 18,445 words and initiated 826 queries. Users 1–4 typically queried an unfamiliar word only once, usually on the first encounter, with the queries per queried-word averaging 1.16. Users 2 and 4 completed the entire course, and their progression can be seen in the data below. Screens 2, 12, and 16 (bolded) all had the exact same text, separated by at least 900 words in the intermediating Screens.

The screens are numbered in the chronological order they were presented. You can see the direct relationship between word familiarity and reading pace, particularly from Screen17 to Screen18. Queries were disabled on Screens 19 and 20 to test unsupported reading ability and pace, and these final screens included comprehension assessments after the reading (both users answered 9 out of 9 questions correctly). Note the profound effect of familiarity on Screen20 vis-a-vis Screen19. These users received no explicit instruction on the sound–letter correspondences of the International Phonetic Alphabet.

User2 — Screen2 — 20 wpm — 14% familiar — queries: 11
User2 — Screen3 — 35 wpm — 18% familiar — queries: 3
User2 — Screen4 — 21 wpm — 27% familiar — queries: 15
。。。
User2 — Screen12 — 53 wpm — 69% familiar — queries: 2
。。。
User2 — Screen16 — 91 wpm — 76% familiar — queries: 0
User2 — Screen17 — 70 wpm — 72% familiar — queries: 1
User2 — Screen18 — 55 wpm — 68% familiar — queries: 1
User2 — Screen19 — 84 wpm — 78% familiar — queries: 0
User2 — Screen20 — 51 wpm — 62% familiar — queries: 0

User4 — Screen2 — 25 wpm — 11% familiar — queries: 13
User4 — Screen3 — 40 wpm — 20% familiar — queries: 5
User4 — Screen4 — 30 wpm — 29% familiar — queries: 22
。。。
User4 — Screen12 — 100 wpm — 73% familiar — queries: 0
。。。
User4 — Screen16 — 118 wpm — 78% familiar — queries: 0
User4 — Screen17 — 76 wpm — 74% familiar — queries: 11
User4 — Screen18 — 66 wpm — 69% familiar — queries: 17
User4 — Screen19 — 112 wpm — 78% familiar — queries: 0
User4 — Screen20 — 27 wpm — 63% familiar — queries: 0