thethesisis full of errors!

greetings, gentle readers! i apologize, once again, for the long hiatus. this post will give a brief preview of the fruits of the labor that has kept me from blogging ere these long weeks. my thesis about typos is nearing completion, and, believe it or not, i actually have some interesting conclusions to show for it! what follows is a synopsis of what my thesis will become over the next couple of weeks... *knocks on wood*.

many of the studies of typos i read are inadequate to describe naturalistic typing (that is, speech over IM, for instance), because they test people doing transcription typing, or they gather their corpora (sets of linguistic material--in this case, errors) from published works. neither of these data sets is appropriate for comparing the errors made in typing to the errors made in speaking; published typos are very selective because they are only those that escaped the notice of several proof-readers and editors. transcription studies don't work because they involve a perceptual element, reading, that a) doesn't occur in speech and b) might change the types of errors made in ways that are, as yet, unpredictable.

so i brought 20 undergrads into the lab and made them (paid them to) have IM conversations. they used software that recorded each of their keystrokes including the backspace key. These 10 conversations generated almost 350 typos, including things that were clearly 'edits' (planned utterances that were deleted before being sent). My spreadsheet called 'Typos that Matter' contains 192 rows. These are typos that involve individual letters or sounds. They will be the focus of my study!

one of the main observations made about speech errors is their striking regularity. for example, in speech, consonants only slip with other consonants and vowels only slip with other vowels. Gary Dell, one of the big speech error guys, uses a set of four rules, including this consonant-vowel category effect, to identify 'human-like' speech errors. Other rules include the phonotactic regularity effect--speech errors do not cause combinations of letters that are illegal in the language being spoken. For example, in English, you can't begin a word with 'tk,' so that combination never occurs at the beginning of a word as a result of an error.

it's immediately apparent that typos do not obey these same regularities. adjacent letters often slip, regardless of whether they are consonants or vowels, and illegal combinations occur fairly frequently.

another important difference concerns the volume of errors in speech versus typing. I brought half as many participants into the lab to have conversations over the phone. they produced only 5 Errors that Matter. the ratio of speech to typing errors in this setting is something on the order of 1 to 20.

both of the above facts are evidence that typos occur because of very different processes than speech errors do. however, the fact that made me want to study this in the first place is the similarity of some typos to speech errors. there is a set of typos that behave almost the same way as speech errors--these errors are of a specific type known as 'non-contextual substitutions.' in these cases, a letter is removed completely and replaced with another letter that does not exist in the surrounding utterance. "fright" typed for "bright" is a good example of this, whereas "rbight" for "bright" is a contextual exchange.

when i examined only single letter non-contextual substitution errors, a striking pattern emerged: they follow Dell's rules very very closely, almost entirely, as speech errors do. except one rule to which i'll return later.

since one class of errors follows cognitive constraints while others seem to ignore them completely, i plan to propose that there are two separate mechanisms responsible for typos. non-contextual substitutions (and perhaps other types of errors) are caused by errors in planning and internal speech--the same things that cause spoken errors. exchanges and other types of errors occur after planning at a lower level of processing and aren't susceptible to the filters or structures that make phonotactic and consonant-vowel category violations impossible. that's pretty neat!

the Dell rule that non-contextual substitutions do violate with abandon is known as the initialness effect. in speech errors involving consonants, the first consonant in a word is much more likely to be involved in an error than consonants at the end or middle of words are. the same is not true for typing errors! in fact, of the 22 non-contextual substitutions involving consonants that i've checked for consistency with Dell's rules, 16 violate the initialness effect. this is a striking effect that might point to a major difference in processing and planning across the two media.

so, not only are there different articulatory problems with typing than there are with speech, but perhaps the medium in which we do our articulation actually effects the way we plan and retrieve speech. i would try to claim this as empirical evidence that contemporary media theorists are correct--that communication technologies effect our cognition in a profound way--but that's a lot to say in 20 pages, and linguists don't take kindly to much of my favorite media theory...

look for a future post on plans for further research!

science!

mike, a fellow student in my child language acquisition class, has posted an experiment (part of his final project) online... it strikes me as a fun task, and you can participate, even if you're not a child!

it's about the way we learn to recognize words in speech. i won't say too much else now, or risk biasing potential subjects, but perhaps after finals, i can recruit him to say a bit about his results.

enjoy!
test 1,
test 2,
test 3

ps - seems there's some paypal (as opposed to cash) money involved.

text errors

i'm so excited! forgive me if my fast tyoing leads to a few errors now and again...

iit's official! i'm starting an indepeendent study on typos. a lot of luinguistics research is done on the nature of speech errors, anbd they reveal a lot about the way we store, retrieve, and rpcoscess speech. but it's only recently that typing has become an isntantaneous form of communication. this raisse all kinds of interesting questions about what the nature of our typung errors reveals about those same processes--how is typing differentf rom speech? how si it the same?

i won't be testing a specific hypothesis in my study, partly because there's sso little research available on this tpic right now. i'll just be collecting typing inj a laboratory setting (not just the finished product, but every keystroke using some sort of keylogger software), and trying to categorize the different types of errors epoeple make.

most typos, i assume, will be explainable on the basis of the keyboard's design. that is, most letters that are insterted will probably be near the intended letter on the keyboardd. a lot of metathesis (reversal of two lettres) too, probably. these kinds of things can be explained by the motor processes involved in using the keyobard itselv.

however, what if there are other errors, maybe ones that odn't freqyuently make it into the final text (even in IMs or other informal settings), that couldn't be eexplained by proximity or other motor functions? what if someone swithced a 'b' for a 'p' or typed 'thear' instead of 'there'? these kinds of errors would reveal something really really relaly intersting about the processes we go through before generating text. specifically, we could hypothesize that certain errors would reveal that a phonological (sound) pattern is generated before motor control is initiated; that is, it might demonstrate that we do hear words, in some sense, before typing them.

other errors could shed light on just how motor (i just typed 'mother' there, i swear it!) control is accomplished. for instance, if someone frequently typed 'the' and then deleted the 'e' in words that used 'th' like 'that' or 'this', we might hypothesize that letters are stored and retrieved in 'chunks' of some kind rather than letter-by-letter. or typing 'buutton' for 'button' (typing mistaken double letteres more frequently in words that have actual double letteres) might mean that there is a command for 'type two letters' that is supposed to go with the command 'type a 't',' but gets misaligned.

mayube i'm the only one that maes these kinds of errors... but i think we've all had that mysterious 'why on earth did i type that?' experieince. maybe there are other even more itneresting errors out there!

there's also a lot of research on self-monitoring and repair in speech--what errors do we catch ourselves making? what errors do we block before we make them? how do we monitor our own speech; do we hear it the way we hear other people's speech? thed same questions cam be casekd about typing, so i'll be looking at which errors people are likely to correct/notice before they push 'send', how long it takes to catch different types of errors, and so on.

lots of big questions!, so don't be surprised if you see more posts on this one!

sentence complexity

let me briefly and profusely apologize for our extended absence. the last month or so has been filled with vacations, moving, job hunts, comments on my last post, and so on. enough with this 'sorry' blather! on to language...

speaking of jobs, i'm spending the summer as a research assistant in a cog sci lab. i'm helping write/assess a government adult literacy and vocabulary test. that's right! i'm one of those people that makes standardized tests! i think there was a point in my life when i thought it would be an interesting career... i wasn't totally wrong.

so far i've written 20 questions. each focuses on one word with multiple meanings (usually including more than one part of speech). i write three fill-in-the-blank sentences that would make sense with the target word in the blank, and then i write three 'matching' sentences that would NOT make sense with the target word in the blank. but in all cases, the wrong sentences have to match in missing part of speech and all the sentences have to use the same form of the word (that is, no plurals or past-tense or suchwhat).

now i'm collecting a wide range of statistics on each possible answer sentence so that we can make sure, for example, that the wrong answers aren't systematically different from the right answers.

i'm running each sentence and all the sentences for each question together through a readability statistics website that gives me the Flesch-Kincaid reading ease, grade level, Gunning-Fog index, and average number of syllables per word.

i'm also taking a list of statistics about what i'm calling the 'syntactic complexity' of each sentence. these include number of prepositional phrases, number of strings of modifiers, number of non-main clauses and so on. it would be really interesting if any of these statistics ended up correlating with the response time we will empirically collect for the sentences.

interestingly, since the standard readability tests are partially based on the average number of syllables per word, lots of grammatical words (prepositions, quantifiers, conjunctions, etc) tend to yield low Flesch-Kincaid grade levels, but tend to make my 'syntactic complexity' score very high. this type of complexity also correlates better with sentence length than the readability stats do, since you can't have a syntactically complex sentence that's only four words.

no real conclusions to draw yet, except to say that a lot of these grammatical categories are tough to define, and sometimes come down to my research-assitant judgement.

if anyone out there in linguo-blog land knows of any official measure of syntactic complexity or any measure that's been used to predict reading times, send it on in!