This time I tried to use the SpeechRater implementation at My Speaking Score to produce the highest scoring “short” answer I could come up with. Once again, my answer is just 74 words in total. You can hear it below:
Here’s a transcript:
Clearly, it’s advantageous to study various subjects while we are at university. This is because it makes it possible to find our true passion. For instance, when I was a freshman, I took courses in chemistry, psychology and history. While I was enamored with all of them, eventually I discovered that I was most fascinated by chemistry. As a result, I ended up majoring in that. Moreover, I gained interdisciplinary insights throughout my journey.
This is almost the same as the last answer. I’ve improved the vocabulary a little by using slightly less common words. I added an -ly adverb to the beginning.
The main difference is that I delivered this in a natural speaking voice at a normal pace. That means I finished after only 30 seconds. As you will recall, last time I slowed it down to a unnatural pace and finished at 45 seconds.
The result… a score of 3.46, or 27/30. Wow. This isn’t a fluke. I recorded a similar (but slightly different) answer of 80 words and got a score of 3.6.
The fluency markers are all fantastic:
But how can that be? I scored in the 87th percentile for response length, but scored in the 14th percentile for response rate last time… with the same number of words. My guess is that the SpeechRater compares the number of words in the answer only to responses of the same length. Do you get what I mean? My answer was compared only to other 30 second answers and I included more words than most of them. Again, I want to stress that this isn’t a fluke. My 80-word answer got a similar response length score.
Perhaps we as teachers we should stress pace rather than word count.
Pronunciation and Vocabulary
Moving on, here are the pronunciation and vocabulary scores:
The pronunciation is better this time, but it still didn’t really like my rhythm. That’s a flukier score, I think. My 80-word answer was in the 98th percentile. I think the SpeechRater isn’t perfect when it comes to measuring that dimension. Perhaps I will make 20 unique recordings of the same answer one day and track the scores.
The vocabulary scores are okay. Perhaps my low word count limited my ability to use a lot of uncommon words in the answer. That makes sense.
Grammar and Coherence
My grammar and coherence scores are as follows
So there you have it. It is possible to get a good SpeechRater score on an independent response with a “short” answer. This does track with what ETS says about how the human raters check answers. ETS speakers have said again and again that it is not necessary to speak quickly on the test. I don’t know if this can be applied to the integrated speaking tasks, but I’ll check my notes about how many actual details need to be included in the answer. At the last presentation I attended it was implied that only half of the details from the lecture of a type three speaking question need to be included. Maybe thirty seconds is enough.
If any of my faithful readers want to sponsor a few test attempts I’ll march down to the test center and run these experiments in an actual testing environment.
Students often ask me how important it is to speak quickly in the TOEFL speaking section. Keen students even ask how many words they should include.
I’ve always said that speaking rate is really important. I’ve urged students to practice speaking quickly, as long as that doesn’t mess with their pronunciation or intonation. But that’s always just been my gut feeling, stated without solid evidence. In an effort to gather some real data on the issue, I submitted a few of my own practice answers to SpeechRater, the automated scoring software used (together with human raters) by ETS to grade the TOEFL. I was able to do this by uploading my answers to My Speaking Score, which has licensed SpeechRater. I encourage both teachers and students to make use of that site. It is comprehensive and fairly affordable. A monthly subscription gives you a bunch of credits to upload answers (or if you prefer you can just record them in your browser). Teacher and student accounts can be linked to facilitate reviews and personal feedback.
I must mention a few disclaimers before I get into the data:
The real test uses both a human rater and the SpeechRater. That means we cannot use SpeechRater alone to completely predict how a given answer would score.
While My Speaking Score is meant to be as close to the real TOEFL as possible, it is a third party implementation, so it cannot be perfect.
This is a third party blog, not associated with ETS. My interpretation of the numbers below might be totally wrong.
My Sample Answer
Now that I’ve got that out of the way, you should listen to my sample answer:
Here’s a transcript:
I think it is much better to study various subjects while we are at university. This is because it helps us to find our true passion. For example, when I was a freshman, I took courses in chemistry, psychology and history. While I loved all of them, eventually I discovered that I was most interested in chemistry. As a result, I ended up majoring in that. Moreover, I gained interdisciplinary insights along the way.
My speaking rate is very low. It is just 74 words (per the word counter in Google Docs). I generally recommend about 120 words if students want a perfect score, and about 100 words if they want an “average” score.
However, everything else is pretty good. My pronunciation and intonation are at a native level. I’ve included a few fancy words like “passion” and “interdisciplinary” and “insights”. I’ve also used transitions like “moreover” and “as a result.” I even included a few conjunctions like “while” and “when.” There aren’t any “umm” breaks, self-corrections or stutters.
When I asked experienced TOEFL tutors how this answer would score, I got responses ranging from 24 points to a perfect 30 points. As you can see, a few of the tutors really liked it!
Meanwhile, SpeechRater gave this one a score of 2.92/4. That converts to 23 points out of 30. Good… but not great.
What did SpeechRater Think about my Fluency?
First up, here’s the score for speaking rate.
Not surprisingly, the answer is all the way down in the 14th percentile. That means I spoke slowly. There is a penalty for that. But speaking rate is just one metric. It can’t account for all the entire seven-point penalty from the SpeechRater.
Here’s a look at the rest of the fluency metrics:
They aren’t good either. As you can see, the SpeechRater gave me a poor score for “sustained speech.” It identified a bunch of disfluencies and I ended up in the 27th percentile. In this case the disfluencies are silent pauses, but in other answers they might include “uhh” breaks. It also gave me a fairly poor score for metrics specifically related to pauses, as you can see. Slowness and pauses usually go hand in hand, as you might expect.
What did SpeechRater Think of my Pronunciation?
My pronunciation scores were a mixed bag:
Despite my slowness, my rhythm was pretty good. However, my pronunciation of vowels was merely average. But how can that be? I’m a native speaker. Well, another source of slowness is the way I sometimes draw out my vowel sounds. Notice my pronunciation of “and history” and “all of them” and “most interested”. The awkwardness is subtle, but noticeable if you are listening for it. The penalty for doing this is likely small, but I think it added up since I did it multiple times in every sentence.
What did SpeechRater Think of my Vocabulary?
Vocabulary was another mixed bag:
As I mentioned above, I think my answer has a few good words in it. However, I’m stuck in the 25th percentile for vocabulary depth. And even though I didn’t really repeat words, my vocabulary diversity score is merely average. Why? Well, my guess is that since my total word count is quite low, it is almost impossible for me to include a lot of “uncommon” words. I mentioned three words I suppose are “uncommon,” but that’s not really enough. In a more quickly delivered answer I might have had time for seven or eight words, and earned a better score in that domain. Likewise, a more quickly delivered would have almost automatically included a more diverse vocabulary… and earned a higher score.
A future experiment might involve jamming as many fancy words into an answer of the same length in an attempt to produce the best possible 75 word answer.
What did SpeechRater think of my Grammar?
Ooof. My grammar score is not good:
I’m all the way down in the 9th percentile. Again, this is despite the fact that my grammar is flawless. Again, I think that the brevity of my answer means that I didn’t have the opportunity to use any advanced grammatical structures. I have a couple of subordinating conjunctions, but that’s about it. I don’t have any coordinating conjunctions, I don’t have any adverbs and I’m short on adjectives. There are no conditionals in the answer, either. Most of the answer is in the past tense. Some people might be able to fit a lot of grammatical conventions into just 75 words, but it isn’t easy. I think my limited use of grammar is common in answers that are delivered slowly.
What did SpeechRater think of my Coherence?
SpeechRater didn’t like my coherence either:
Again, my impression is that my answer was too short to include enough connective devices to please SpeechRater. There are three obvious transitional phrases in my answer… but an answer with more words overall would naturally have more than that. Likewise, it would probably have a few compound sentences (my answer has none).
The point I’m trying to make here is an obvious one, but it is important. A TOEFL response delivered slowly may draw a low score from the SpeechRater. In addition to being short overall, it will likely be missing some of the key features the SpeechRater wants to see. Be careful on test day.
As I indicated above, a future experiment will be to create the best possible 74 word answer, to see the best-case result for a slow answer.
Part of your TOEFL speaking score comes from the SpeechRater engine, which is an AI application that scores your responses during the speaking section of the TOEFL. Basically, every one of your answers is graded by one human scorer, and by the SpeechRater engine. These scores are combined to produce your final score for each response. We don’t know how the human rater and the SpeechRater are weighted. I assume that the human rater is given greater weight, but I don’t have any evidence to support that claim.
How does the SpeechRater engine work? It is hard to answer this question with any certainty, since ETS doesn’t provide all of the details we want to read. However, an article published recently in Assessment in Education provides some helpful information.
The article describes the twelve features used to score the delivery of a TOEFL response, and the six features used to score the language use of a TOEFL response in one study. It also describes the relative impact of each feature on the final score.
It is really important to note that the article only describes how the SpeechRater engine was used in a specific study. Remember: when the SpeechRater engine is used to grade real TOEFL tests the feature set and impact of each feature might be different from this study.
So. Let’s dig into those features and their relative impact. First, the 12 delivery features:
stretimemean (15% impact). This feature measures the average distance between stressed syllables. Researchers believe that people with fewer stressed syllables overall are less expressive in using stress to mark important information (source). SpeechRater measures this variable in time between stressed syllables, rather than in syllables themselves. I would like to experiment with this using implementations of the SpeechRater (Edagree, My Speaking Score) but I find it difficult to eliminate stresses from my own speech.
wpsecutt (15% impact). This is your speaking rate in words per second. If you say more words, you get a better score. This has been confirmed by my experiments with the above implementations.
wdpchk (13% impact). This is the average length of uninterrupted speech (chunks) in words. A chunk is a word or group of words that is separated by pauses (source). Note that other implementations of SpeechRater have measured chunks in seconds rather than words (source).
wdpchkmeandev (13% impact). This is the “mean absolute deviation of chunk length in words.” The absolute deviation is important because obviously the average mentioned above can be skewed by the presence of one really long chunk and a bunch of short chunks. This feature seems to reward people who give answers containing chunks of sensible lengths.
conftimeavg (12% impact). This one is described as “mean automated speech recogniser confidence score; confidence score is a fit statistic to a NNS reference pronunciation model.” I don’t know what that means. But the article says that it relates to your pronunciation of segmentals, so I suppose it measures how well you pronounce vowel and consonant sounds.
repfreq (8% impact). This measures the repetition of one or more words in sequence. As in: “I like like ice cream.” I have experimented with this a bit, and was able to reduce my score by about two points (out of 30) by inserting a bunch of such repetitions.
silpwd (6% impact). This measures the number of silences in your answer of more than 0.15 seconds. Pauses hurt scores! Note that I’ve also seen this referred to as measuring pauses of greater than 0.20 seconds, but don’t ask me for a citation.
ipc (6% impact). This is said to measure the “number of interruption points (IP) per clause, where a repetition or repair is initiated.” I’m not quite sure what that means. Obviously, though, it has something to do with moments when the speaker backtracks to correct an error in grammar or usage (like: “Yesterday, I go… went to school.”)
stresyllmdev (5% impact). This is “the mean deviation of distances between stressed syllables.” Again, it encourages the speaker to have sensible distances between stressed syllables, rather than merely having a nice average. I think. I’m not much of a mathematician.
L6 (3% impact). This is described as “normalised acoustic Model (AM) score, where pronunciation is compared to a NS reference model.” I am not sure how this differs from “conftimeavg” above. Again, though, it relates to your pronunciation of segmentals. Teachers and students need to know that proper pronunciation is probably a good thing.
longpfreq (3% impact). This measures the number of silences greater than 0.5 seconds. It is interesting that the SpeechRater engine has a separate category for really long pauses. Some implementations seem to combine these into a single reported result, while others provide two separate pause-related results. This certainly warrants some experimentation.
dpsec (1% impact). This measures all of the “umm” and “eer” disfluencies. Interestingly, these don’t seem to matter at all! I suppose, though, there is a risk that disfluencies can impact the pause and chunk related features. I will experiment.
Next, the 6 language use features:
types (35% impact). This measures “the number of word types used in the response.” There is no definition for “word types” but we can assume it refers to: adjectives, adverbs, conjunctions, determiners, nouns, prepositions, pronouns and verbs. I guess the SpeechRater engine rewards answers that include all of those. Most of those will be used naturally in an answer, but it is easy for students to forget about adjectives and adverbs. And, obviously, lower-level students will not be able to use conjunctions properly. I don’t really know if SpeechRater is looking for a certain distribution of types.
poscvamax (18% impact). Oh, dammit, this is another hard one. It is described as “comparison of part-of-speech bigrams in the response with responses receiving the maximum score.” It is touted as measuring the accuracy and complexity of the grammar in an answer. A bigram is a sequence of two units (source). I would assume, in this case, that it is two adjacent words. Perhaps SpeechRater purports to measure grammar by comparing how you paired words together to how other high scoring answers paired words together. Yes… you are being compared to other people who answered TOEFL questions. In my experience, SpeechRater’s grammar results have been wonky and some implementations don’t bother showing them to students. I think EdAgree removed this from their results recently.
logfreq (15% impact). This measures how frequently the words in your answer appear in a reference corpus (the corpus is not named). It purports to measure the sophistication of the vocabulary in the response. I guess this means that the use of uncommon words is rewarded… but surely there is a limit to this. I don’t think one can get a fantastic score by using extremely uncommon words (as they would sound awkward).
lmscore (11% impact). This “compares the response to a reference model of expected word sequences.” I’m not sure what this means, but it seems like you will be rewarded for stuff like proper subject-verb agreement. One imagines that “Most cats like cheese” is a more expected sequence than “Most cats likes cheese.” Teachers and students should probably just assume that proper grammar is rewarded, and improper grammar is penalized.
tpsec (11% impact). This measures the “number of word types per second.” Again, we don’t have an official definition of “word types” but my assumption is that students are rewarded for using a greater variety of word types in the answer. That is to say, the SpeechRater may not be looking for a specific distribution, but rewards a simple variety of types.
cvamax (10% impact). This compares the number of words in the given answer with the number of words in other answers that got the best possible score. Popular wisdom seems to be that the best scoring answers are 130 words in the independent task and 170 words in the integrated tasks.
I think I will leave it at that, but please consider this post a work in progress. I’ll add to it as I continue to carry out research.