Sunday May 15, 2005

"Gulf" by Robert Heinlein

[This post is part of an ongoing series about linguistics in science fiction.  Fair warning: spoilers follow.  Page references are to the tenth printing of the Signet paperback of Assignment in Eternity.]

A couple of weeks ago, I was attending a meeting of the Machine Translation reading group.  During a discussion of Shannon's work on information theory and how it applied to the problem of coding information in speech, the professor who runs the group said, "That reminds me of an old science fiction story."  My ears pricked up.  "The one with Newspeak."

"Newspeak was in 1984," I said, not seeing the connection between Orwell's novel and information encoding.

"No, not 1984.  It was a story about a language that you could speak really fast because it used one phoneme per word..."

"Ah, that's in Robert Heinlein's 'Gulf'," I said.  I hadn't thought about the story in years, but I made a mental note to reread it and write a post about it.

"Gulf", originally published in 1949, is a story about espionage, weapons of mass destruction, supermen, and, last but certainly not least, language.  (It also has a sequel, Friday, that Heinlein wrote many years later.)  The main character, Joe, is an agent for the future Federal Bureau of Security whose last name might be Gilead, Briggs, or Greene.  The story begins with him traveling undercover, transporting microfilm from the Moon to Earth.  He realizes he's being followed, but manages to hand-draw the electronic routing labels and drop the microfilm in the mail before he's captured.  His captors place him in a cell with a man he comes to know as "Kettle Belly Baldwin", who communicates with him in an improvised code using a pack of cards.  Joe is suspicious at first, but comes to believe that Baldwin represents neither his captors nor him employers, but rather a mysterious third organization.  The two men escape together.

When Joe reports to his headquarters, he finds that the microfilm spools have not arrived at their expected destination.  The films contain the plans for the nova bomb, a new weapon capable of destroying an entire planet, and it becomes clear that Joe's boss doesn't believe his story and intends to pin the blame for their loss on him.  Joe, ever a man of action, punches him in the throat, drugs him, and slips out of the building.  Now a free agent, he contacts Baldwin, reasoning that he must have intercepted the microfilm and hoping that whatever organization he works for will be able to help him.

They meet at Baldwin's remote ranch, which is abuzz with activity.  Gilead-Briggs-Greene is unable to determine what Baldwin's people are working on, and occasionally he overhears bits of conversation in a language he doesn't understand and can't place:

There were two voices, one male, one female, outside the door; the female was Thalia Wagner, the man he could not place.

Male:  "tsʉmaeq?"

Female:  "nø!"

Male:  "zUlntsɨ."

Female:  "ɨpbitʹ New Jersey."

Those are not precisely the sounds that Gilead heard, first because of the limitations of phonetic symbols, and second because his ears were not used to the sounds.  Hearing is a function of the brain, not of the ear; his brain, sophisticated as it was, nevertheless insisted on forcing the sounds that reached his ears into familiar pockets rather than stop to create new ones.  (pp. 36-37)

Heinlein's explanation of what humans do with unfamiliar speech sounds is exactly right; that's a nice, clear little description of what linguists call categorical perception.  Joe is only able to recognize the units of sound in the strange language, which exist in a continuous acoustic space, by assigning them into discrete categories he's familiar with.  Those categories don't fit the new language, though, as Joe finds out when he asks a young woman about what he's heard:

"Gail, what does 'tsʉmaeq?' mean?"

"Come again?" she answered.  "I had water in my ear."

He repeated all of the conversation he had heard.  She looked incredulous, then laughed.  "You didn't hear that, Joe, you just didn't."  She added, "You got the 'New Jersey' part right."

"But I did."

"Say it again."

He did so, more carefully, and giving a fair imitation of the speakers' accents.

Gail chortled, "I got the gist of it that time."  (pp. 37-38)

Joe heads off to an interview with Baldwin.  Before Baldwin has time to offer a full explanation, he receives a message through the intercom on his desk ("œnɪe ʀ ħøg rylp", to which he replies, "nU"), and informs Joe that the ranch is being raided.  They head into an underground hiding place, where Baldwin reveals that he did intercept the nova bomb microfilms, and then gains Joe's trust by destroying them.  Asking Joe to join him, Baldwin reveals who he is and what his organization is about:

"You asked what I was.  I'm sort of the executive secretary of this branch of an organization of supermen."

"I thought so."

Joe is a smart guy, you see, which is why Baldwin is trying to recruit him.  The organization is made up of what Baldwin considers an emerging new species within the human race, homo novis.  They're not different physically, but rather mentally—as he puts it, "supermen are superthinkers" (p. 45)—and he suspects Joe is one of them.  To find out, Joe will be subjected to the novel training methods the organization has developed.  Gail is assigned as his teacher:

"First we must teach you to see and hear, then to remember, then to speak, and then to think."

Joe looked at her.  "What's this I'm doing with my mouth at this moment?"

"It's not talking, it's a sort of grunting.  Furthermore English is not structurally suited to thinking.  Shut up and listen."  (p. 51)

Gail begins to train him on a device that flashes numbers he must memorize up on a screen for a fraction of a second, which she calls it a Renshaw tachistoscope.  (It's apparently based on the work of a real scientist, Samuel Renshaw, about whom more here—but don't follow the link yet, since it also talks about "Gulf").  After Joe has been "rather thoroughly Renshawed" (which sounds uncomfortable), Gail begins to teach him the organization's language using a machine that displays the components of a sound on a screen (think spectrogram, although Heinlein doesn't use that word).  His first attempts aren't very promising:

"How's that, teacher?" he said triumphantly.

"Terrible, by several decimal places.  You held the final gutteral too long—"  She pointed.  "—the middle vowel was formed with your tongue too high and you pitched it too low and you failed to let the pitch rise.  And six other things.  You couldn't possibly have been understood.  I heard what you said, but it was gibberish.  Try again.  And don't call me 'teacher'."

...

"Confound it, if you would tell me what the words mean instead of treating me the way Milton treated his daughters about Latin, I could remember them easier."

She shrugged.  "I can't, Joe.  You must learn to hear and speak first.  Speedtalk is a flexible language; the same word is not likely to recur.  This practice word means: 'The far horizons draw no nearer.'  That's not much help, it it?"

The definition seemed improbable, but he was learning not to doubt her.

...

Speedtalk was a structurally different speech from any the race had ever used.  Long before, Ogden and Richards had shown that eight hundred and fifty words were sufficient vocabulary to express anything that could be expressed by "normal" human vocabularies, with the aid of a handful of special words—a hundred odd—for each special field, such as horse racing or ballistics.  About the same time phoneticians had analyzed all human tongues into about a hundred-odd sounds, represented by the letters of a general phonetic alphabet.

On these two propositions Speedtalk was based.

To be sure, the phonetic alphabet was much less in number than the words in Basic English.  But the letters representing sound in the phonetic alphabet were each capable of variation several different ways—length, stress, pitch, rising, falling.  The more trained an ear was the larger the number of possible variations; there was a limit to variations, but, without much refinement of accepted phonetic practice, it was possible to establish a one-to-one relationship with Basic English so that one phonetic symbol was equivalent to an entire word in a "normal" language, one Speedtalk word was equal to an entire sentence.  The language consequently was learned by letter units rather than by word units—but each word was spoken and listened to as a single structured gestalt.

But Speedtalk was not "shorthand" Basic English.  "Normal" languages, having their roots in days of superstition and ignorance, have in them inherently and unescapably wrong structures of mistaken ideas about the universe.  One can think logically in English only by extreme effort, so bad it is as a mental tool.  For example, the verb "to be" in English has twenty-one distinct meanings, every single one of which is false to fact.  (pp. 52-53)

Speedtalk is therefore both more efficient in terms of time, because the information content of an entire sentence in a natural langauge can be expressed in a single word, and more logical than natural languages.  Both of these features of Speedtalk have consequences.  The New Men can not only communicate faster, but because (Heinlein claims) all thought must take place in the symbols provided by langauge, they can also think faster and better than ordinary humans because Speedtalk is (handwavingly) more logical, more rational, and free of the built-in errors of natural languages.  At one point (p. 56) Heinlein says that since a homo novis thinks about three times faster than you or I, and since Speedtalk allows another sevenfold speedup of thought, the New Men live mental lives fully twenty-one longer than ours.  I'll bet they get really bored on the long flight to Tokyo, though.

Speedtalk sounds pretty cool, huh?  I don't believe it for a minute, of course.  Speedtalk may be more compressed in principle, but phonemes are phonetically distinct for a reason—speakers need to be able to pronounce them distinctly even if they've got food in their mouths, and hearers need to be able to tell them apart even it's raining loudly.  That business about all thought taking place in symbols provided by language strikes me as plain wrong, too—when I remember what the Mona Lisa looks like, what language am I thinking in?

Heinlein's conception of Speedtalk was heavily influenced by a theory called General Semantics that was developed by Alfred Korzybski in the early 20th century.  Heinlein goes on at some length about Korzybski's ideas, which were first laid out in Science and Sanity: And Introduction to Non-Aristotelian Systems and General Semantics (1933).  At the mention of "non-Aristotelian" sharp SF fans will be thinking of A. E. Van Vogt's The World of Null-A, another story built around Korzybski's theories.  I actually have a copy of Science and Sanity that I picked up years ago at a thrift shop when its title rang a bell, and I've been using it as bedtime reading for the last few nights.  It's more effective than sleeping pills, which is not to say it's completely uninteresting—it's becoming clear to me as I read that Korzybski was a first-class kook, and I love kooks.  They're so entertainingly detail-oriented.  After this, I may finally get around to reading Julian Jaynes' magnificently-titled The Origin of Consciousness in the Breakdown of the Bicameral Mind, which I understand is both kooky and also the basis of Neal Stephenson's Snow Crash.

"Gulf" is suprisingly dense with linguistics and philosophy considering that it's basically an action-adventure story.  Heinlein crams a lot of information into the rather talky middle part of the story—phonetics and phonology, Basic English, General Semantics, and Samuel Renshaw—but it's a mixed bag of stuff I can believe and stuff that's completely improbable.  Take, for instance, the Speedtalk automatic dictaphone that Heinlein says the New Men have (p. 59).  If you thought speech recognition was hard on natural languages, imagine how much harder it would be on Speedtalk, which contains many very slight variations on every possible speech sound.  So, while I enjoyed the story for its exciting plot and free-wheeling speculation, I have to say that its linguistic content is largely unconvincing.  It's also full of political and social speculation that I find at best suspect and at worst plain wrong.  Heinlein's secret organization of supermen is ruthless and extremely effective, but they conveniently have a sentimental attachment to democracy, human dignity, and freedom—so I guess that makes their willingness to assassinate their enemies OK, then?  Still, its worth reading "Gulf" if you never have—Heinlein's stories rarely fail to entertain.

I am The Tensor, and I approve this post.
07:23 AM in Linguistics in SF | Submit: | Links:

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341c88ad53ef00d8345fb2d269e2

Listed below are links to weblogs that reference "Gulf" by Robert Heinlein:

» Paging James Burke from Strangely Silent: De Doc`s Ventures
Between link-surfing and newsfeeds, Ive come across some fascinating stuff today all the more so because of the connections between them. Im not James Burke, but Ill have a go at this, for the sheer fun of making the connect... [Read More]

Tracked on Apr 20, 2006 4:22:27 PM

Comments

Nicely done. Gonna tackle Babel-17 sometime?

Posted by: Dorothea Salo at May 15, 2005 10:57:12 AM

If you thought speech recognition was hard on natural languages, imagine how much harder it would be on Speedtalk, which contains many very slight variations on every possible speech sound.

Couldn't you make the opposite argument, that it would be easy for a machine to interpret since, as very slight variation in the sound of a word will change its meaning, speakers are forced to pronounce each phoneme consistently? Which is the bigger challenge in speech recognition - making fine distinctions between different sounds, or recognizing which different sounds represent the same phonemes? (Seriously, I don't know.)

At least it should be easy enough to recognize morpheme boundaries, if there's a 1-1 mapping between morphemes and phonemes.

Posted by: Tim May at May 15, 2005 1:52:55 PM

Interesting. If H. novis is that much more mentally adept than ordinary human beings, then perhaps the information-theoretic side of the story isn't so completely far-fetched; in the fictional context, I can imagine that the übermenschen can train themselves to distinguish sounds well enough to dispense with the redundancy that's built into natural languges. (The rational language/rational thought business is a depressing load of hooey, though.)

Speedtalk makes for an interesting contrast with Ptydepe, the artificial language in Vacláv Havel's The Memorandum. Ptydepe is designed to have quite a lot of redundancy; all words are required to be n% distinct from one another (I forget what n is--something like 60 or 80, I think) so that, if one is a poor but not disastrously poor typist, the language is effectively an error-correcting code. This makes for some very long words, but the inefficiency is mitigated by an inverse correlation between length and token frequency. (The longest word in Ptydepe is 319 characters long and means 'wombat'.)

Of course, Havel imagined Ptydepe as part of a bureaucratic dystopia, not as a language for Heinleinian post-humans....

Posted by: Q. Pheevr at May 15, 2005 5:50:10 PM

Gah! For Vacláv, read Václav, of course. (I am no H. novis, and sometimes have trouble with vowel length.)

Posted by: Q. Pheevr at May 15, 2005 5:52:41 PM

As soon as I got to the "twenty-one distinct meanings, every single one of which is false to fact" I knew Korzybski was in the air. Speedtalk is a typical Heinlein idea, clever, unlikely-to-impossible, and on some level repellent -- Heinlein loved repellent ideas.

Interesting that the phrase "nova bomb" did not occur in the 1949 Astounding story but in the 1953 book version -- see here.

And I'm a little surprised that Heinlein would make such an elementary error in Latin as novis for novus.

Posted by: language hat at May 16, 2005 10:43:56 AM

Fooking synchronicity. This post has nearly nothing to do with the topic at hand, but your link to the article about Renshaw above links forward to Panshin's longer essay on how Rite of Passage relates to the work of Heinlein. Of course, I hadn't read or thought about Rite of Passage in years until I took it off the shelf at a whim last week. Of course, all of science fiction is incestuous, but it's still amusing to be in the middle of rereading a novel and have more info about it plop into my lap.

Posted by: Scott Schulz at May 16, 2005 2:27:16 PM

I just checked, and it really is spelled homo novis in the story. I didn't even notice that it was bad Latin—I must be getting rusty. Interestingly, my Latin-English dictionary lists homo novus, 'self-made man (first man of a family to reach a curule office)'.

As for the speech recognition question, it's true that Heinlein posited the New Men had superior cognitive abilities and could therefore process language faster, but that wouldn't help you build an automatic dictation machine. Suppose that, as Heinlein implies, Speedtalk has several different versions of each speech sound distinguished by tone level. If a short utterance (one "word") happened to contain a sequence like high-mid-high-high, how can the machine know what the baseline tone is? Even assuming only three tone levels, that sequence could be H-M-H-H, or M-L-M-M, or H-L-H-H. A (super-)human listener can figure out from context that you meant "disarm the nova bomb immediately" and not "curious green ideas sleep furiously", but how is the machine to know?

I think I agree with the statement "Heinlein loved repellant ideas" in the sense that he loved bringing up unorthodox ideas and then poking at them from all sides (including from the inside). I still have trouble pinning down exactly what he believed, and what he's just having his characters assert to tweak us. Here's another example of the same technique from a recent Language Log post:

And given Chomsky's taste for intellectual provocation, I can imagine him flirting with such an alliance. In the first lecture I ever heard him give, in 1965, he asserted that psychology had in no way improved on Plato's theory that learning is remembering past lives. Hilary Putnam interrupted from the back of the room: "Wait a minute. You're not seriously suggesting that reincarnation is a plausible explanation?" Chomsky held his ground: "Why not? It certainly makes more sense than associative learning does."

(Hmm, now done it—I've compared Chomsky and Heinlein's debating styles. I'd better quit while I'm ahead...)

Posted by: The Tensor at May 16, 2005 7:46:54 PM

As for the speech recognition question, it's true that Heinlein posited the New Men had superior cognitive abilities and could therefore process language faster, but that wouldn't help you build an automatic dictation machine.

Who said it would? (Although in a general sense, superior cognitive abilities would presumably be useful to any kind of inventor.)

Suppose that, as Heinlein implies, Speedtalk has several different versions of each speech sound distinguished by tone level. If a short utterance (one "word") happened to contain a sequence like high-mid-high-high, how can the machine know what the baseline tone is? Even assuming only three tone levels, that sequence could be H-M-H-H, or M-L-M-M, or H-L-H-H. A (super-)human listener can figure out from context that you meant "disarm the nova bomb immediately" and not "curious green ideas sleep furiously", but how is the machine to know?

I'm at a disadvantage, not having read the story, and I'm in no way an expert on speech recognition. But this may not be such a problem if the machine can be calibrated to individual users - how much does the baseline tone vary in a given person's speech? How much of a problem is this in speech recognition of real-world tonal languages? Or perhaps Speedtalk just uses contour tones.

Posted by: Tim May at May 17, 2005 5:12:35 AM

It's the "furthermore, English is not structurally suited to thinking" sort of thing that always cracks me up about early Heinlein. It's the linguistic equivalent of the 1930s aeronautical engineer's claim that bumblebees can't fly.

I had to do a panel on Heinlein and linguistics at Norwescon this year, and had a lot of trouble trying to figure out how to say "Korzybski was a first-class kook" without offending the president of the Heinlein Society, two chairs down. As I recall, I got by with some handwaving about Derrida and Sapir-Whorf.

Posted by: David Moles at May 17, 2005 11:18:24 AM

Of course, all of science fiction is incestuous [...]

And none more so than Heinlein's! (I hope....)

Posted by: Q. Pheevr at May 17, 2005 12:23:21 PM

It's hard to beat "All You Zombies".

Posted by: David Moles at May 17, 2005 5:09:17 PM

But don't forget Sturgeon's "If All Men Are Brothers, Would You Let One Marry Your Sister?"

Posted by: language hat at May 17, 2005 5:45:47 PM

I think David Gerrold's The Man Who Folded Himself is in the running, too.

Posted by: The Tensor at May 17, 2005 5:54:06 PM

So basically, with some handwaving, Heinlein has his characters speaking mathematics.

Posted by: agm at May 22, 2005 2:25:21 AM

Or APL.

Posted by: David Moles at May 23, 2005 9:38:46 AM

Cute, I'm old enough to have used APL just a little.

Posted by: Stephen M (Ethesis) at Aug 12, 2005 7:52:42 PM

The longest word in Ptydepe is 319 characters long and means 'wombat'.

The problem with a language like that is that it strongly discourages specialization. Can you imagine being a biologist who specialized in the study of wombats?

Posted by: Chris Ball at Sep 29, 2005 8:49:55 AM

Think of what a powerful tool of social control such a language would be if you designed it such that all the words associated with changing the government were really long:

Can't...foment...revolution. Jaw muscles...too tired...

Posted by: The Tensor at Sep 29, 2005 2:13:44 PM

Hi,

My name is Janko.
I'm collecting numbers from various systems in different languages.
Could you please send me numbers from 1-10 in your conlangs (artlangs).
About my project you can find on:
www.ataltane.net/jg/
www.ataltane.net/jg/JANKO%20GORENC.htm
www.ataltane.net/jg/CONSTRUCTED%20LANGUAGES.htm
www.ataltane.net/jg/NUMBERS.htm
www.ataltane.net/jg/SOME%20INTERESTING%20PAGES.htm
www.ataltane.net/jg/CONLANGERS.htm

I wish you a lot of success at your work!

Thank you for your help!

Janko Gorenc.

Posted by: JANKO at Mar 8, 2006 9:48:30 AM

2 years later and its always humorous to see children who repetitively state that they know nothing of linguistics, biology, psychology, mathematics, evolution or even bothered to do any basic research giving plight to those who have:

From The Tensor
"but it's a mixed bag of stuff I can believe and stuff that's completely improbable. Take, for instance, the Speedtalk automatic dictaphone that Heinlein says the New Men have"

Response:
http://en.wikipedia.org/wiki/Speech_recognition
Performance of speech recognition systems

The performance of a speech recognition systems is usually specified in terms of accuracy and speed. Accuracy is measured with the word error rate, whereas speed is measured with the real time factor.

Most speech recognition users would tend to agree that dictation machines can achieve very high performance in controlled conditions. Part of the confusion mainly comes from the mixed usage of the terms "speech recognition" and "dictation".

Speaker-dependent dictation systems requiring a short period of training can capture continuous speech with a large vocabulary at normal pace with a very high accuracy. Most commercial companies claim that recognition software can achieve between 98% to 99% accuracy if operated under optimal conditions. These optimal conditions usually means the test subjects have:

* matching speaker characteristics with the training data,
* proper speaker adaptation, and
* clean environment (e.g. office space).

This explains why some users, especially those whose speech is heavily accented, might actually perceive the recognition rate to be much lower than expected. Speech recognition in video has become a popular search technology used by several video search companies.

Limited vocabulary systems, requiring no training, can recognize a small number of words (for instance, the ten digits) as spoken by most speakers. Such systems are popular for routing incoming phone calls to their destinations in large organizations.

Both acoustic modelling and language modelling are important studies in modern statistical speech recognition. In this entry, we will discuss the use of hidden Markov model (HMM) which is widely used in many systems. (Language modelling has many other applications such as smart keyboard and document classification; to the corresponding entries.)

The Carnegie Mellon University has made some good steps in increasing the speed of speech chips by using ASICs (application-specific integrated circuits) and reconfigurable chips called FPGAs (field programmable gate arrays).


From Scott Schulz
"I just checked, and it really is spelled homo novis in the story. I didn't even notice that it was bad Latin—I must be getting rusty. Interestingly, my Latin-English dictionary lists homo novus, 'self-made man (first man of a family to reach a curule office)'."

Response:
He probably was trying to create a new word or a new way of describing something new. But I suppose that is above you or beyond your ability to comprehend.


From The Tensor
"I'm at a disadvantage, not having read the story, and I'm in no way an expert on speech recognition."

Response:
And yet you will comment on this. I know nothing of surgery, but hey, lets give a thesis on it shall we...


From Tim May
"It's the "furthermore, English is not structurally suited to thinking" sort of thing that always cracks me up about early Heinlein."

Response:
English is considered one of the hardest languages to learn, its constant abstract thought processes and multi use words are what leave most new learners and even children at a loss. The "mis-understandings" that often occur between 2 people happen due to the inconsistencies within the language itself. As someone who speaks fluently 9 languages, I can assure you that english is not for the thinking man.

From Chris Ball
"Think of what a powerful tool of social control such a language would be if you designed it such that all the words associated with changing the government were really long Can't...foment...revolution. Jaw muscles...too tired..."

When one does not have a word for a thing, either they will make a new word or attempt to assemble other words together to make the conversant understood.

From Tensor
"when I remember what the Mona Lisa looks like, what language am I thinking in?"

Response:
You are thinking in picture, and when we speak, we send as vibrations to the listener vocal images, the word cat is meaningless if you haven't seen a cat before or have no point of reference. When we learned english we were presented with books that had the word and image for cat. And thus we learned that cat meant not only this http://en.wikipedia.org/wiki/Image:Bengal_Cat_%28Fia%29.jpg but also this http://dictionary.reference.com/browse/cat i.e. http://www.amazon.com/Cat-Hat-Dr-Seuss/dp/039480001X

A "critic" is a man who creates nothing and thereby feels qualified to judge the work of creative men. There is logic in this; he is unbiased – he hates all creative people equally. - Robert A. Heinlein

Posted by: Michael at Oct 16, 2007 4:56:48 PM

Oh and silly me I almost forgot..


http://home.inreach.com/sl2120/ilaksh/Ilaksh_Intro.html

The language. Do you have what it takes to learn it? Doubtful.

Posted by: Michael at Oct 16, 2007 5:01:32 PM