Language in the fast lane — the end of an epoch?

Throughout history, humans have tried to accelerate their language communication skills in various ways. Some standard structural aspects of language behavior (using lips and larynx, simplifying speech to children, etc.) don’t seem to change much, but others have been selected for technical fixes to accelerate speaking/writing/reading/signing/receiving linguistic content (words and sentences).

Now we have pivoted into a digital age of machines that can process speech and text content at electronic velocities, the pursuit of human speed around language has lost most of its relevance. The combination of machine learning and large data bases is marking a major flexion point in the history of human language. Until now, trying to execute linguistic acts more quickly was a deliberate field of innovation. Let’s see how.

Speed-writing and speed-reading, for example, became a minor fashion fairly recently (post-war 1950s) in the global north. Anxious people were encouraged to believe you could increase your intake of information in education or at work by learning to read books or documents quicker or type more words per minute. Speed meant more work achieved in less time, and eventually more income. It was part of a broader managerial time-and-motion fad as the economy got going again after World War II and office bureaucracies expanded.

These time-conscious practices are less front-of-mind today. We use compute systems and networks that determine most of the basic velocities in our information exchanges. Yet speed of execution and communication is still an implicit delivery parameter in business, science, diplomacy, sport, military action, and more. Does this mean that specifically human linguistic abilities could reach a natural velocity barrier, whereas new AI-driven work-arounds will either collect or create and deliver far more information at scale more quickly?

We will certainly interact verbally with robots of some kind in the near future. The logical choice is that these robots are trained to converse with us at our ”natural” human comprehension speed. But should we make this speed adjustable on such a machine, so it adapts to the natural speed of each interlocutor? Or should the bots maintain a steady drone, rather like the robot in Space Odyssey, to avoid uncanny valley phenomena?

These and other questions point to the varieties of linguistic road-racing over the past couple of millennia.

Speaking at speed

Obviously, people speak at different natural speeds (within a fairly narrow range) and there is evidence that certain communities/language groups/individuals are particularly fast talkers who increase their average speed from the typical 145 to around 160 words per minute or more., Professional broadcasters and commentators can often deliver at up to 200–250 words per minute for a given period. Naturally this raises the question about the “length” or identity of a word, as this is usually a visual entity only available through some form of writing. But interestingly, different languages all tend to transmit information at about the same rate of 39 bits per second, whatever the form of their words — about twice the speed of sending Morse code!

We all know individuals who seem to speak at the speed limit of understanding. And no, you can’t speed up your hearing — i.e. absorb acoustic information faster than a certain standard value, determined by your brain processes rather than your will. But you can learn to speak more quickly if this is deemed socially or professionally beneficial, and speech trainers exist to help us with this.

Employers might vet speech speed behavior among job candidates to detect speed deficits and other potentially negative signals, but children in many societies are typically vetted and treated for serious speaking deficiencies such as apraxia, dysarthria, orofacial myofunctional disorders, and stuttering, which can pose serious communicational barriers.

Yet deciphering rapid speech is still hard. You can learn over time to adapt to a particular fast-talking individual in your entourage, but probably not extend this skill to just anyone you encounter. Think for example, of trying to learn from those speeded-up spoken medical product ads on the TV or radio, designed to fulfill legal requirements within a short time window. Rapid speaking is also used during horse- or motor-racing commentaries, as well as when nervous public speakers gabble their prepared speech.

This kind of fast-moving monologue can of course impede interpreting fluency for professional spoken translators or note-takers and more generally make content understanding difficult in any communication situation where meaning needs to be relayed to a third party in some way. In the case of recorded speech, we are now able to slow down the machine output speed to understand rapid speakers if necessary.

Overall, though, formalizing or controlling speaking speed has never been a major issue in discussions about global communication and education. Nor do we systematically prevent people from speaking fast, though reasonable speed shared universally by our brain mechanisms appears to be part of good communicational manners.

It will be interesting to observe whether we will be able to control speaking speeds in foreign languages, for example, when more spoken language is mediated globally across the upcoming metaverse.

Dictating flat out

Producing an accurate written record of normal-speed speech has traditionally been a hard task. Before audio recording technology, which can modify the tempo of the recording to make listening easier for a typist to copy out the content, the only way we mastered the written recording of speech was to use stenography — an abbreviated sign system (more below). Noting down content from another human has a long history, even though many early writing systems were not in themselves quick fixes for inscribing full semantic content.

Writing in ancient Sumer and Egypt was not primarily designed to take down connected speech but initially to note down information about crop volumes, accounting facts, royal decrees, and similar numeracy-related issues. Much later, after alphabets were developed (mainly post-1200 BCE), it became easier to note down spoken content; a phoneme-type notation made it possible to shorten the amount of actual physical work compared to non-alphabetic script: abbreviation is easier with an alphabet than with character-based systems in terms of learning time, as you can often delete the vowel letters “wtht hndring mst ndrstndng”.

In the domain of writing texts ab initio (not just copying other people’s speech), communication has often demanded speed of execution. Typically, handwriting speeds range widely from 20 to 100 words a minute, with an average of around 45 wpm, though legibility will vary enormously. Writing longhand in office usage eventually gave way to typing, borrowing ideas from early experiments to help blind writers in the late 18th century in Europe.

Typing radically increased the production of clean, readable output by the late 19th century. This compared favorably with generations of scribbled notes in pen or pencil which later had to be copied again and again for public circulation by professional scribes.

Again from the late-19th century, you could maximize message transfer speeds by using the electric telegraph (literally long-distance writing). This led to expressing brief sequences of words — “telegramese” — that relied on human understanding for decoding. The point was that the shorter the message the cheaper the cost, as each telegram had to be tapped out in coded signals. But it meant that linguistic information began to travel at an unprecedented rate by adapting linguistic form to boost velocity.

Prior to that, an invention such as Chappe’s telegraph in France in the 1790s using flag signaling between two field stations to convey messages over long distances (telescopes were used to identify the content from a sending station) depended on shortening messages to a fixed minimum and then either spelling out complex or hard-to-guess words using a flag code. This meant agreeing ahead of time on a number of rapid flag signals for highly repetitive words or special expressions.

Speed again was of the essence as this telegraph device was first designed for military or intelligence work before becoming briefly more multifunction. And then replaced by wires. Sending military and diplomatic secrets in special messages for transmission goes back of course to early antiquity, and (usually) worked against rapid decoding for the receiver. No doubt techniques were developed to speed up the encoding process, as well as the decoding process among your enemies.

Stenography

Literally “narrow writing” and better known as “shorthand”, this is an age-old technique designed to capture continuous language content spoken at normal speed rather than being dictated in short bursts. It goes back at least to Tiro’s shorthand system used to report speeches in the Roman senate around the 1st century BCE. But steno systems of various kinds were no doubt reinvented many a time by different scribes who needed to record crucial spoken content in many different geographies and languages.

Eventually formalized and taught, steno is currently used in some courtrooms and critical-testimony meetings to live-record speech using a specially designed typewriter that accelerates capture using a multi-finger “chording” technique. In terms of speed, this is real time “speech-writing” but is now gradually being replaced by automated solutions.

Speech-to-text technology can either be used to subtitle speech on a video stream or generate a written transcript as a text file. So far this tech only covers a few dozen languages around the world, is not always as fast as it should be on the screen, and not yet 100% reliable as a source of mission-critical testimony. Yet it is clearly a way forward for capturing cross-media content expression on a massive scale. Shorthand will gradually disappear.

Standard writing speeds (i.e. not just for copying speech) have obviously increased radically over human history due to improvements in the technology of both instrument and surface media. Chiseling stone tablets in ancient Egypt took longer than scribal writing in the Chinese imperial court using brush and ink on long rolls, which in turn took longer than pen nibs on scrubbed parchment sheets in medieval Europe.

Eventually moveable-type printing on paper took over part of the text copying process and radically speeded up widespread access to standardized written knowledge for the literate. When the typewriter, first patented in 1714, became widely used as a copying and distribution technology in the offices of alphabetic (European) and later non-alphabetic (Chinese) cultures during the late 19th century, a tradition even developed in the profession to hold typing speed competitions for alphabetic “copy” typists — not actual writer-producers. The first was held in Cincinnati in 1888, and these continued until the early 21st century.

The initial alphabetic English language keyboard layout was — ironically — later redesigned in order to slow down the tapping fingers so that adjacent keys did not jam. Hence the QWERTY (and similar for other languages) layout that has long been mainstream in much of the global North. Interestingly, the invention in 1936 of the Dvorak keyboard — a major layout reboot to help professionals type faster and better — failed to catch on, even though it is still available for computer keyboards.

In fact, writing words down — a major activity across vast swathes of personal, social, educational, and professional life since the Renaissance in Europe and now globally — has constantly been the target of innovations to increase speed, but also legibility, understandability, and later searchability.

Pen and pencil design has evolved to ensure constant ink or graphite flows, and in the case of Lewis Carroll in the 19th century, inventions included a method for writing down notes in the dark at night! Yet human writing output might now be declining in absolute quantities as voice and compute gradually power our communication networks and content production.

Speed readers

Accelerating your reading speed has been a significant target in knowledge-based societies, due to the historically recent extension of literacy and education to children in most societies. Dyslexia is the corresponding disability to that skill, causing difficulty in reading and writing due to misidentifying letter signs. Many individuals try to improve their reading — and therefore learning — speed so that they can gain more knowledge from content more quickly and compete at skills or deliver services better.

There are a number of special techniques available for speed reading, but these too are probably mostly used in the global North, if at all these days. Most people learn themselves how to read faster, mainly by skimming for relevant or interesting information words. The emergence starting in the 1990s of rapid electronic delivery of chunks of written content for everyone (emails, text messages, social media, etc.) has meant that information is arriving ever faster straight to the consumer. Reading speeds have adapted accordingly, but the nature of the written message has also helped this.

For example, there is a range of stylistic techniques exploited by writers — i.e. information producers — to render written content easier and quicker to absorb for any reader. These mainly include simplifying sentence structure and using a more generalized or explanatory vocabulary to prevent semantic hiccups. Avoiding ambiguous phrases also helps, as does the use of bullet point layouts, logically signaled narratives, numbered lists, and similar visual cues to accelerate content absorption in practical chunks.

In other words, improved layout, visual aids such as variable font sizes, and other interface design practices have probably done most to accelerate content reading and retention for most people in the digital age.

“Controlled” or simplified language

One of the more interesting research efforts to speed up “technical” readability, but which has failed to scale, is what became known in the last decades of the 20th century as controlled language. This approach to writing design for better readability involved restricting the rules of grammatical patterns, vocabulary, and sentence length to a small set of active descriptions or instructions, rather as we have typically used in children’s literature.

The invention by C.K. Ogden in the 1930s of Basic English (a set of 850 vital words for commercial communication, plus restricted grammatical structures) was aimed at non-native users. In fact, the British government purchased the copyright to Basic English in WW2 and it was used (by George Orwell!) to author part of BBC broadcasts to India during WW2 in the 1940s.

More systematic efforts for “controlled language” began in the 1970s to draft large industrial machine documentation (e.g. building tractor plants, airplanes, or oil refineries). The aim was that everything should be read and understood unambiguously quickly and effectively by non-native technicians faced with English-text plans and workbooks for mission-critical constructions. There was also a French system.

Today controlled discourse is being applied (if at all) to communicating with machines — e.g. computer-related knowledge processing uses, for example by providing a special systematic vocabulary to describe object words in knowledge graphs when processing meaning in large text data collections.

Similar post-WW2 experiments for the general public, such as using “plain language” also tried to make language content (such as legalese, technical government information, content for immigrant targets, etc.) easier and therefore quicker to understand for either non-natives or general readers.

Many of these “inclusive” solutions have partially found their way into general information design today, as in government efforts to inform citizens during the Covid pandemic. Internet websites have also experimented with new layouts to display sets of instructions or newsletters or business communications about seemingly complex processes that can be read and digested quickly. Some newsletters even give the expected reading time for each of their content chunks.

Visual design for faster reading

Page and print design to simplify reading of course reach back far into history, starting with the crucial step of word separation on the page, which in the West began in the late 7th century. Until then, we must remember that words in Roman or Hebrew texts were typically separated by a dot, making it much harder to speed read or scan.

Note that in this crucial case of word presentation in Europe, it was Irish monks who developed the separated word approach to religious text copying, and by the 11th century this practice had become standard among northern Europe scribes in general. These were therefore non-Latin speaking communities that introduced this practical innovation to a Latin language display process that has been a feature of all written text in Europe ever since. A further question, therefore, is whether this word spacing trick actually helped speed up understanding or memorizing.

Many readers in earlier centuries in fact read their texts by speaking sotto voce, which would have inevitably reduced the speed at which they gathered and retained knowledge. Modern-style fully silent reading is achieved by eliminating this process of subvocalization. Using purely visual perception of signs without a murmuring vocal channel enables a reader to rapidly skim over less-informative words and zap onto the semantically juicy bits or search for specific key words along a line of print.

However, the reading of novels — longish story texts usually without illustrations and organized as a sequence of chapters that began to emerge as a popular literary media in the late 18th century in Europe — continues to exploit the traditional “paragraph of sentences” approach. Spoken exchanges by characters are signaled by quote marks or similar depending on national print conventions. Reading speed was not really an issue here: a literate population would plunge into a silent reverie as they turned the pages of their latest novel, or reread it with pleasure.

Silent reading rates typically range from some 140 to 170 wpm for most adult readers, and up to 290 and over for the well-practiced user. If you read a novel out loud to your siblings or an aged parent, you might drop to 150 wpm. And as generations grew more proficient, these rates would rise.

However, few of us would be able to match a certain Anne Jones who in 2005 read a short novel in UK English of 26,152 words (around 100 pages) in 17 minutes — a reading rate of over 1,500 words a minute. However, she only “comprehended” — i.e. remembered the data to answer questions about — 56.7 percent of what she read, which meant adjusting her speed to a still rapid 869.2 words per minute. For her, though, that was plain slow: two years later, she read Harry Potter and the Deathly Hallows silently in public at a bookshop in London in a record-breaking 47 minutes and one second, or 4,251 words per minute. She then wrote a review for a newspaper! But with only 50% of the content retained, where is the real pleasure in speed-reading fiction?

Today, reading speed can be partially monitored by eye-tracking technology, which notes where readers slow down or whizz forward, and deduces that either they find something interesting, or spot a linguistic problem with a word or phrase. This kind of intelligence gathering can be used for example in translation production, enabling the machine to spot hesitations in an editor’s “gaze”, which might suggest an error or awkward phrasing.

Language and sensory disability

Signing for deaf people is yet another media that engages with communication and reception speeds. The consensus opinion is that the communication of linguistic meaning using signing operates faster than in typical speaking situations, partly because visual signs can be combined to create complex meanings simultaneously or carry richer information that would take more time if uttered in linear speech. But it all depends on who is doing it for whom. A good sign language interpreter can transcribe spoken English at rates of 225 words per minute or higher.

Pre-lingually deaf people (deaf from birth) are almost always quicker at interpreting these signs than those who become deaf later in life. By definition, a visual sign for a word/meaning is instantaneous, whereas a spoken form takes time to utter in the acoustic sphere (faster or slower). So in signing, arms and hands act as articulators which can usually move faster than a sequence of sound units.

The interesting question is whether in future signing for all deaf speakers can be effectively automated, using a camera and software that can read spoken discourse and instantaneously transform it into a sequence of visual words over a phone or special spectacles. Will signing survive this technology shift over the longer term?

Braille users (the blind) on the other hand tend to read at approximately a third to a half the speed of a seeing reader of print. But there seems to be some divergence of opinion about these figures. Averages of 70–100 wpm are often contrasted with 200–300 wpm among sighted readers. Braille reading speed is a controversial topic and the data vary.

Towards an age of neuron speed

In the end, it looks as if the semantic core of a linguistic form takes about the same time to express, whatever the media. Speed as a dimension of human language production and reception simply ties in with the constant concern in human affairs to save time, boost efficiency, and gain in productivity, while also addressing human frailties and limitations. The critical feature facing us now is this: in an AI-fueled content future, where will the critical synapses occur for acceleration or compensation in language production, reception, and comprehension?

One such seismic shift in extreme communication could be intra-brain language manipulation, still in its infancy and highly science-fictional. Experiments have been carried out to link brain stimulation directly to language processing. The idea would be to enable brain- or otherwise damaged individuals to send messages straight from their brain to a display device, bypassing such organs of eloquence as mouths, tongues, and larynxes, or fingers, keyboards, and pens.

The further ambition is that we will somehow be able to exchange “speech” (or linguistic content) directly — and hence super-fast — from one brain via an artifical network to another brain. In a sense this would be the end of articulated, expressive, linear language as we know it.

This potential technology also raises the critical issue of the “deprivatization” of thought. Brain-to-brain could mean there would be no personal ego left to fashion the rhetoric of what we wish to say to someone: our raw “thoughts” would in a sense be uncontrollable. And would there be different languages to negotiate?

Future brain scenarios raise the prospect of sharing at speed a conceptual idiom, abolishing the need for uttering a linear sequence of sounds and intonations. Yet there would surely have to be some vehicle for negotiating our relations between two brains — the neural equivalent of a conversation. It is also possible that the brain doesn’t just represent existing information in some form, but actively constructs it in a constant flow. What kind of world would it be if we are all swirling our thought lines through virtual space instantaneously?

This should have taken about 23 minutes to read…!!