The Birth of the Semicolon.

Cecelia Watson, a historian and philosopher of science who teaches at Bard College, writes for the Paris Review about one of the many lasting products of the Renaissance:

The semicolon was born in Venice in 1494. It was meant to signify a pause of a length somewhere between that of the comma and that of the colon, and this heritage was reflected in its form, which combines half of each of those marks. It was born into a time period of writerly experimentation and invention, a time when there were no punctuation rules, and readers created and discarded novel punctuation marks regularly. Texts (both handwritten and printed) record the testing-out and tinkering-with of punctuation by the fifteenth-century literati known as the Italian humanists. The humanists put a premium on eloquence and excellence in writing, and they called for the study and retranscription of Greek and Roman classical texts as a way to effect a “cultural rebirth” after the gloomy Middle Ages. In the service of these two goals, humanists published new writing and revised, repunctuated, and reprinted classical texts.

One of these humanists, Aldus Manutius, was the matchmaker who paired up comma and colon to create the semicolon. Manutius was a printer and publisher, and the first literary Latin text he issued was De Aetna, by his contemporary Pietro Bembo. De Aetna was an essay, written in dialogue form, about climbing volcanic Mount Etna in Italy. On its pages lay a new hybrid mark, specially cut for this text by the Bolognese type designer Francesco Griffo: the semicolon (and Griffo dreamed up a nice plump version) is sprinkled here and there throughout the text, conspiring with colons, commas, and parentheses to aid readers. […]

Nearly as soon as the ink was dry on those first semicolons, they began to proliferate, and newly cut font families began to include them as a matter of course. The Bembo typeface’s tall semicolon was the original that appeared in De Aetna, with its comma-half tensely coiled, tail thorn-sharp beneath the perfect orb thrown high above it. The semicolon in Poliphilus, relaxed and fuzzy, looks casual in comparison, like a Keith Haring character taking a break from buzzing. Garamond’s semicolon is watchful, aggressive, and elegant, its lower half a cobra’s head arced back to strike. Jenson’s is a simple shooting star. We moderns have accumulated a host of characterful semicolons to choose from: Palatino’s is a thin flapper in a big hat slouched against the wall at a party. Gill Sans MT’s semicolon has perfect posture, while Didot’s puffs its chest out pridefully. (For the postmodernist writer Donald Barthelme, none of these punch-cut disguises could ever conceal the semicolon’s innate hideousness: to him it was “ugly, ugly as a tick on a dog’s belly.”)

There are images, including a side-by-side comparison of the various early varieties, and a discussion of “hand-wringing sages [who] forecast a literary apocalypse precipitated by too-casual attitudes about punctuation”; Watson warns against mistaking the –que abbreviation for a semicolon, something that is rarely a problem in these days of fallen Latinity. (Semicolons previously on LH: 2002, 2004, 2008, 2009, 2012 — inter, haud dubie, alia.)


  1. Can anyone explain when, how, and why modern Greek adapted the semicolon to indicate a question, instead of the question mark?

  2. The Greek and Latin question marks are roughly contemporaneous (the 8C). Old Church Slavonic used the Greek one; I don’t know if it was Peter the Great who switched to the Latin one for secular (“civil”) use, but it wouldn’t surprise me. The typographical convergence with the Latin semicolon probably began after1453 and was hastened when Greek printing (at that time almost entirely in the West) took off, as the earliest Greek typefaces were based on Italian humanist models.

  3. Russian Wikipedia has a lamentably short and sketchy article, but they do say the ? was first used for questions in the 18th century, so I’m guessing you’re right.

  4. Athel Cornish-Bowden says

    While we’re at it, when did Spanish start to enclose questions in ¿ … ? ? It has the advantage of telling you where the question begins, which isn’t necessarily the beginning of the sentence. I’ve sometimes been surprised that it hasn’t spread to other languages, not even Catalan and Portuguese.

    The other day my daughter gave me a pill container with a row of little boxes labelled LU MA ME JE VE SA DI. The Catalans would need to label them DI DI DI DI DI DI DI , which might be confusing.

  5. I could swear I’ve seen rhetorical ⸮ somewhere in modern times as well, but then perhaps as an independent reinvention.

  6. The typographical convergence with the Latin semicolon probably began after 1453

    Given that we’ve just heard that there weren’t any Latin semicolons to converge with until 1494, this seems an odd way of putting it.

  7. I encountered an unexpected appearance of the “¿” character on my old Apple IIGS computer. The IIGS was a sort of hybrid. It had an OS with a graphical desktop interface like a Macintosh. However, unlike the Mac, it was backwardly compatible with all the software and peripherals designed for earlier Apple II generations. In order to ensure this compatibility, there was a “control panel,” accessible at any time via a keystroke combination*, that let the user change the machine’s settings. This could be used to throttle down the speed for older games, or change settings that would, on older machines, have been set via hardware jumpers. One of the settings that could be changed this way was the keyboard layout. I played around with the alternative layouts, but they weren’t that interesting, except for the French-Canadian keyboard layout, which also changed the cursor prompt from “]” to “¿”; no other keyboard layout did anything similar.

    * For reasons that I never understood, interrupt request handling on Apple II and early Macintosh computers was much more robust that on PCs of the same era. It was almost trivially easy to lock up a PC so completely that it could not be restarted with anything short of cycling the power. That was not true of the Apples, which could be rebooted from the keyboard practically without fail. (Actually, I am not actually sure whether the other IRQs on Apples were as robust, since I never got into interrupt programming on Apple computers the way I did on PCs. My first commercial programming work was a program to manage conflicting TSRs, which required a lot of exploratory trial and error, since the vector swapping involved produced a lot of undefined behavior.)

  8. Stu Clayton says

    @Brett: For reasons that I never understood, interrupt request handling on Apple II and early Macintosh computers was much more robust that on PCs of the same era.

    I remember that crap with PCs. This seems to be relevant for Apples (it’s been a long time since I had to deal with such things):

    # The [6502] processor’s non-maskable interrupt (NMI) input is edge sensitive, which means that the interrupt is triggered by the falling edge of the signal rather than its level. The implication of this feature is that a wired-OR interrupt circuit is not readily supported. However, this also prevents nested NMI interrupts from occurring until the hardware makes the NMI input inactive again, often under control of the NMI interrupt handler.

    The simultaneous assertion of the NMI and IRQ (maskable) hardware interrupt lines causes IRQ to be ignored. However, if the IRQ line remains asserted after the servicing of the NMI, the processor will immediately respond to IRQ, as IRQ is level sensitive. Thus a sort of built-in interrupt priority was established in the 6502 design. #

  9. Stu Clayton says

    The Catalans would need to label them DI DI DI DI DI DI DI , which might be confusing.

    That’s cute, even irritating in a “I’m so ashamed for your predicament” way. I didn’t know about the Català words for weekdays.

    Every pastillero and organitzador pastillas advertised in Català that I find in the internet cops out with English, Spanish or German labels.

    One solution is obvious: leave off the initial “DI”, as in English we leave off the final “day”.

  10. PlasticPaddy says

    In Irish there is a similar problem but no standard abbreviations, so calendars use Lu or Lua for Dé Luain, Cé or Céa for Dé Céadaoin, depending on how much space they have (like English the two-letter abbreviations are enough to distinguish days when the initial Dé, when present, is dropped).

  11. Trond Engen says

    Two characters to distinguish seven items is bad design, but three characters to distinguish 12 months in the calendar is worse.

  12. When I was in secondary school, the calendars printed and distributed to show us what classes to go to when (necessary because what was a morning class on Monday might be an afternoon class on Friday) were headed M T W R F. Sometimes people would say “Happy R-day!”

    The reason for the rotation was to reduce the problem of classes first thing in the morning, right after lunch, or right before the end of the day always being the same subjects for a given student, since those are the hours in which students in general are least performant. Scheduling therefore had to be done for the whole school at once, probably on a service-bureau IBM mainframe.

  13. Ja F Mc Ap My Jn Jl Au S O N D: perfectly doable, but then doesn’t look very good.

    (Finnish edition: ta hl ml hu to k hn e s l m j, front-loaded and a poor match with the established capitalization conventions.)

  14. David Marjanović says

    Mo Di Mi Do Fr Sa So, fully established… though I’m not sure about places, like here, where people say Sonnabend instead of Samstag.

    (Outside the US, weeks begin on Monday.)

  15. Dé ¦ Luain = “On [the day] ¦ [of] Monday”

    On calendars and other noticeboard language you would just go with Luan = “Monday”

    The exception is Déardaoin, which is both “Thursday” and “On Thursday”.

    It also initial-clashes with Domhnach “Sunday”; I guess the minimal DOTW abbrevs would be L M C Da A S Do

    The preposition (only used with DOTWs) must be somewhat unusual in being always capitalised.

  16. T. Crofton Croker’s tale “A Legend of Knockgrafton”, in which the Good People were singing the days of the week, and from which we learn that if you improve their tunes they may reward you, but if you murder them they will surely punish you.

    (Published by Yeats in Fairy and Folk Tales of the Irish Peasantry (1888).)

  17. David Eddyshaw says


  18. David Eddyshaw says

    Akismet is angry.
    Trying again, with added padding:


    Padding added, with again trying.
    Angry is Akismet.

  19. Or “Sun Moon Fire Shui Mu Jin Tu”, as GT has it.

    Given inclusive counting, I don’t think that “eight days” is ever reliable evidence of an 8-day week, any more than “seven days” is evidence of a 6-day week.

  20. David Marjanović says

    Sun, moon, and the five elements: fire, water, wood, metal/gold, earth.

    Used in Japan, but not in China, where “week” is “startime”, Sunday is “startime sky” or “startime sun”, and the others are numbered 1–6.

  21. PlasticPaddy says

    I find the old form dia luain (also dia Luain, Dia Luain and Dia luain) in their earliest texts c. 1600. After 1750 Luain is generally capitalised and Dé appears instead of dia, except for Amhlaoibh Ó Súileabháin, the one-eyed Munster contrarian (is dia still correct in your dialect?).

  22. January First-of-May says

    Mo Di Mi Do Fr Sa So, fully established

    Russian tends to go Пн Вт Ср Чт Пт Сб Вс (sic – not the first two letters, though they would have been unambiguous).

  23. What I like about Dé [Dia] Luain for Monday is that it combines a transparent IE word for “day” and a transparent Italo-Celtic word for “moon,” both of which were obsolete as common nouns by the Classical Old Irish period.

  24. Of course homonymous with Dia/Dé “God” nominative/genitive.

  25. (Outside the US, weeks begin on Monday.)

    In Australia the week begins on Sunday. That is what calendars tend to show as well.

    The working week, on the other hand begins on a Monday.

  26. not the first two letters, though they would have been unambiguous

    Russian has strong aversion to ending abbreviations on a vowel.

  27. Lars Mathiesen says

    When I was wee, Danish calendars also started the week with Sunday, because after all God rested on the seventh day, which was Saturday. I blame various popes, the ISO and the Welsh Gnomes for upsetting the order of the world so grievously.

  28. The preposition Dé

    For what it’s worth, Wikt.en says it’s just a capitalized version of literary Irish ‘day-GEN’, so the gloss for Dé Luain is not ‘on [the day] | [of] Monday’ but ‘[on the] day | [of] Monday’, at least historically, and that dé luain is a direct borrowing with native inflections (as opposed to a calque) of dies lunae. If this is the only use of , it’s hard to say whether it’s a noun or a preposition synchronically.

  29. Giacomo Ponzetto says

    Athel, Stu: in Catalan it’s DL DT DC DJ DV DS DG, or more exactly: dl. dt. dc. dj. dv. ds. dg.

  30. Stu Clayton says

    Thanks, Giacomo. But how does DT arise from dimars ? The others make sense systematically.

  31. Giacomo Ponzetto says

    The spelling, at least the current correct spelling, is more Latin than that: dimarts.

  32. Hmmm, I had stopped at the first picture (like a “slide” in a slideshow) I saw under photos after a search for “catalan dias de la semana” – a slide showing days of the week. There “dimarts” is misspelled as “dimars”. All the *text* links I looked at just now spell it correctly.

  33. If this is the only use of Dé, it’s hard to say whether it’s a noun or a preposition synchronically.

    FWIW Ó Dónaill’s dictionaries (1977 and 1991) plump for noun.

  34. Kate Bunting says

    (Outside the US, weeks begin on Monday.)

    Not quite true
    (In Australia the week begins on Sunday. That is what calendars tend to show as well.)

    In the UK some calendars follow the tradition of beginning the week on Sunday, some the modern custom of thinking of the weekend as a unit.

  35. In the U.S., art calendars begin on Sunday and so do electronic ones (by default), but calendar books that show each hour separately generally begin on Monday and divide the sixth space between Saturday and Sunday.

  36. Incidentally, and returning to the original topic, I think it should be possible to construct a joke from the following elements: the discovery of a new world in 1492; the invention of a new punctuation mark two years later; Venetian-Genoese rivalry; “Cristóbal Colón”. Actually getting a laugh out of a human being using this material is left as an exercise for the reader.

  37. Trond Engen says

    How universal is the system of week numbers? Norwegian calendars have the weeks of the year numbered from 1 (the first week of 4 days or more) to 52 or 53 (the last week of four days or more). When planning ahead we can often say things like “Friday in Week 48” rather than dates.

  38. I know nothing of week numbers.

  39. I have occasionally seen week numbers printed on English-language day planners, but even there they are uncommon, and I have never known anyone to actually use them

  40. Trond Engen says

    I should have written “How widespread …”. I know it’s not known in the U.S., but the U.S. is hardly a reliable indicator of international standards. We use it mostly for planning purposes. Holiday plans. The schools’ winter break is in week 8 or 9 (depending on county). A meeting schedule may be planned for odd- or even-numbered weeks. Etc.

    I cane to think of week numbers because I read in Sunde and Mundal (the other thread) that the Sami had named weeks, and they speculated that the Norse may have had as well. I don’t mean to suggest a continuous line from one week system to the other, but it does show that using weeks rather than months for recurring annual events is a natural idea.

  41. David Eddyshaw says

    Never heard of it in the UK.

    Surely months are much more natural than arbitrary things like weeks, on account of that pretty shiny thing we see changing shape in the night sky (if the cloud cover ever breaks)?

    The Kusaasi have months; offhand, I can’t think of a culture which doesn’t, though I’m sure there must be some somewhere. (I expect the Pirahã don’t …)

    Now that I think of it, the Kusaasi don’t have named months (AFAIK); Mooré seems to, but the times just correspond to the European ones even though the names have no connection, and I suspect the whole system is just a sort of calque. “October” is “Millet-harvesting Month”, for example, and so on throughout.

  42. David Eddyshaw says

    Scratch that: I was wrong. Kusaal does have named months: there’s a complete set in Niggli’s Toende Kusaal dictionary. The names given there are farmer-oriented, like the Mooré ones (“Harmattan Month” for February, for example, and “Gleaning Month” for April.) Again, they don’t look very ancient, though: “Christmas Month” for December does not suggest high antiquity to me (the word for “Christmas” is an Akan loan, incidentally, itself of unknown origin.)

    Interesting. I wonder if this is a Francophone thing? None of my dictionaries for Ghanaian Western Oti-Volta languages gives any month names. But then my Buli dictionary does …

    The names don’t match in meaning: Buli October is “Groundnut Month” (perhaps related to Groundhog Day?)

    Further research needed …

  43. David Marjanović says

    Week numbers are often printed in day planners and the like in mainland Europe, but this is the first time I learn they’re actually used anywhere. I have no idea what number the current week is.

  44. The week numbers are an ISO standard: you can define week 1 equivalently as the first (Monday-based) week with a majority (four or more) of its days in the new year, as Trune says, or the week containing the first Thursday in January, or the week containing January 1, or the week with the first workday (assuming that Saturday, Sunday, and New Year’s Day are all non-work days).

    The ISO week-based year (an awkward term, but there it is) begins on Monday of week 1 and ends on Sunday of either week 52 or week 53. So for example the ISO week-based year 2020 begins on December 30 and ends on January 3. Only when December 31 is a Thursday (or a Wednesday in leap years) are there 53 weeks. The notation for writing such dates is 2020-W01-1 and 2020-W53-7 (you can leave out the hyphens).

  45. Trond Engen says

    I didn’t mean to suggest that we go around remembering the week number — being able to deduce the week number from a date is even seen as some sort of superpower, almost on par with division by decimal numbers — but it is used regularly and considered an integral part of the calendar.

  46. ~~ sigh ~~

    I guess I’ll have to include them after all in the date-time package for the Scheme programming language that I’m going to be writing in a few years, God willin’ and the oceans don’t rise.

  47. In Germany, week numbers are used regularly in business scheduling; setting a meeting or a task deadline for Kalenderwoche XY is a normal thing in German organisations. You normally don’t know them by heart, because you can always look them up in a calendar. My Outlook calendar has them, and came with them in when I received my notebook, but that may be a default setting for Germany, seeing that the Americans among us seem to be so unfamiliar with this.

  48. I just took a look at my pocket calendar (they used to be Week-at-a-Glance, but I now get free ones from my dentist) and a careful inspection shows no indication of week number.

  49. But I just checked an older model and it did have week numbers in small type. So there you go: America is a land of contrasts.

  50. Kalenderwoche XY is a normal thing in German organisations. You normally don’t know them by heart

    Bürohengste do know them by heart. I remember only what month it is, and on which weekday the first of that month fell. From those I can quickly reconstruct any dates I need. I use this to find the KW at any given instant of Echtzeit.

  51. I find that in Google Calendar you can turn on week numbers with (Gear) / Settings / View Options / Show week numbers. However, they are visible in the year-by-year and month-by-month calendars only; fortunately, they are also shown in the little month display in the corner of the week-by-week calendar, from which I can instantly determine that this is week 48. I just have to be careful to remember that Sunday belongs to the previous week, because my calendar is set to Sunday-first and week numbers are always Monday-first. This wouldn’t affect people who hide Saturday and Sunday.

  52. Athel Cornish-Bowden says

    How universal is the system of week numbers?

    At Oxford University there are (or were 50 years ago) exactly eight weeks in the undergraduate term. The were numbered from 1 to 8 (not from 1 to 52 or 53). It was completely standard to refer to particular events as occurring in, say, fourth week (not week 4).

  53. The Bard whom pilf’red Pastorels renown,
    Who turns a Persian tale for half a crown,
    Just writes to make his barrenness appear,
    And strains, from hard-bound brains — eight lines a year
         —A. Pope

  54. Stu Clayton says

    “hard-bound” = constipated, I bet.

  55. JC 2019:

    I guess I’ll have to include them after all in the date-time package for the Scheme programming language that I’m going to be writing in a few years, God willin’ and the oceans don’t rise.


  56. For more on history of inverted semicolons and question marks:
    The Question of Question Marks in Greek Manuscripts
    by Brent Nongbri, Variant Readings blog

  57. Thanks! An excerpt:

    But I have wondered for a while: When, exactly, do we start seeing the ; symbol as question mark in Greek manuscripts? The textbooks have not been especially helpful. The clearest statement I have seen is Edward Maunde Thompson writing in 1912 (An Introduction to Greek and Latin Palaeography):

    “The mark of interrogation also first appears about the eighth or ninth century” (p. 60).

    Frustratingly, Thompson does not provide any examples. The earliest instance I know of is a manuscript of Genesis in the Bodleian Library at Oxford (Bodleian Library MS. Auct. T. inf. 2). This is a parchment manuscript copied partially in majuscules and partially in minuscules. Some authorities assign it to the 9th century, others to the 10th century.

  58. David Marjanović says


    The oceans do rise… Florida is already in Egypt, and we should start planning the evacuation of Bangladesh.

  59. So what you’re saying is that JC hasn’t written the date-time package for the Scheme programming language.

  60. John Cowan says

    I have a proposal, but I’m not happy with it yet. It’s hard to get this domain right.

  61. Stu Clayton says

    Why not implement it as a runtime call into an open source “DLL” where date-time has already been correctly implemented ? Only counsels of perfection and purity could stand in your way.

  62. John Cowan says

    No portable FFI, basically. Not even a portable FFI model. Anyway I have yet to find a satisfactory implementation anywhere, although Java is getting better.

  63. Stu Clayton says

    The Java 8 Date/Time API has been around for 8 years now.

    The date/time classes are the Java model. Do you already have a date/time model in Scheme ? If not, why not use the Java model ? If doch, then you write yourself a converter between (the input/output of) the two models. That should keep you busy for a few months.

    In any case, any Java web server provides convenient wherewithal to call into Java. You should have a VM running all the time, otherwise it will need to start on every call from Scheme and that’s slow. Even a Scheme program must run on a VM allowing you to call into, or out of it, by whatever nifty mechanisms.

    I’m not clear on what you mean by “portable”. If you mean “autonomous, consisting only of Scheme code”, prepare to die an inglorious Heldentod.

    Why does Scheme need fancy date/time at all ?

  64. Stu Clayton says

    Instead of using a free-standing web server, just put the whole kaboodle into a docker container.

  65. John Cowan says

    Do you already have a date/time model in Scheme?

    I’m not sure if “you” is sg. or pl. here. There are many Scheme date-time libraries, each with its own model; since there are some 77 incompatible Scheme implementations, this is unsurprising. (Scheme is not a single language like Java.) I am attempting to construct a satisfactory model that can then be implemented by all, with as much of it as possible implementable portably. Fortunately, given a notion of the current time (or several notions), almost everythng else can be done portably.

    any Java web server provides convenient wherewithal to call into Java

    Portable HTTP, or even TCP, is something else Scheme doesn’t have yet.

    If you mean “autonomous, consisting only of Scheme code”, prepare to die an inglorious Heldentod.

    I do mean that, but I do think it is practical; the important point is that the result is right.

    Why does Scheme need fancy date/time at all ?

    Why does Java?

  66. Lars Mathiesen says

    Allow me to point out at this point my frustration with the mindset that leads Java programmers to do their Transport Layer Security processing with an old set of classes from Sun that were found in a box in the closet and haven’t seen a patch in fourscore and seven years, instead of calling out to OpenSSL that has had a stable ABI for ages, is patched regularly at the OS level and produces error messages that can be understood without asking the original programmer. Of course Scheme dudes would never be that benighted.

    (And then they are surprised when everything in the world, except their own test servers but including the load balancer for the production data base, starts to refuse the TLS 1.0 connections that are all their legacy code can produce. And want me to “fix” the world).

    TL;DR: Developers!


  1. […] Hat examines the origins of the semicolon, in Venice in […]

Speak Your Mind