# When should I not use a ligature in English typesetting?

Typesetting that goes beyond the scope of basic MS Word (e.g. LaTeX, or even modern Word versions with a good OpenType font) often uses ligatures for certain glyph combinations, the most common being

• f + f = ﬀ
• f + i = ﬁ
• f + l = ﬂ
• f + f + i = ﬃ
• f + f + l = ﬄ.

I would assume, however, that there are cases in which a ligature shouldn't be used. Are there any rules or standards telling me when I should use a ligature and when I shouldn't?

-
I don't think this is your question, but you shouldn't use ligatures if you expect people to be searching your document in an electronic format. For example, if you use a ligature for the double f in the word efficient then any searches of your document for that word will not produce any results. –  SigueSigueBen Dec 7 '11 at 16:07
@SigueSigueBen: It depends on the program that creates the pdf file: Some will indeed insert unsearchable stuff, whereas others don't and hence don't cause problems when you're searching the file for words such as "fit", "fly", "off", "baffle", or "difficult". :-) Obviously, if you have a choice in pdf-creating programs, you should choose one of the latter type. :-) –  Mico Dec 7 '11 at 17:34
@SigueSigueBen: If you compile a minimal \documentclass{article}\begin{document}ff fi fl\end{document} with an up-to-date pdfLaTeX, the resulting ligatures in the Computer Modern font will indeed be searchable. For other setups and fonts, adaptions may be necessary, though, see e.g. tex.stackexchange.com/questions/4397/…. –  doncherry Dec 7 '11 at 17:52
@SigueSigueBen: I think the key is that one should be using a modern TeX engine which creates a pdf file directly from the tex input file(s), rather than go through the older, and no longer necesssary process from .tex -> .dvi -> .ps -> .pdf. –  Mico Dec 7 '11 at 22:48
@tchrist I've found the problem, and in fact it lies in the font, not TeX and not the reader. There is a StackExchange question which deals with this exact issue: Make ligatures in Linux Libertine copyable (and searchable). Ultimately, it is something to be aware of when creating documents which use typographic ligatures. –  SigueSigueBen Jan 20 '12 at 17:05

At the end of this answer is a list of words for which various f-ligatures should not be employed if using them might impair the words' readability. Obviously, I make no claim as to this list's completeness. Comments and suggestions for additions welcome. (For the curious: I've come up with this list in preparation for a LaTeX package I'm creating that will provide a turnkey solution to the need to more-or-less automate the exclusion of some words from LaTeX's ligation algorithms.)

Aside: The "standard five" f-ligatures -- ff, fi, fl, ffi, and ffl -- that are provided by many fonts and the ft ligature provided by some fonts are the main subject of this answer,. That said, there are naturally many other though less commonly available ligated characters -- such as tt, st, and ct -- to which the following discussion applies as well. I.e., I'd argue that even if the wonderfully charming st ligature (drawn with an arc from the top of the s to the top of the t) is available in a given font, one should probably not use it in the word painstaking -- unless one wishes to make the reader think about pain inflicted by staking (a body part?) ...

As I understand it, the principle that suggests not using certain ligatures in certain words is that ligatures should not cross morpheme boundaries. Loosely speaking, morphemes are the smallest semantically meaningful units inside a word, i.e., the particles that carry/convey distinct meaning. Why? If in a given word two (or more) letters that could be ligated actually belong to separate morphemes, one risks decreasing the readability of the word (and of the entire text) by ligating them anyway, because the visual unity created by the ligature may conflict with the characters' separateness in terms of the morphemes they belong to. For instance, one should wish to avoid employing the ft ligature in "rooftop" because the f and t belong to distinct morphemes.

Of course, just how much readability is impaired in practice by a use of ligatures across morpheme boundaries depends both on the morphemes involved and on the way the glyph that contains the ligature is drawn in a given font. Comparing, for instance, the appearance of the words

with the appearance of

I daresay that few people will argue that the readability of the words shown in the upper row is enhanced by the use of the fi and fl ligatures. This is because the two ligatures unnecessarily risk making your readers puzzle -- if only briefly -- what's going on with these baffling fish and fly species. (Incidentally, the "fish"-words are typeset in Baskervald ADF and the "fly"-words are typeset in EB Garamond.) In contrast, I'll opine confidently that even if the ft letter pair in "fifty" does get ligated, no readers will be misled, slowed down, or otherwise inconvenienced by the use of the ligature. (Of course, "fty" is not AFAIK an English language morpheme -- unlike the "fly" and "fish" cases noted earlier.)

Clearly, using ligatures is not (and never has been) an absolute requirement for making a text readable. Otherwise, virtually all documents typeset with general-purpose word processing programs wouldn't be readable, right? Nevertheless, a consistent use of the ligatures provided by a font family does constitute a mark of good typography, in part because avoiding visual interference of glyph pairs (and triples) improves the readability of the words in question. That said, depending on the word and morphemes involved, I'd argue that not all letter combinations that can be ligated should always be ligated, i.e., for some words the use of ligated glyphs instead of the constituent glyphs need not be a good thing (and may even be a bad thing). In particular, if the letters involved in the ligature belong to separate morphemes, the word's overall legibility may be reduced by the use of a ligature because the visual unity created by the ligatures may impair, however fleetingly, the reader's ability to perceive the underlying morphemes.

Be sure to also include the appropriate plural forms of the nouns in this list.

ff to f-f

shelfful bookshelfful mantelshelfful

fi to f-i

elfin

chafing leafing loafing sheafing strafing vouchsafing; beefing reefing, briefing debriefing; coifing fifing jackknifing knifing midwifing waifing wifing; (air-, child-, fire-, flame-, moth-, rust-, sound-, water-, weather-) proofing goofing hoofing (re)roofing spoofing whoofing woofing; (be-, en-, in-) gulfing, golfing gulfing rolfing selfing wolfing; barfing (be-) dwarfing enserfing kerfing scarfing snarfing (wind-) surfing turfing wharfing

deafish dwarfish elfish oafish selfish serfish unselfish wolfish (plus -ness nouns and -ly adverbial forms)

beefier comfier goofier gulfier leafier surfier turfier (plus -iest superlative forms)

beefily goofily, goofiness

fl to f-l

beefless briefless hoofless leafless roofless selfless turfless (plus adverbial and noun forms: -ly -ness, e.g., selflessly)

aloofly briefly chiefly deafly liefly

calflike dwarflike elflike gulflike hooflike leaflike rooflike serflike sheaflike shelflike surflike turflike waiflike wolflike

halflife shelflife, halfline roofline

leaflet leaflets leafleted leafleting leafletting leafletted leafleteer

pdflatex (Ok, you've probably got to be a TeXie in order to "get" this word!)

ffi to ff-i

baffing biffing (out-) bluffing boffing buffing chaffing cheffing chuffing coffing coiffing (hand-, un-) cuffing daffing doffing (en-, in-) feoffing fluffing gaffing gruffing huffing luffing miffing muffing offing piaffing puffing quaffing rebuffing reffing restaffing restuffing riffing (cross-, over-, under-) ruffing sclaffing scoffing scuffing shroffing sluffing sniffing snuffing spiffing (over-, under-) staffing stiffing (over-) stuffing tariffing tiffing waffing whiffing yaffing

draffish giraffish gruffish offish raffish sniffish standoffish stiffish toffish (plus adverbial and noun forms: draffishly giraffishly etc.)

buffier chaffier chuffier cliffier daffier fluffier gruffier huffier iffier miffier puffier scruffier sniffier snuffier spiffier stuffier (plus superlative forms: buffiest, chaffiest, etc.)

daffily fluffily gruffily huffily puffily scruffily sniffily snuffily spiffily stuffily

fluffiness huffiness iffiness puffiness scruffiness sniffiness spiffiness stuffiness

baffies biffies jiffies taffies toffies

waffie

Pfaffian Wolffian Wulffian

ffl to ff-l

bluffly gruffly ruffly snuffly stiffly

cuffless stuffless (and adverbial forms: cufflessly, ...)

scofflaw, cufflink, offline, offload (plus -s, -ed, and -ing affixes)

rufflike clifflike

ffi to f-fi

chaffinch; wolffish

ffl to f-fl

safflower

ft to f-t

chieftain chieftaincy chieftainship

fifteen fifteens fifteenth fifteenths fifth fifthly fifths fifties fiftieth fiftieths fifty fiftyish

halftime halftone

rooftop rooftree

fft to ff-t (not a common ligature, but Minion Pro and EB Garamond do provide it)

offtrack

-
That's awesome, I'm excited for your package! Is there any typographic reference manual or the like stating this rule? (I've heard the morpheme rule as well, but some reference would be nice.) Are you interested in undesired ligatures in other languages as well? I'd assume there are many German LaTeX aficionados who'd be willing to help you by collecting words. –  doncherry Dec 7 '11 at 17:59
It has examples of words with two ligatures: fluffier firefly fisticuffs, flagstaff fireproofing, chiffchaff and riffraff. Notice "fluffier" and "fireproofing" where you would not ligate the ff in the former and the fi in the latter. –  egreg Dec 9 '11 at 21:42
@egreg: Thanks for this. I will admit that, from a typographical standpoint, not all suffix morphemes may be deserving of equal respect (of not being ligated). What should matter most in this regard, as with all other aspects of good typography, is enhancing readability of the text. For the f-ly and f-ish suffix cases, I'd argue strongly that by not ligating the f-l and f-i pairs one is being very kind to the reader. The f-ier and f-ing suffix cases are a bit more debatable, and the ff-ier case may well be a case where any benefits of being a linguistic stickler are just negligible. :-) –  Mico Dec 9 '11 at 22:05
@egreg: That's my main point too, I believe: Knuth doesn't explain explicitly why "shelfful" looks better to him, but it can't be because Computer Modern's ff ligature isn't sufficiently good-looking on its own, right? I can only surmise that the reason he thinks "shelfful" looks better without the ligature is because he somehow thinks there's a chance that the reader might get tripped up briefly by the presence of a ligature that spans morpheme boundaries. Naturally, this is just my "educated" guess. Do you happen to know if DEK has clarified his thinking on this topic elsewhere? –  Mico Dec 10 '11 at 0:06
@Mico I am referring to how proper typesetting software will, or can be asked to, implicitly select the ligatures provided by the actual font used. That way you never have to do anything special. Even troff had the .lg directive. OpenType font support discretionary ligatures and other sorts of contextual glyphs. The ligatures are usually divided into two classes, standard and historical. (Not all software gives you convenient access to these at any finer granularity.) On a Mac, use an OpenType font that supports them, & either FontBook or TextEdit to play with these implicit ligatures. –  tchrist Jan 17 '12 at 21:48

(Edited for typos and to hand-tweak the typography. See new appendix at bottom.)

Are there any rules or standards telling me when I should use a ligature and when I shouldn’t?

Yes.

In version 3.2 of Robert Bringhurst’s The Elements of Typographic Style (Hartley & Marks, 2008), the main discussion of ligatures occurs on pages 50–53, under the section ‘3.3 Ligatures’, which has two subjections I shall describe below. Signiﬁcant mentions of ligatures also occur at perhaps a half-dozen other points in this book.

This is one text that truly cannot come too highly recommended. If you do not own it, you should. It comes in both trade-paper and hardcover editions. Buy it. No scanned version is available via Google Books, and it is unlikely that it ever shall be — for reasons that should become quickly apparent once you crack its spine. The hardcover has a sewn binding with a built-in red silk bookmark, and costs only \$10 more than the trade paper; seems worth the upgrade to me. But both bindings are remarkably well-made books. Just make sure to get the most recent version; as of this writing, that is version 3.2 from 2008.

The ﬁrst subsection of ‘3.3 Ligatures’, which is on page 50, is entitled:

3.3.1 Use the ligatures required by the font, and the characters required by the language, in which you are setting the type.

In most roman faces the letter f reaches into the space beyond it. In most italics, the f reaches into the space on both sides. Typographers call these overlaps kerns. Only a few kerns, like those in the arm of the f and the tail of the j, are implicit in a normal typefont, while others, like the overlap in the combination To, are optional reﬁnements, independent of the letterforms.

Reaching into the space in front of it, the arm of the f will collide with certain letters – b, f, h, i, j, k, l – and with question marks, quotation marks or parentheses, if these are in its way.

As the craft of typography spread through Europe, new regional ligatures were added. An fj and æ were needed in Norway and Denmark for words just as fjeld and fjord and nær. In France an œ, and in Germany an ß (eszett or double-s) were required, along with accented and umlauted vowels. Double letters which are read as one – ll in Spanish, ij in Dutch, and ch in German, for example – were cast as single sorts for regional markets. An ffj was needed in Iceland. … Purely decorative ligatures were added to many fonts as well.

If your software is inserting ligatures automatically, take a moment to verify two things: (1) that the software is inserting all the ligatures you want and none that you do not want; (2) that all these ligatures are staying where they’re put.

That goes on for quite some time, and is important. I’ll try to summarize all this below instead of typing it all in. The other subsection, which isn’t until page 52, is entitled:

3.3.2 If you wish to avoid ligatures altogether, restrict yourself to faces that don’t require them.

It is quite possible to avoid the use of ligatures completely and still set beautiful type. All that is required is a face with non-kerning roman and italic f – and some of the ﬁnest twentieth-century faces have been deliberately equipped with just this feature. Aldus, Melior, Menoza, Palantino, Sabon, Trajanus and Trump Mediäval, for example, all set handsomely without ligatures. Full or partial ligatures do exist for these faces, and the ligatures may add a touch of reﬁnement – but when ligatures are omitted from these faces, no unsightly collisions occur.

As always in Bringhurst, you should pay careful attention to the thoughtful and revealing sidenotes in this section. They contain myriad wonderful examples of typefaces and effects that are completely impossible to reproduce here. It’s much too much for me to type in, but do please read it. Here’s just one from page 51:

Decorative ligatures such as and Th are now deservedly rare. The ligature ffﬂ is rarer still, but it has been cut for at least one typeface(Jonathan Hoeﬂer’s Requiem italic).

What I will type in in toto, however, is the deﬁnition from the glossary from page 312:

ligatures Basic ISO fonts are limited to typographic ligatures, and . Rigid deﬁnitions of the glyph set, leaving no provision for additional ligatures (such as ff, fﬁ, fﬂ, fj) are a hazard to typography. Ligatures required by the design of the individual typeface should always reside on the basic font.         The lexical ligatures ligatures æ, Æ, œ, Œ and ß are bonaﬁde Unicode characters, separately listed in this appendix. Typographic ligatures such as and are glyphs, not characters, but many are now included in Unicode as “presentations forms.” [U+FB00—FB06]

When Bringhurst writes ‘glyph’ above, he means what a hand compositor would call a ‘sort’: something that would have its own slot in the compositor’s case. Most (but not all) digital fonts are distressingly impoverished when it comes to glyph variants. For his 42-line Bible, Gutenberg needed perhaps 40 sorts; he used 290. That tells you how important they are.

## Typographic Ligatures

However, you should not encode attempt to encode any typographic ligatures into your text, even for those few for which Unicode code points exist. (Edit: but see the bottom of this posting for why I have been forced to do so here.) Rather, this is something that must be done for you automatically by your typesetting software. You may be able to enable or disable such implicit selection, or distinguish between basic (typographic) ligatures and discretionary (typographic) ligatures.

Here are several discrete font suggestions with good typographic ligatures:

• One Western (Latin, Greek, and Cyrillic) OpenType face particularly rich in typographic ligatures, and which also includes the best Unicode support (highest number of glyphs, including many variants) seen to date, is Richard Slimbach’s Arno Pro, released in 2007. Slimbach is also the designer of Minion and Poetica. I can’t begin to describe how nifty this font is. (Well, I suppose I could, considering that I have an entire 19-page write-up of it which still only barely scratches the surface. But I have no space for that here.) It is based on early Renaissance forms, and is quite possibly already on your computer, as it was included with Adobe’s CS3. If you like fonts cut by Claude Garamond (or Garamont), Robert Granjon, or Jean Jannon, then you will like Slimbach’s Arno Pro.

• For typographic ligatures in general roman and italic faces, little can compare with Zuzana Ličko’s lavish Mrs Eaves (1996), which sports some 71 typographic ligatures. It is based on (the original) Baskerville. That’s for basic Latin, though. It does not have anything like the Unicode support that Slimbach’s much more recent Arno Pro offers.

• For script faces, master punchcutter and designer Hermann Zapf ’s Zapﬁno and its OpenType version, Zapﬁno Extra (2004), are without peer. In Zapﬁno, the very word Zapﬁno is itself a ligature, and comes with several outstanding variants. There really is no other font like Zapﬁno; Bringhurst calls it a calligraphic tour de force: “It is an exemplary marriage of artistic and technical ability. Effective use of such a type requires considerable patience and skill.”

• For Greeks, Matthew Carter’s Wilson (2005) is highly recommended. Like Zapﬁno, Wilson is sui generis. It is an exceptionally well-cut chancery Greek, which means it is highly embellished, full of elaborate ligatures and alternate forms. In this it is reminiscent of Robert Granjon’s chancery Greek that he cut in the 1560s, but Wilson gets its name from the fonts it is more directly based on, those of the eighteenth-century master, Alexander Wilson of Glasgow, who was also an astronomer and a physician.

Those fonts I’ve listed above all handle typographic ligatures exceptionally well. But as I’ve said, you should not encode typographic ligatures in your own text. (Edit: But see the bottom of this posting for why I’ve been forced to do so here.) That is for the font to decide for you. See also the Unicode FAQ on Ligatures, Digraphs, and Presentation Forms, and take especial note of this answer:

A: The existing ligatures exist basically for compatibility and round-tripping with non-Unicode character sets. Their use is discouraged. No more will be encoded in any circumstances.

Ligaturing is a behavior encoded in fonts: if a modern font is asked to display “h” followed by “r”, and the font has an “hr” ligature in it, it can display the ligature. Some fonts have no ligatures, some (especially for non-Latin scripts) have hundreds. It does not make sense to assign Unicode code points to all these font-speciﬁc possibilities.

However, you should deﬁnitely let the font use them if they are needed in that font. It’s the same as with kerning. If the font needs it to avoid collisions, you must use it or else use a different font; see the second of Bringhurt’s ligature subsections.

For example, in Georgia (which I believe you are reading this in right now), you cannot write things like f ’ and f ” without unacceptably horrible collisions producing f’ and f”. That’s just one reason web pages, even this one, frankly look like crap. Fonts designed with typographic ligatures and kerning in mind cannot be acceptably used without them.

## Lexical Ligatures

Lastly we come to lexical ligatures. Here the story is quite different. There are absolutely required by the language. For example, the name of famed Icelandic programmer Ævar Arnfjörð Bjarmason cannot be written any other way. Writing **AEvar* or **Aevar* would simply be wrong. In the same way, if you were talking about the Old English word that became modern wrath, you would have to write wrǽððu or wrǽþþu (both were known to occur). Similarly, French words and phrases like un œuf, hors d’œuvres, and ma jolie sœur are not supposed to be written broken up; it is highly frowned upon in French. The Encyclopædia Britannica actually has that lexical ligature in its registered trademark, and so must be maintained. And IPA uses both œ and ɶ (the small caps version) to mean separate vowels; you mustn’t mess with those. All those are lexical ligatures.

For other cases in English, the style has largely shifted, although it does still depend to some degree on the publication and nation of origin. Where we once wrote œnophile and amœba, Cæsar and archæology, we tend now to write those with the letters ‘broken apart’, so to speak.

My own recommendation is to use lexical ligatures if you are quoting another source that uses them, especially if they are originally in another language than English, or, if in English, you are trying to reproduce the original. This is the same rule of thumb used for diacritics.

For example, in a poem written in honor of the 19th centenary of Virgil’s death, Alfred, Lord Tennyson wrote such lines as Thou that seëst Universal / Nature moved by Universal Mind alike with Now thy Forum roars no longer, / fallen every purple Cæsar's dome— and you would be remiss not to reproduce those as the poet wrote them. Similarly in Lord Byron’s Episode of Nisus and Euralyus when he wrote Yet wakeful Rhæsus sees the threatening steel or When great Æneas wears Hesperia's crown.

If the original had them, then you likely should, too.

However, I dare say that maintaining those historical forms when writing your own new material upon Caesar or Aeneas is a bit too precious. There’s a reason that some ligatures get called ‘quaints’ by typesetters. You might not care to appear so quaint.

Then again, perhaps you might. It’s really up to you.

## Edit

Because ELU is presented to us in Georgia, I have hand-tweaked some of the typesetting, because with extremely rare exception, web browsers are far too stupid to use the mandatory ligatures and kerns that are absolutely required with this font.

I have therefore actually used the only three Latin typographic ligatures that Georgia provides: ﬁ for fi, ﬂ for fl, and although it appears to be unnecessary, ﬀ for ff. Only the first of those appears to have been strictly necessary for this font, and so it is a bug that browsers don’t do it.

I have also in several places hand-kerned the f so that a following quotation mark, single or double, no longer collides with it. I have done this by placing a U+200A HAIR SPACE after the f and before the quotation mark, making f’ ﬁnally correctly render as f ’, and f” as f ”.

I should not have to do these things, but web browser technology is extremely primitive when it comes to typography. Gutenberg was far superior to anything we have yet achieved in web browsers, which means that we have still not caught up to the very ﬁrst instance of movable type of almost half a millennium ago; pretty sad, isn’t it? However, browser awareness of OpenType hinting is apparently coming, at which point all this silliness should no longer be needed.

I’m sure Georgia is used here because it is the only universally available font that by default uses text ﬁgures instead of titling ﬁgures, as is of course properly done. Unfortunately, Georgia is also one of those fonts that cannot be correctly rendered without typographic ligatures and kerning hints being automatically applied — which is why I’ve been forced to make the various hand adjustments described in this edit.

Unfortunately, Georgia does not have very good Unicode support at all, nor does it have small caps, which are also required for tolerable typesetting.

-
Wow, I'm looking forward to reading this, but now it's really time to go to bed. –  doncherry Jan 18 '12 at 2:24
@Mico You let the font choose the ligatures. You never suppress them. Otherwise you make things that look bad to read. –  tchrist Mar 20 '12 at 14:10
@Mico Yes of course. It’s completely ridiculous to pretend that automatically applying the typographic ligatures built into a font harms legibility. Read Bringhurst. A well-designed font is made with that in mind, so that those letters correctly fit together as implicit typographic ligatures. To not use them is to invite ugly collisions that ruin the text. It has nothing to whatsobloodyever to do with semantics; it’s a simple matter of font design, so just use all the standard built-in typographic ligatures. Lexical ligatures are a different matter. –  tchrist Mar 20 '12 at 16:02
@Mico Bringhurst is commenting on the way Adobe puts the Th ligature in the standard set in their OpenType fonts. His point is that it is a decorative ligature like ct, and that it doesn’t belong in the standard set any moreso than quaints like ct or st do — at least if you have a sense of the historical uses of these things. He isn’t commenting on things like your examples of cufflink, wolfish, or selfless. You’re right that I utterly reject the notion that using the fonts’ implicit typographic ligatures in such words can ever ‘harm legibility’ as you allege. This isn’t German. –  tchrist Mar 20 '12 at 17:02
@Mico Precisely which part of Bringhurst’s “3.3.1 Use the ligatures required by the font, and the characters required by the language, in which you are setting the type.” is it that you disagree with? And yes, despite your protestations to the contrary, I very most certainly do see him supporting my claim that one should use the ligatures required by the font, because that is precisely what he says. Your allegation that one should not do so in words like wolfish is nowhere supported by Bringhurst. To not do so in some fonts will screw up your type, which is his whole point. –  tchrist Mar 20 '12 at 17:24

Many ligatures of Latinate letters are nothing more than scribal abbreviations which have been preserved in type. Readability of the text is often improved by the use of these ligatures, such as ff, fi, and fl. Without ligature, these letters can partially overlap and require manual kerning. With ligature, the spacing is perfect. Ligatures such as ct and st sometimes have stylistic pen loops which are considered to contribute to the beauty of the line.

By contrast, some non-Latinate written languages have hard and fast rules for ligature.

Chicago Manual of Style advises that you should avoid ae and oe ligatures "in Latin words or transliterated Greek words", and "in words adopted into English from Latin, Greek, or French (and thus to be found in English dictionaries)".

Otherwise, I am aware of no authority which prohibits ligature in English text, such as between syllables. So it is up to the typesetter to determine what will make the most readable product.

-
You’re mostly right. However, the main scribal abbreviation preserved in modern type that comes readily to mind is the ampersand proper: &. Especially in humanist Renaissance faces and their latterday reimplementations, including the one in the banner at the top, the scribal connection to the Latin et is quite clear. The Unicode Latin compatibility ligatures, meaning ﬀ, ﬁ, ﬂ, ﬃ, and ﬄ, plus the two quaints, ﬅ and ﬆ, were never scribal abbreviations. Rather, they are hand compositors’ ‘abbreviations’, dedicated sorts needed to avoid collisions in metal type. –  tchrist Jan 17 '12 at 21:39
+1 this comment, I can't learn enough about typography. Now what you say seems to be contradicted by Wikipedia and this author. Both say that cast ligatures directly reproduce scribal handwriting, going right back to Gutenberg. So who's right and how do we know? –  MετάEd Jan 17 '12 at 22:29
Incidentally, the author I linked to above points out another scribal abbreviation preserved in modern type: the question mark, originally q over o, short for questio. –  MετάEd Jan 17 '12 at 22:30
Insofar as scribes wrote sequences like ff as a single glyph of two letters tied together with a common stoke, or fi in a way that was different than what they did when the i did not follow the f, it is true that these are reflected in metal. However, the main reason for essential ligatures to exist is because some fonts require them, because otherwise you will get collisions, especially when unkerned. I have a long answer I need to write up. –  tchrist Jan 17 '12 at 22:45
That will be great. Based on what you say here, it's not strictly true that the ff ligature, for example, wasn't originally a scribal ligature. It was, but when the transition was made to type the ligature survived because it was as useful in typesetting as in handwriting, albeit for different reasons. –  MετάEd Jan 17 '12 at 23:08

A rule (not always observed) is not to use a ligature spanning syllables. Chaffinch has syllables chaf-finch, for example. So don't use the ffi ligature in there.

-
Also in "efficient" the syllables are "ef-fi-cient", but the ligature is used. You probably meant a ligature spanning components of a compound word. –  egreg Dec 5 '11 at 16:03
I said: not always observed. Some would say that ligature ff should not be used in efficient. Personally, I never worry about this. –  GEdgar Dec 5 '11 at 16:07
Can you show examples where a non final "ff" doesn't span two syllables? –  egreg Dec 5 '11 at 16:09
@egreg Take any of dozens of short verbs (buff, cuff, huff, ruff, bluff, stuff...) and add an -ed suffix (so, buffed, cuffed...) –  jwpat7 Dec 5 '11 at 19:36
We've gotta be careful not to intermix the terms syllable and morpheme. These units may, but don't necessarily have to coincide, e.g. purple has two syllables, but it is monomorphemic. The Latin compound words are a tricky case indeed, but I think I'd agree with you that {effic} is one morpheme, my reason being that no native speaker of English would use *{fic} as a morpheme productively, not even as a joke. Maybe we should get Alan Munn involved in this discussion, he could certainly provide some interesting insights, or help with your package. –  doncherry Dec 7 '11 at 23:26

There is no rule that ever requires or prohibits the use of a ligature in English - it is purely a matter of style.

-