% language=uk \startcomponent languages-hyphenation \environment languages-environment \startchapter[title=Hyphenation][color=darkmagenta] \startsection[title=How it works] Proper hyphenation is one of the strong points of \TEX. Hyphenation in \TEX\ is done using so called hyphenation patterns. Making these patterns is an art and most users (including me) happily use whatever is available. Patterns can be created automatically using \type {patgen} but often manual tweaking is needed too. A pattern looks as follows: \starttyping pat1tern \stoptyping This means as much as: you can split the word \type {pattern} in two pieces, with a hyphen between the two \type {t}'s. Actually it will also split the word \type {patterns} because the hyphenation mechanism looks at substrings. When no number between characters in a pattern is given, a zero is assumed. This means as much as {\em undefined}. An even number inhibits hyphenation, an odd number permits it. The larger the number (weight), the more influence it has. A more restricted pattern is: \starttyping .pat1tern. \stoptyping Here the periods set the word boundaries. The pattern dictionary for us english has smaller patterns and the next trace shows how these are applied. \starthyphenation[traditional] \showhyphenationtrace[en][pattern] \stophyphenation The effective hyphenation of a word is determined by several factors: \startitemize[packed] \startitem the current language, each language can have different patterns \stopitem \startitem the characters, as some characters might block hyphenation \stopitem \startitem the settings of \type {\lefthyphenmin} and \type {\righthyphenmin} \stopitem \stopitemize A place where a word can be hyphenated is called a discretionary. When \TEX\ analyzes a stream, it will inject discretionary nodes into that stream. \starttyping pat\discretionary{-}{}{}tern. \stoptyping In traditional \TEX\ hyphenation, ligature building and kerning are tightly interwoven which is quite effective. However, there was also a strong relationship between the current font and hyphenation. This is a side effect of traditional \TEX\ having at most 256 characters in a font and the fact that the used character is fact a reference to a slot in a font. There a character in the input initially ends up as a character node and eventually becomes a glyph node. For instance two characters \type {fi} can become a ligature glyph representing this combination. In \LUATEX\ the hyphenation, ligature building and kerning stages are separated and can be overloaded. In \CONTEXT\ all three can be replaced by code written in \LUA. Because normally hyphenation happens before font logic is applied, there is no relationship with font encoding. I wrote the first \LUA\ version of the hyohenator on a rainy weekend and the result was not that bad so it was presented at the 2014 \CONTEXT\ meeting. After some polishing I decided to add this routine to the standard \MKIV\ repertoire which then involved some proper interfacing. You can enable the \LUA\ variant with the following command: \starttyping \setuphyphenation[method=traditional] \stoptyping We call this method \type {traditional} because in principle we can have many more methods and this one is (supposed to be) mostly compatible to the built-in method. This is a global setting. You can switch back with: \starttyping \setuphyphenation[method=default] \stoptyping In the next sections we will see how we can provide alternatives within the traditional method. These alternatives can be set local and therefore can operate over a limited range of characters. One complication in interfacing is that \TEX\ has grouping (which permits local settings) and we want to limit some of the above functionality using groups. At the same time hyphenation is a paragraph related action so we need to enable the hyphenation related code at a global level (or at least make sure that it gets exercised by forcing a \type {\par}). That means that the alternative hyphenator has to be quite compatible so that we could just enable it for a whole document. This can have an impact on performance but in practice that can be neglected. In \LUATEX\ the \LUA\ variant is 4~times slower than the built-in one, in \LUAJITTEX\ it's 3~times slower. But the good news is that the amount of time spent in the hyphenator is relatively small compared to other manipulations and macro expansion. The additional time needed for loading and preparing the patterns into a more \LUA\ specific format can be neglected. You can check how words get hyphenated using the patterns management script: \starttyping >mtxrun --script patterns --hyphenate language hyphenator | hyphenator | . l a n g u a g e . . l a n g u a g e . hyphenator | 0a2n0 0 0 2 0 0 0 0 0 0 hyphenator | 2a0n0g0 0 2 2 0 0 0 0 0 0 hyphenator | 0n1g0u0 0 2 2 1 0 0 0 0 0 hyphenator | 0g0u4a0 0 2 2 1 0 4 0 0 0 hyphenator | 2g0e0.0 0 2 2 1 0 4 2 0 0 hyphenator | .0l2a2n1g0u4a2g0e0. . l a n-g u a g e . hyphenator | mtx-patterns | us 3 3 : language : lan-guage \stoptyping \stopsection \startsection[title=The last words] Mid 2014 we had to upgrade a style for a \PDF\ assembly service: chapters from (technical) school books are combined into arbitrary new books. There are some nasty aspects with this flow: for instance, all section numbers in a chapter are replaced by new numbers and this also involves figure and table prefixes. It boils down to splitting up books, analyzing the typeset content and preparing it for replacements. The structure is described in \XML\ files so that we can generate tables of contents. The reason for not generating from \XML\ sources is that the publisher doesn't have a \XML\ workflow and that books already were available. Also, books from several series are combined and even within a series structure (and rendering) differs. What has this to do with hyphenation? Writing a style for such a flow always results in a more complex one that estimated and as usual it's in the details. The original style was written in \MKII\ and used some box juggling to achieve reasonable results but in \MKIV\ we can do better. Each chapter has a title and books get titles and subtitles as well. The titles are typeset each time a new book is composed. This happens within some layout constraints. Think of constraints like these: \startitemize[packed] \startitem the title goes on top of a shape that doesn't permit much overflow \stopitem \startitem there can be very long words (not uncommon in Dutch or German) \stopitem \startitem a short word or hyphenated part should not end up on the last line \stopitem \startitem the left and right hyphenation minima are at least four \stopitem \stopitemize The last requirement is a compromise because in most cases publishers seem to want ragged right not hyphenated rendering (at least in Dutch schoolbooks). The arguments for this are quite weak and probably originate in fear of bad rendering given past experiences. It's this kind of situations that drive the development of the more obscure features that ship with \CONTEXT\ and a (partial) solution for this specific case will be given later. If you look at thousands of titles and turn these into (small) paragraphs \TEX\ does a pretty good job. It's the few exceptions that we need to catch. The next examples demonstrate such an extreme case. \startbuffer[example] \dorecurse{5} { % dejavu \startlinecorrection[blank] \bTABLE \bTR \bTD[align=middle,width=2em,foregroundstyle=bold] #1 \eTD \bTD[align={verytolerant,flushleft},width=15em,offset=1ex] \hsize \dimexpr11\emwidth-#1\dimexpr.5\emwidth\relax \dontcomplain \lefthyphenmin=4\righthyphenmin=4 \blackrule[color=darkyellow,width=\hsize,height=-3pt,depth=5pt]\par \begstrut\getbuffer[long]\endstrut\par \eTD \bTD[align={verytolerant,flushleft},width=15em,offset=1ex] \sethyphenationfeatures[demo] \hsize \dimexpr11\emwidth-#1\dimexpr.5\emwidth\relax \dontcomplain \blackrule[color=darkyellow,width=\hsize,height=-3pt,depth=5pt]\par \begstrut\getbuffer[long]\endstrut\par \eTD \eTR \eTABLE \stoplinecorrection } \stopbuffer \definehyphenationfeatures [demo] [rightwords=1, lefthyphenmin=4, righthyphenmin=4] \startbuffer[long] a verylongword and then anevenlongerword \stopbuffer \starthyphenation[traditional] \enabletrackers[hyphenator.visualize] \getbuffer[example]\par \disabletrackers[hyphenator.visualize] \stophyphenation Of course in practice there need to be some reasonable width and when we pose these limits the longest possible word should fit into the allocated space. In these examples the rule shows the width. In the right columns we see a red colored word and that one will not get hyphenated. \stopsection \startsection[title=Explicit hyphens] Another special case that we needed to handle were (compound) words with explicit hyphens. Because often data comes from \XML\ files we can not really control the typesetting as in a \TEX\ document where the author sees what gets done. So here we need a way to turn these hyphens into proper hyphenation directives and at the same time permit the words to be hyphenated. \definehyphenationfeatures [demo] [hyphens=yes, lefthyphenmin=4, righthyphenmin=4] \startbuffer[long] a very-long-word and then an-even-longer-word \stopbuffer \starthyphenation[traditional] \enabletrackers[hyphenator.visualize] \getbuffer[example]\par \disabletrackers[hyphenator.visualize] \stophyphenation \stopsection \startsection[title=Extended patterns] As with more opened up mechanisms, in \MKIV\ we can extend functionality. As an example I have implemented the extensions discussed in the article by László Németh in the Proceedings of Euro\TEX\ 2006: {\em Hyphenation in OpenOffice.org} (TUGboat, Volume 27, 2006). The syntax for these extension is somewhat ugly and involves optional offsets and ranges. \footnote {I'm not sure if there were ever patterns released that used this syntax.} \startbuffer \registerhyphenationpattern[nl][e1ë/e=e] \registerhyphenationpattern[nl][a9atje./a=t,1,3] \registerhyphenationpattern[en][eigh1tee/t=t,5,1] \registerhyphenationpattern[de][c1k/k=k] \registerhyphenationpattern[de][schif1f/ff=f,5,2] \stopbuffer \typebuffer \getbuffer These patterns result in the following hyphenations: \starthyphenation[traditional] \switchtobodyfont[big] \starttabulate[|||] \NC reëel \NC \language[nl]\hyphenatedcoloredword{reëel} \NC \NR \NC omaatje \NC \language[nl]\hyphenatedcoloredword{omaatje} \NC \NR \NC eighteen \NC \language[en]\hyphenatedcoloredword{eighteen} \NC \NR \NC Zucker \NC \language[de]\hyphenatedcoloredword{Zucker} \NC \NR \NC Schiffahrt \NC \language[de]\hyphenatedcoloredword{Schiffahrt} \NC \NR \stoptabulate \stophyphenation In a specification, the \type {.} indicates a word boundary and numbers indicate the weight of a breakpoint. The optional extended specification comes after the \type {/}. The values separated by a \type {=} are the pre and post sequences: these end up at the end of the current line and beginning of the next one. The optional numbers are the start position and length. These default to~1 and~2, so in the first example they identify \type {eë} (the weights don't count). There is a pitfall here. When the language already has patterns that for instance prohibit a hyphen between \type {e} and type {ë}, like \type{e2ë}, we need to make sure that we give our new one a higher priority, which is why we used a \type{e9ë}. This feature is somewhat experimental and can be improved. Here is a more \LUA-ish way of setting such patterns: \starttyping local registerpattern = languages.hyphenators.traditional.registerpattern registerpattern("nl","e1ë", { start = 1, length = 2, before = "e", after = "e", } ) registerpattern("nl","a9atje./a=t,1,3") \stoptyping Just adding extra patterns to an existing set without much testing is not wise. For instance we could add these to the dutch dictionary: \starttyping \registerhyphenationpattern[nl][e3ë/e=e] \registerhyphenationpattern[nl][o3ë/o=e] \registerhyphenationpattern[nl][e3ï/e=i] \registerhyphenationpattern[nl][i3ë/i=e] \registerhyphenationpattern[nl][a5atje./a=t,1,3] \registerhyphenationpattern[nl][toma8at5je] \stoptyping That would work oke well for words like \starttyping coëfficiënt geïntroduceerd copiëren omaatje tomaatje \stoptyping However, the last word only goes right because we explicitly added a pattern for it. One reason is that the existing patterns already contain rules to prevent weird hyphenations. The same is true for the accented characters. So, consider these examples and coordinate additional patterns with other users so that errors can be identified. \stopsection \startsection[title=Exceptions] We have a variant on the \TEX\ primitive \type {\hyphenation}, the official way to register a specific way to hyphenate a word. \startbuffer \registerhyphenationexception[aaaaa-bbbbb] aaaaabbbbb \par \stopbuffer \typebuffer This code is self explaining and results in: \blank \starthyphenation[traditional] \setupindenting[no]\hsize 1mm \lefthyphenmin 1 \righthyphenmin 1 \getbuffer \stophyphenation There can be multiple hyphens and even multiple words in such a specification: \startbuffer \registerhyphenationexception[aaaaa-bbbbb cc-ccc-ddd-dd] aaaaabbbbb \par cccccddddd \par \stopbuffer \typebuffer We get: \blank \starthyphenation[traditional] \setupindenting[no]\hsize 1mm \lefthyphenmin 1 \righthyphenmin 1 \getbuffer \stophyphenation \stopsection \startsection[title=Boundaries] A box, rule, math or discretionary will end a word and prohibit hyphenation of that word. Take this example: \startbuffer[demo] whatever \par whatever\hbox{!} \par \vl whatever\vl \par whatever$x$ \par whatever-whatever \par \stopbuffer \typebuffer[demo] These lines will hyphenate differently and in traditional \TEX\ you need to insert penalties and|/|or glue to get around it. In the \LUA\ variant we can enable that limitation. \startbuffer \definehyphenationfeatures [strict] [rightedge=tex] \stopbuffer \typebuffer \getbuffer Here we show the three variants: traditional \TEX\ and \LUA\ with and without strict settings. \starttabulate[|p|p|p|] \HL \NC \ttbf \hbox to 11em{default\hss} \NC \ttbf \hbox to 11em{traditional\hss} \NC \ttbf \hbox to 11em{traditional strict\hss} \NC \NR \HL \NC \starthyphenation[default] \hsize1mm \getbuffer[demo] \stophyphenation \NC \starthyphenation[traditional] \hsize1mm \getbuffer[demo] \stophyphenation \NC \starthyphenation[traditional] \sethyphenationfeatures[strict] \hsize1mm \getbuffer[demo] \stophyphenation \NC \NR \HL \stoptabulate By default \CONTEXT\ is configured to hyphenate words that start with an uppercase character. This behaviour is controlled in \TEX\ by the \typ {\uchyph} variable. A positive value will enable this and a negative one disables it. \starttabulate[|p|p|p|p|] \HL \NC \ttbf \hbox to 8em{default 0\hss} \NC \ttbf \hbox to 8em{default 1\hss} \NC \ttbf \hbox to 8em{traditional 0\hss} \NC \ttbf \hbox to 8em{traditional 1\hss} \NC \NR \HL \NC \starthyphenation[default] \hsize1mm \uchyph\zerocount TEXified \dontcomplain \stophyphenation \NC \starthyphenation[traditional] \hsize1mm \uchyph\zerocount TEXified \dontcomplain \stophyphenation \NC \starthyphenation[default] \hsize1mm \uchyph\plusone TEXified \dontcomplain \stophyphenation \NC \starthyphenation[traditional] \hsize1mm \uchyph\plusone TEXified \dontcomplain \stophyphenation \NC \NR \HL \stoptabulate The \LUA\ variants behaves the same as the built-in implementation (that of course remains the reference). \stopsection \startsection[title=Plug-ins] The default hyphenator is similar to the built-in one, with a couple of extensions as mentioned. However, you can plug in your own code, given that it does return a proper hyphenation result. One reason for providing this plug is that there are users who want to play with hyphenators based on a different logic. In \CONTEXT\ we already have some methods to deal with languages that (for instance) have no spaces but split on words or syllabes. A more tight integration with the hyphenator can have advantages so I will explore these options when there is demand. A result table indicates where we can break a word. If we have a four character word and can break after the second character, the result looks like this: \starttyping result = { false, true, false, false } \stoptyping Instead of \type {true} we can also have a table that has entries like the extensions discussed in a previous section. Let's give an example of a plug-in. \startbuffer \startluacode local subset = { a = true, e = true, i = true, o = true, u = true, y = true, } languages.hyphenators.traditional.installmethod("test", function(dictionary,word,n) local t = { } for i=1,#word do local w = word[i] if subset[w] then t[i] = { before = "<" .. w, after = w .. ">", left = false, right = false, } else t[i] = false end end return t end ) \stopluacode \stopbuffer \typebuffer \getbuffer Here we hyphenate on vowels and surround them by angle brackets when split over lines. This alternative is installed as follows: \startbuffer \definehyphenationfeatures [demo] [alternative=test] \stopbuffer \typebuffer \getbuffer We can now use it as follows: \starttyping \setuphyphenation[method=traditional] \sethyphenationfeatures[demo] \stoptyping When applied to one the tufte example we get: \startbuffer[demo] \starthyphenation[traditional] \setuptolerance[tolerant] \sethyphenationfeatures[demo] \dontleavehmode \input tufte\relax \stophyphenation \stopbuffer \blank \startnarrower \getbuffer[demo] \stopnarrower \blank A more realistic (but not perfect) example is the following: \startbuffer \startluacode local packslashes = false local specials = { ["!"] = "before", ["?"] = "before", ['"'] = "before", ["'"] = "before", ["/"] = "before", ["\\"] = "before", ["#"] = "before", ["$"] = "before", ["%"] = "before", ["&"] = "before", ["*"] = "before", ["+"] = "before", ["-"] = "before", [","] = "before", ["."] = "before", [":"] = "before", [";"] = "before", ["<"] = "before", [">"] = "before", ["="] = "before", ["@"] = "before", ["("] = "before", ["["] = "before", ["{"] = "before", ["^"] = "before", ["_"] = "before", ["`"] = "before", ["|"] = "before", ["~"] = "before", -- [")"] = "after", ["]"] = "after", ["}"] = "after", } languages.hyphenators.traditional.installmethod("url", function(dictionary,word,n) local t = { } local p = nil for i=1,#word do local w = word[i] local s = specials[w] if s == "after" then s = { start = 1, length = 1, after = w, left = false, right = false, } specials[w] = s elseif s == "before" then s = { start = 1, length = 1, before = w, left = false, right = false, } specials[w] = s end if not s then s = false elseif w == p and w == "/" then t[i-1] = false end t[i] = s if packslashes then p = w end end return t end ) \stopluacode \stopbuffer \typebuffer \getbuffer Again we define a plug: \startbuffer \definehyphenationfeatures [url] [characters=all, alternative=url] \stopbuffer \typebuffer \getbuffer So, we only break a line after symbols. \startlinecorrection[blank] \starthyphenation[traditional] \tt \sethyphenationfeatures[url] \scale[width=\hsize]{\hyphenatedcoloredword{http://www.pragma-ade.nl}} \stophyphenation \stoplinecorrection A quick test can look as follows: \startbuffer \starthyphenation[traditional] \sethyphenationfeatures[url] \tt \dontcomplain \hsize 1mm http://www.pragma-ade.nl \stophyphenation \stopbuffer \typebuffer Or: \getbuffer \stopsection \startsection[title=Blocking ligatures] Yet another predefined feature is the ability to block a ligature. In traditional \TEX\ this can be done by putting a \type {{}} between the characters, although that effect can get lost when the text is manipulated. The natural way to do this in a \UNICODE\ environment is to use the special characters \type {zwj} and \type {zwnj}. We use the following example lines: \startbuffer[sample] supereffective \blank superef\zwnj fective \stopbuffer \typebuffer[sample] and define two featuresets: \startbuffer \definehyphenationfeatures [demo-1] [characters=\zwnj\zwj, joiners=yes] \definehyphenationfeatures [demo-2] [joiners=no] \stopbuffer \typebuffer \getbuffer We limit the width to 1mm and get: \startlinecorrection[blank] \bTABLE[option=stretch,offset=.5ex] \bTR \bTD \tx \type{method=default} \eTD \bTD \tx \type{method=traditional} \eTD \bTD \tx \type{method=traditional}\par \type{featureset=demo-1} \eTD \bTD \tx \type{method=traditional}\par \type{featureset=demo-2} \eTD \eTR \bTR \bTD \hsize 1mm \dontcomplain \starthyphenation[default] \getbuffer[sample] \stophyphenation \eTD \bTD \hsize 1mm \dontcomplain \starthyphenation[traditional] \getbuffer[sample] \stophyphenation \eTD \bTD \hsize 1mm \dontcomplain \starthyphenation[traditional] \sethyphenationfeatures[demo-1] \getbuffer[sample] \stophyphenation \eTD \bTD \hsize 1mm \dontcomplain \starthyphenation[traditional] \sethyphenationfeatures[demo-2] \getbuffer[sample] \stophyphenation \eTD \eTR \eTABLE \stoplinecorrection \stopsection \startsection[title=Special characters] The \type {characters} example can be used (to some extend) to do the same as the breakpoints mechanism (compounds). \startbuffer \definehyphenationfeatures [demo-3] [characters={()[]}] \stopbuffer \typebuffer \blank \getbuffer \blank \startbuffer[demo] \starthyphenation[traditional] \sethyphenationfeatures[demo-3] \dontcomplain \hsize 1mm we use (super)special(ized) patterns \stophyphenation \stopbuffer \typebuffer[demo] \blank \getbuffer[demo] \blank We can make this more clever by adding patterns: \startbuffer \registerhyphenationpattern[en][)9] \registerhyphenationpattern[en][9(] \stopbuffer \typebuffer \blank \getbuffer \blank This gives: \blank \getbuffer[demo] \blank A detailed trace shows that these patterns get applied: \starthyphenation[traditional] \ttx \showhyphenationtrace[en][(super)special(ized)] \stophyphenation \unregisterhyphenationpattern[en][)9] \unregisterhyphenationpattern[en][9(] The somewhat weird hyphens at the edges will in practice not show up because there is always one regular character there. \stopsection \startsection[title=Counting] There is not much you can do about patterns. It's a craft to make them and so they are shipped with the distribution. In order to hyphenate well, \TEX\ looks at some character properties. In \CONTEXT\ only the characters used in the patterns of a language get tagged as valid in a word. The following example illustrates that there can be corner cases. In fact, this example might render differently depending on the patterns available. First we define an extra language, based on French. \startbuffer \installlanguage[frf][default=fr,patterns=fr,factor=yes] \stopbuffer \typebuffer \getbuffer Here we set the \type {factor} parameter which tells the loader that it should look at the characters used in a special way: some count for none, and some count for more than one when determining the min values used to determine if and where hyphenation is to be applied. \startbuffer \startmixedcolumns[n=3,balance=yes] \hsize 1mm \dontcomplain \language[fr] aesop oedipus æsop œdipus \column \hsize 1mm \dontcomplain \language[frf] aesop oedipus æsop œdipus \column \startexceptions æ-sop \stopexceptions \hsize 1mm \dontcomplain \language[frf] aesop oedipus æsop œdipus \stopmixedcolumns \stopbuffer \typebuffer We get three (when writing this manual) different columns: \getbuffer The trick is in the \type {factor}: when set to \type {yes} an \type {æ} is counted as two characters. Combining marks count as zero but you will not find them being used as we already resolve them in an earlier stage. \startluacode context.startcolumns { n = 2 } context.starttabulate { "|Tc|c|c|l|" } for u, data in table.sortedhash(languages.hjcounts) do if data.category ~= "combining" then context.NC() context("%05U",u) context.NC() context("%c",u) context.NC() context(data.count) context.NC() context(data.category) context.NC() context.NR() end end context.stoptabulate() context.stopcolumns() \stopluacode It is very unlikely to find an \type {ffi} in the input and even an \type {ij} is rare. The \type {æ} is marked as character and the \type {œ} a ligatyure in \UNICODE. Maybe all the characters here are dubious but al least we provide a way to experiment with them. \stopsection \startsection[title=Tracing] Among the tracing options (low level trackers) there is one for pattern developers: \startbuffer \usemodule[s-languages-hyphenation] \startcomparepatterns[de,nl,en,fr] \input zapf \quad (\showcomparepatternslegend) \stopcomparepatterns \stopbuffer \typebuffer The different hyphenation points are shown with colored bars. Some valid points might not be shown because the font engine can collapse successive discretionaries. \getbuffer \stopsection \stopchapter \stopcomponent