As may be seen from the references, gu alone is required in only two rules, R121-2; gue is used in similarly few places, R22-3; but the triple option presented by , which may be executed by a gu, pause/comma or frequently by nothing at all, is extremely widely used. In fact is the most widely-used grameme in Loglan grammar. It provides an opportunity to mark the right boundary of phrases or clauses whenever such a boundary would be unclear without it. Loglanists tend to use pauses and commas at such points when dealing with human readers and interlocutors in conditions of low noise, and to use gu's in high noise or when addressing computers or other unforgiving auditors. See the numerous rules listed above for the occasions on which these punctuators may be used.
Optional punctuators are not the only "punctuation marks" in Loglan. There are also six punctuators which are essential to the structures they mark and so may never be omitted. These are the "grouping operator" ge, which may be found in R22-3 along with the optional gue; the "inversion operator" go, found in R48; goa, which marks the V-O-S word-order in R155; and the "fronting operators" gi and goi, found in R165-66. In addition, there is a punctuator-like member of the PA-Lexeme, ga, which is used exclusively for marking the left boundaries of otherwise unmarked predicate expressions when these would be absorbed by just-preceding descriptions; see R139. For example, ga in Le mrenu ga sadji = 'The man is wise' prevents Le mrenu sadji = 'The man (sort of) wise one' from being heard. would work here—for example, *Le mrenu, sadji also parses in the required way—but is regarded as bad usage (and so is *-ed) because there are some kinds of descriptive arguments after which fails to produce the required separation. Gaalways works and so is preferred.
This is the entire punctuation system of Loglan. See the rules cited for the details.
Group B. Linked Arguments
Je and jue are the two preposition-like words that attach strings of one or more arguments to predicate words. Sometimes the predicate word involved is buried in a predicate expression (Da kukra je lo litla, grobou = 'It's a faster-than-light ship'); sometimes it is the last word in a description (Le selrispe farfu je la Djek = 'The proud father of Jack'). In either case, the linking words have the effect of binding a string of elements into a single unit. Thus if the je and jue of S8 were removed, the string remaining would be composed of three distinct arguments: To ketpi da de = 'Two tickets, X, Y'. The two linking operators bind this triad into a single argument: 'Two tickets to destination X from point-of-departure Y'. The present group of rules shows how to construct the right part of such expressions, the "linked arguments". The predicate units, like ketpi, to which they are attached are constructed in the next group; see R26.
=> JUE argument
To ketpi je da jue de = Two tickets to X from Y.
Jue de = From Y.
=> juelink linksl gap
Jue de jue di = From Y on W.
Jue de = From Y.
=> links M1 A linksl
Jue de, a jue di = From Y or from W. (The pause after de is a "morphemic pause", necessary for the resolution of a.)
When links are connected, as they are for example in S18, the linking words need not be repeated. Thus, Loglan Je da, e de conveys the same notion as S18 does just as English 'To X and Y' conveys the same notion as 'To X and to Y'. Sometimes in both languages, however, this degree of explicitness is desired.
Links and linkargs are used in only three non-local rules. Links are used in R172, and linkargs in R26 and R173. In R172-3 both links and linkargs appear as fragmentary utterances, such as might be answers to Jue hu and Je hu ('To/from/by whom?') questions. Linked arguments enter the main stream of the grammar at just one place, however, namely in R26 of the next group, where they are attached to predicate words.
Notice that two machine lexemes have been introduced in Group B. They are M1 which marks the "eks" (the a-form afterthought connectives) which connect both links and linkargs; and M2 which marks the "keks" (the ka-form forethought connectives) which connect these objects. The reader will recall that machine lexemes are put in place by the preparser, one of whose tasks is to extend the limited 1-element lookahead of the machine. Human brains do not need these warnings since our capacity to remember and inspect long strings is so much greater than that of any parsing algorithm used currently by machines.
Group C. Predicate Units
Predicate units, or "predunits" as we will sometimes call them, are either single predicate words, with or without certain inflecting operators such as nu, no or po; or they are strings of such possibly inflected predicate words which have been made into a predicate unit by ge, or by a ge matched with a subsequent gue (R22-3); or they are arguments that have been turned into predunits by the "predification" operator me (R24); or they are predicate words which have been augmented by the attachment of one or more linked arguments (R26).
Ba sucmi = Something swims.
=> NU PREDA
Ba nu sucmi = Something is swum to.
=> GE despredE gue
Da briga ge musmu janto = X is brave for a mouse hunter. (The is not activated here; it would be redundant if it were.)
Ge is the grouping operator; it is only meaningful if it precedes a string of two or more predunits. Thus *briga ge musmu, while grammatical, is proscribed as bad usage, for it can mean nothing other than what briga musmu ('brave mouse') already means. Again, redundant marks are avoided. , which is the operand of ge, is made in the next group; see R45-6.
=> NU GE despredE gue
Da nu ge briga janto = X is a quarry of brave hunters. (Nu is one of the conversion operators; and ge extends the scope of nu over the entire string, in this case briga janto, which then takes its place-structure from its last term, in this case janto. Thus X is a kind of quarry.)
=> ME argument gap
Ba mela Ford = Something is a Ford. (Me turns any argument into a predicate unit with a meaning associated with that argument.)
Da kukra grobou = X is a fast ship.
=> predunitl linkargs
Da kukra je lo litla, grobou = X is a faster-than-light ship.
R26 is the primary use of the linked arguments made in the preceding group. Their use in descriptions also passes through this rule. Usage imposes a certain restraint on the use of R26, however. The
to which linkargs is attached by it must not, by convention, be the last unit in a string unless that string is a descriptive predicate, i.e., one used in making arguments. Thus Le mutce kukra je lo litla = 'The thing which is very much faster than light' is an acceptable use of linked arguments. In fact the link word is necessary if this description is to be heard as one argument rather than two. The link word is also necessary in Da kukra je lo litla, grobou in which the argument is linked to a non-final predicate unit in a string. (We will call this the "internal specification" of a predicate.) But arguments are not allowed to be linked to final units in predicate strings when these are being used as sentence predicates. Thus while *Da mutce kukra je lo litla (presumably to render 'X is very much faster than light') is perfectly grammatical, it is a proscribed usage because there exists a preposition-free form that does the same job more elegantly: Da mutce kukra lo litla. Again, usage in Loglan avoids redundant markings, such as this quite unnecessary je. The same link words are not of course redundant inside predicate strings, or to link a descriptive argument to other arguments. But they are redundant if used to link a sentence predicate to its argument set, or the arguments in that set to one another.
This is the first instance we have encountered of the "grammatical superset" phenomenon. By this I mean that we have written a rule of grammar that generates a domain of grammatical utterances which is larger than the domain of "good utterances" allowed by usage. Another way of saying this is that not all grammatical utterances are acceptable as good usage. Any economically written set of grammar rules will occasionally generate such effects. They are perfectly harmless. We need only add a set of usage rules to such a grammar to ensure that any grammatical utterance that meets the usage rules will also be interpretable. We have at present no way of interpreting the difference between *Da kukra je de and Da kukra de; so we do not use the former. There are other supersets in Loglan grammar; I will call the reader's attention to them as we encounter them.
Da sadja = X is wise.
=> NO predunit3
Da no bimbo = X is no fool. (This is short-scope negation; it applies to a single predunit. Long-scope negation is accomplished elsewhere.)
Ti nigro = This is black.
=> predunit4 ZE predunit3
Ti nigro ze babe = This is black-and-white (mixed). (Ze is a special connective which has the sense of mixing properties.)
Ti blanu = This is blue.
=> PO predunit4
Ti po blanu = This is a state of being blue. (This is short-scope abstraction; other uses of po take whole sentences as operands; see R130-1.)
=> ZO predunit4
34, 36, 49-50.
Ti zo blanu = This is an amount by which something is blue.
Zo is kept out of the PO-Lexeme only because the preparser needs to recognize quasi-ZO in lexing acronyms. This is because -z- is the acronymic hyphen, and when it precedes /o/, as it does in CaiIzO [CIO], the result looks like a compound of TAI + I + ZO to the preparser. If zo were part of PO, it could not lex this acronym correctly. For this entirely mechanical reason—which probably has no meaning for human lexers—PO and ZO must be kept lexemically separate.
Predunits are the building blocks with which "descriptive predicates" (predicates used as "nouns") and "sentence predicates" (predicates used as "verbs") are made. These will be constructed in the next two groups. No machine lexemes have been necessary in the construction of predicate units.
Group D. Descriptive Predicates
These are the predicate expressions which, when used as operands of any of the descriptive operators—le and kin, for example, or the name operator la, or any quantifier—create that special kind of argument known as nouns or noun-phrases in Indo-European grammar. We call such arguments "descriptions" because they employ a predicate expression to "describe" some feature of the designated thing.
The following sequence of rules differs from that in the next group, in which sentence predicates are made, primarily in one feature: descriptive predicates may have kekked head units, e.g., the ke forli ki sadji part of S35; sentence predicates may not. It turns out that ambiguities are generated if sentence predicates are permitted to have kekked head units, while the same constructions are quite unambiguous in descriptions. So kekked head units are provided in this rule group—the of R35 and 37-8—and absent from the next.
Le denro simba ci janto = The dangerous lion-hunter. (Without ci the predunits in the string would group left, thus 'The dangerous-lion hunter'.)
=> M3 KA descpred KI despredA
Le ke forli ki sadji = The one who is both strong and wise.
=> NO kekpredunit
Le no ke forli ki sadji = The one who is not both strong andwise.
Another machine lexeme is introduced here, M3 which extends the parser's lookahead over KA again, this time to announce to the parser that a predicate is being kekked. Unmarked keks connect arguments.
Le sadji = The wise one.
=> CUI despredC CA despredB
Le mutce cui fizdi forli ce sadji = The one who is very physically-strong and (very) wise.
Cui is the left boundary mark of left connectands formed of more than one predunit. Without cui the single units on either side of the "shek" (the CA-connective) are taken as its connectands. No similar mark for right connectands is necessary.
This is the rule by which predicate strings are generated. Because it is left-recursive, all unmarked predicate strings are left-grouping.
Le sadji = The wise one.
=> despredD CA despredB
Le forli canoi sadji = The strong if wise one.
Le mrenu = The man.
=> despredE despredD
Le forli canoi sadji mrenu = The strong if wise man.
Le troku hasfa = The stone house.
=> despredE GO descpred
58, 72-4, 76, 97
Le hasfa go troku = The house of stone.
Notice that several grammar rules in this group besides the last one are non-local. All these other non-local rules are employed at similar points in the next group, where sentence predicates are made. But as all descriptive predicates may have kekked head units, care is taken to ensure that these structures borrowed from the descriptive sequence do not end up as the heads of sentence predicates; for there they would generate ambiguities.
The descriptive predicate grameme itself, , is used as the right part of R58, the final rule in the construction of sentence predicates, and at various points (R72-4, 76 and 97) in the construction of arguments, which are made in Group G.
This group of grammar rules is structurally parallel to the preceding one, the only difference being the one already noted, namely that sentence predicates may not have kekked head units. Note that some "despred" gramemes figure in these "senpred" rules, but that they are never initial in an allogram; this renders them harmless. Thus mutce ke briga ki ckano is a permissible sentence predicate while ke briga ki ckano mrenu, with its kekked head unit, is not. The reason for this proscription is not hard to find. If ke briga ki ckano mrenu were permitted as a sentence predicate, as in Da ke briga ki ckano mrenu, presumably to mean 'X is both a brave and a kind man', then the parser could not distinguish between this use of keks to connect predicate units—it is briga and ckano that are being connected here and not briga and ckano mrenu—and the later use of the same keks in Group I to connect whole predicate expressions. This would generate an ambiguity between the two rules, for they would both be capable of generating the same string. So the auditor could not tell whether the speaker meant his keks to connect just preddunits, as above, or whole predicate strings, as in Group I. The latter possibility would give an entirely different parse tree, one with the interpretation 'X is both a brave person and a kind man'. As things stand, the second interpretation is the only legitimate one…precisely because kekked head units are not allowed in sentence predicates.
Da mrenu = X is a man.
=> predunit CI senpred1
Da simba ci janto mrenu = X is a lion-hunting man.
Da sadji = X is wise.
=> CUI despredC CA despredB
Da grada cui simba janto ce sadji = X is a great lion-hunter and (a great) wise one.
Da prano = X runs.
=> senpred3 CA despredB
Da prano a fleti = X runs and/or flies.
Da mrenu = X is a man.
=> senpred4 despredD
Da simba mrenu = X is a lion man.
Da simba fumna = X is a lion woman.
=> senpred4 GO descpred
Da fumna go simba = X is a woman of lions.
All the gramemes in this tiny group are local except the last one, and that one, , figures in just one later rule: R128 in Group I, which makes the objects called simply "predicates". It is in this later group that the sentence predicates made here will be supplied with their "termsets": the strings of modifiers and/or arguments that complete them as predicates. Before that can be done the termsets must be constructed. The modifiers and arguments of which termsets are composed will be prepared in the next two groups.
The making of sentence predicates has involved no machine lexemes.