*Slinkuiis a disallowed word, typical of its class, that on the surface seems perfectly permissible as a borrowing. It is certainly a resolvable predicate. It is certainly not either primitive or complex. But then one notices that if *slinkuiwere allowed in the language, a certain perfectly regular complex, namely paslinkui (which means 'ancestor', from 'past-linear-kin'), would not uniquely resolve. For if *slinkui were allowable, paslinklui would break up as the phrase pa *slinkui.
On the other hand, if we disallow *slinkui, and all words of its type, then paslinkui, and any word of its type, will not break up. The resolver would first obtain *slinkui; note that it was disallowed; look at the prequel; find pa; and then it would say, Right! Paslinkui must be it, then. That's the only thing /pasLINkui/ can be, since *slinkui is not allowed. (This, incidentally, is how Aslinkui-type words would also get resolved…if we allowed them. But these left-extensions of the basic Slinkui-form are disallowed for a slightly different reason which I will take up presently.)
The principle underlying the exclusion of Slinkui-type words from the domain of borrowings (some apparent Slinkui-words, by the way, e.g., smabru, are quite acceptable as complexes) is that complex predicates are so much more important in Loglan semantics than borrowings are—it is only the imitativeness of borrowings, after all, and not their number, that we intend to enhance by giving them a large word-space—that whenever there is a competition for a given portion of the word-space between these two varieties of predicate, it is the complexes that must always win out.
Thus, Slinkui-type borrowings (but not complexes) are not only excluded from the class of allowable borrowings, they are excluded from the language. That is why the "perfect complement set" I once thought we could achieve for borrowings is not perfect. It has a big empty swath cut right down the middle of it, namely the space occupied by all those resolvable predicates that are neither primitives nor complexes yet fail to pass the "Slinkui Test"; see below.
A second, similarly large emptiness in the domain of resolvable predicates is caused by our disallowing Aslinkui-type words as well. While these words are definitely resolvable once Slinkui's have been disallowed (but not before), the great variety of those possible leftward extensions of any Slinkui-form makes them difficult to see or hear. Indeed, they impose a "double double-take" on the resolver, a twisted loop in the resolution algorithm. So for these reasons—or this reason, if it is the same one—Aslinkui's too have been disallowed.
Let us finally consider the reasons behind the third test for allowable borrowings. We said it must contain no imitation hyphens, that is, no y- or r-involving sequences that imitate the hyphens of a complex. This exclusion is not always strictly necessary for resolution. It is not necessary, for example, when the word containing the apparent hyphen will not resolve as a complex. But in anticipation of the day when borrowings may also be involved in complexes—when some Innuit-speaking loglanist invents the word i'glymao for 'igloo-maker', for example—in which case hyphenated segments in a complex may be exceedingly irregular, let us now exclude all those imitation hyphens from "simple" borrowings that might confuse such constructions later. What this means is that no /y/, no /r/ in the context /CVVrC/, and no /n/ in /CVVnr/ is permitted in a borrowing.
2.58 The "Slinkui Test": Any CC-initial trial borrowing must be given this test; no other need be. It is simply performed. Put a test prefix, say pa, on your trial word. Let's say it's actually ?slinkui. Then try to resolve ?paslinkui as a complex; see Sec. 29. If the effort succeeds, as it does with pas/lin/kui, then the trial-word fails. If the term-resolving effort fails, as it would, for example, with pas/lin/kuti, then no coalescence with a preceding CV operator is possible; so the trial word passes.
Suppose your trial word is ?spe'a. ?Spea has a form that was once accepted as an allowable borrowing. But ?spe'a obviously fails the Slinkui Test. Paspe'a immediately resolves as complex and so *spea fails as a borrowing. [Thus a whole set of words, once accepted as legitimate borrowings—albeit in a narrower, modularly defined lineage—now vanishes from the domain of allowable borrowings. And so from the language.] But trial words that fail the Slinkui Test can usually be easily repaired. Any disturbance in the smooth structure of the imitation complex that takes over after the first letter—e.g., the linkui-part of *slinkui or the pea-part of *spea—will also suffice to make it pass the test. For example, *sli'nkui could be repaired as slie'nkui, slinkui'a, slini’kui, or, as noted, slinku'ti. Indeed, dozens of variations on each trial word are usually possible; one of them might imitate its natural model almost as well as the word that failed.
Some Slinkui-form sequences discovered by the resolver in the course of word-resolution are actually complexes. For example, if one presented ?sma'bru to the Slinkui Test, one would find out it was a Slinkui. No question of "failing" as a borrowing is involved here…merely the identification by the resolver of certain legitimate sequences as having Slinkui-form. There is more on this diagnostic use of the Slinkui Test in the next two sections.
2.59 The Resolution & Partial Classification of Predicates: Under the new morphology, the resolver proceeds very much as before, except that it must now deal with the possibility of vowel-initial predicates and it must perform the Slinkui Test on any CC-initial predicate-form sequence it encounters. Let us see why.
Suppose some such sequence "fails the test", that is, turns out to be a Slinkui. Since Slinkui-type borrowingsare not allowed, and therefore do not exist in any well-formed string the resolver will encounter, the discovery that it has found something that looks like one—that this particular CC-initial sequence is indeed of Slinkui-form—can only mean (1) that it's a complex (e.g., sma'bru), or (2) that the sequence is part of some larger predicate. Slinkui-form sequences can only exist as words if they are complexes; so if some sequences exist that are not complexes, they must be parts of words. The kind of word of which they are parts must obviously be predicates. Moreover, the part they are is a right-hand part since the right juncture of the Slinkui-sequence will already have been fixed; see next section. Therefore there must be a left-hand part out there somewhere.
What kind of a left-hand part could it be? Well, if weren't for the Aslinkui exclusion, it could be anything that could precede an impermissible initial consonant pair that turned up as the first CC of a predicate. For in truth, the sl of a *slinkui that is not complex cannot begin a word. Since Aslinkui-type borrowings have been excluded from the language as well (see above), there is only one allowable form for the left-hand part of that larger word, and that is the CV-form that will by definition turn any Slinkui into a complex. There will always be one, of course, or the non-complex Slinkui would not be there. So the resolver looks for the CV ahead of each such Slinkui; always finds one (in any well-formed utterance); and puts left-juncture to the left of that CV and hence the word.
The word that results from this manoeuvre will always be a well-formed complex; the Slinkui Test will have seen to that. In a sense, the resolver short-circuits the normal recognition procedure in these Slinkui-cases. For normally the recognition of a complex is subsequent to its resolution as a predicate.
2.60 The Predicate Resolution: The new algorithm for resolving predicate words is in some ways simpler than the 1975 one. The resolution of a predicate still commences, of course, with the discovery of a CC in the left-to-right scan of some breathgroup. That first CC—there may be others, of course—may be either a permissible initial (sp) or impermissible one (rk). If the FIRST CC is IMPERMISSIBLE, the algorithm records that fact as a prevailing condition and then performs the following steps:-
Is there a stressed V' or vv’ immediately ahead of the first CC? If so (e.g., i'sp, ia'rk), mark it and record STRESS FIRST as a prevailing condition. If not, find the first instance of a V' or vv' that follows the first CC (e.g. spe', spudei', rida', rkoti') and mark it.
Find the first instance of a V or vv after the marked V' or vv' (i'spai, ia'rko, spe'a, spudei'tai, rkia'mpa, etc.). Put right juncture to the right of that V or vv (i'spai|, ia'rko|, spe'a|, spudei'tai|, rkia'mpa|, etc.).
If the condition STRESS FIRST and/or FIRST CC IMPERMISSIBLE obtains, find the prequel by locating the first instance of a C or /./ to the left of the first CC (.aoi'spai|, tia'rko|, morkia'mpa|). Put left-juncture at that /./ or before that C (|aoi'spai|, |tia'rko|, |morkia'mpa|) and exit.
Perform the Slinkui Test on any sequence that remains. (Stress will follow its first CC, and it will be a permissible initial.) If it is a Slinkui (e.g., spe'a|, spa'rgu|, sma’bru|), send it to the term resolver to discover whether it resolves as a complex. If its terms do resolve (e.g., sma'/bru|), put left-juncture immediately to its left (|sma'bru|)). If it doesn't resolve as a complex, find the CV that will always be to its left and put left-juncture to the left of that CV (|tospe'a|, |mispa'rgu|). In either case, tag it as a recognized complex and exit. If it is not a Slinkui (e.g., spudei'tai|, spu'ta|), put left juncture immediately ahead of the first CC (|spudei'tai|, |spu'ta|) and exit.
Clearly term-resolution is involved in at least some cases of word-resolution. We must now consider how the resolver will "parse" the terms of a complex when the predicate is one.
2.61 Term-Resolution: Once a predicate has resolved as a word, the recognition of its primitiveness, if it is primitive, is trivial: it is either of matma- or brudi-form, or it is not. If the resolved word is not primitive, or if a Slinkui-form sequence has been identified by the resolver, an attempt is made to parse the word or sequence as a complex, that is to say, to resolve its terms if it has terms. If it has no hyphenation sequences and is 0 mod 3, then, if it is complex, its terms will just "fall out" in 3's, and each triple will be a legitimate affix: sma/bru, nil/boi, tar/ses/mao, roj/mad/ses/mao, and so on. If they don't or aren't (e.g., tai/rko, mor/kia/mpa), it is not complex. If it has no hyphens and is not 0 mod 3, then it will have a 5-letter final affix if it is complex; and the rest will fall out in 3's. If they don't and/or there is no long final affix, it is not complex. If it has hyphens or hyphen-like sequences, e.g., /y/, /CVVrC/ or /CVVnr/, it is complex. (Borrowings, remember, may contain no imitation hyphens.) So the algorithm then examines the regions between the hyphens, and between any hyphen and the ends of the word, and the lengths of these regions, together with their positions in the word, will always uniquely determine what lengths of affixes will be found in them if the complex is well-formed. (It will also identity whatever irregularly formed sequences it finds among the affixes as derived from borrowings, when this development is upon us.) In this way, every predicate word that is a well-formed complex will uniquely resolve into its terms; and every word that resolves into terms will be a well-formed complex.
2.62 The Recognition of Borrowings: The recognition of a borrowing thus always follows on the failure to hear or see it as a complex once it has resolved as a word. Thus, one knows one has a borrowing only by failing to resolve it as a complex. Thus tia'cro is complex because its two affixes, tia and kro, are well-formed. Tia'rko, morkia'mpa and fainstoi'a are not complex because their partition as complexes fails. Rko and mpa are not well-formed CCV-affixes; and fainstoi'a, while 0 mod 3 and hyphenless, does not fall out into legitimate 3's: fai/nst/oia. And so all these words are borrowings.
The construction of a borrowing thus often proceeds by putting impermissible initials (rk mp ns) in judicious places, or by making sure that its affix-length pieces do not have affix-form shapes (nst oia). Thus, we may say that whatever is an allowable predicate word and not a primitive or a complex is a borrowing.
Observe, by the way, the operation of the pair-from-the-left rule of Sec. 1.15 in determining the character of the stressed syllable of fainstoi'a: /fainSTOIa/ (fighn-STOY-aa), which is quite a different sound from (fighn-STOH-yaa).
2.63 Making Borrowed Predicates: The procedure for making a predicate based on some natural language model involves four distinct steps:
Build a preliminary trial word that imitates the model word as well as possible…either its sounds or its letters, or rarely, both. When the model predicate is a scientific word, the word-maker may be guided by the transcription system given in Sec. 2.14 for pronouncing Linnaean names. Thus the predicate 'cephalopod'—which is either derived from or the source of the Linnaean name Cephalopoda—might well be initially transcribed as ?cefalopo'd; for this is our recommended pronunciation, and hence a rewriting in Loglan phonemes, of the portion it shares with the related Linnaean name. Even if there is no related Linnaean term, transcribing it as if it were Linnaean is a good procedure to follow when borrowing scientific words.
Make sure the trial word is a resolvable Loglan predicate, repairing it as necessary to give it the necessary properties; see Sec. 2.38. ?cefalopo'd (sheh-faa-loh-POHD) lacks two of these properties. It is not V-final and it has no CC. Let us repair the first defect by giving it the final -a of the corresponding Linnaean name, and the second by inserting r before the second C. /rC/ is always an impermissible initial, and having one sufficiently early in the word is crucial for preventing the first C and its following one or more V's from "falling off the word". ?cerfalopo'da (shehr-faa-loh-POHD-aa) is the word that results from these moves.
Make sure that the developing word is not primitive or complex, and that it doesn't resolve as a phrase. This means checking its form against the two forms reserved for primitives—which require a more cautious procedure of derivation than you may wish to get involved with here—and attempting to resolve it first as a phrase, and when that fails, as a string of affixes. Both attempts fail with ?cerfalopo'da. Nothing falls off; and term resolution starts out with cer/fal/.. but then encounters a sequence which is impossible in a complex, namely …opoda. So the word passes the third test.
Check it for false hyphens. ?cerfalopo'da has no /y/s, no /CVVnr/s, and the /cerf/ sequence just misses being an instance of /CVVrC/; so the developing word passes this test as well.
Test it for Slinkui Failure and the Aslinkui Condition. The first test is relevant only if the trial word commences with CC, as this one doesn't. The second condition can exist only in a trial word whose first CC is non-initial, is a permissible initial, comes before the stress, and has a prequel that resolves as words. Only in such words will the prequel be loosely-attached. The first CC, namely /rf/, is not a permissible initial; so the trial word avoids the Aslinkui Condition as well.
Cerfalopo'da passes all tests and is thus an allowable borrowing. Eventually, for at least the borrowed portion of the Loglan scientific vocabulary, this word-borrowing procedure can easily be made algorithmic. The borrowing algorithm is under development now. But much of the Loglan scientific vocabulary will not be borrowed. Large portions of it may be captured Chinese-style by local metaphor. For example, rather than borrow the sounds of 'xenophobe' one might prefer to "borrow" its idea, the idea of the "stranger-fearing" person. That would produce a scientific complex that would be decipherable by any Loglan reader, not just those learned in the vocabulary of Western science. If this were our policy, gu'rfia might well become the word for 'xenophobe/-ic' in Loglan. It is made from gutra firpa = 'stranger-fear' and is thus a literal translation from the Greek. Whence lopo gu'rfia would be 'xenophobia', and so on. Still, the borrowing zernofo'bi—made by much the same procedure as cerfalopo'da was made above—is also possible, and has the advantage of being immediately recognizable to those who know the psychiatric vocabulary. But even in English—one might say especially in English—such Graeco-Latin words exclude from understanding precisely those whose educations have not permitted them to know.
To borrow or not to borrow. Lehnwort vs. lehn übersetzung. There are many good arguments on both sides of this question. As yet, The Institute has taken no position. We prefer to wait until the loglaphone community has extended itself internationally—and also intellectually, one would hope—beyond the narrow confines of anglophone computer science and those related Western disciplines of logic, linguistics and anthropology where interest in Loglan began some decades ago.
CHAPTER 3 LEXICON (WORDS & SPEECH PARTS)
3.1. Definitions and Conventions
: In this chapter" the vocabulary of Loglan is partitioned into its 69 "parts of speech", or sets of grammatically interchangeable words. These sets are called lexemes, and we will consider them one at a time in the alphabetic order of their simplest or most representative members. These representative members, written in upper case letters, serve as lexeme names. We commence with Lexeme A, the Afterthought Connectives, and end with Lexeme ZO, the Quantity Abstractor.
If a lexeme has more than one member, these are called its allolexes. The allolexes of a lexeme are its equally permissible alternative expressions; they are the "interchangeable elements" of which that lexeme is composed. Lexemes which have only one member are called monolexic. In the section devoted to each lexeme, our object will be to show how its various allolexes are formed morphologically, how it is used by the grammar, and the range of meanings of its allolexes. Whenever possible, the list of allolexes is complete.
The allolexes of all the lexemes of a language are the lexes of that language. They are the words or phrases which, like English 'nevertheless', the grammar treats as a single word. In Loglan all lexes are words; that is, they are spoken pauselessly and written without internal spaces, The lexer is that part of the human central system, or of some computer program designed to process language, which identifies lexes as such, and which assigns them to their lexemes. In the system of computer programs that constitute the machine grammar of Loglan, the lexer is part of the preparser subsystem.
From the point of view of the grammar, each allolex of a non-monolexic lexeme is an equally legitimate occupant of whatever place any of them occupies in any utterance. Thus it makes no difference to the grammar which allolex of a lexeme a speaker chooses; for the grammatical structure of an utterance does not change when one allolex of a lexeme replaces another. So in a certain sense, if a learner has learned how to use one allolex of a lexeme, da has learned how to use them all. For they all have the same grammatical privileges. Thus, there are only semantical differences between allolexes of the same lexeme, not grammatical ones. It is in that sense that allolexes of the same lexeme are grammatically interchangeable.
The lexemes of Loglan are exclusive. That is, if a lex is a member of a certain lexeme, then it is a member of no other. This is not true of English, where the word 'bank', for example, belongs to at least three lexemes, each with distinct grammatical roles. The uniqueness of lexemic assignment in Loglan removes a major source of linguistic ambiguity from the language. This is the lexemic ambiguity ('They are flying planes') which seems to be both massive and universal in natural languages.
Lexemes whose names are 'M' followed by a numeral are called M-lexemes or machine lexemes. These and certain other machine-oriented lexemes do not appear in human Loglan and are visible and audible only to the machine. The human user of the grammar need not concern daself with them. The role of machine lexemes in the machine grammar is explained in the next chapter. The names of lexemes used only by the machine are *-ed in the section headings. A total of 17 lexemes are occasioned only by the machine's needs. So from the human user's point of view, there are only 52 lexemes in Loglan.
Nine lexemes, or about one out of six in the human lexicon, differentiate connective words from one another. This is an extraordinarily high proportion compared to any natural language. As a logical language, Loglan makes great use of the principle of connectivity, the principle by which the truth-values of several or many sentences may be related to one another in a single sentence.
The PREDA lexeme is the largest lexeme in the lexicon, containing about 80% of all dictionary entries. PREDA's are all the predicates of the language, that is, its noun-, verb- or adjective-like words as well as many of its adverbs and prepositions.
Many allolexes of the non-PREDA lexemes are compound structure words. The compounding formulas which show how these lexes are generated are composed of lexeme names, the sign of alternation [/], of concatenation [+], and of identity [=]. Thus 'NI/TAI + FI = UI' means that members of the NI (nee) or TA1 (tigh) lexemes may be joined together with members of the FI (fee) lexeme to produce compound structure words which will be lexed as members of the UI (wee) lexeme.
The words 'operand', 'modificand' and 'connectand' are originally Latin words meaning 'that which is operated upon', 'that which is modified' and 'that which is connected', respectively; they are used freely throughout the text.
'R' followed by a numeral, e.g., 'R12', refers to a particular rule of grammar that will be found under that number in the next chapter.
The parse of an utterance is a specification of how the grammar generates or understands it. A parse may be partially shown by a full parenthesization of the utterance once it is stripped of any other punctuation. The nested parentheses show the order in which the lexes in the utterance are to be grouped. Thus the parse of Da, a de, e di = 'X or Y and Z' is partially given by ((da a de) e di). This shows that the expression is to be understood by first grouping da a de together, and then grouping the result with di. Note that this structure is independent of the semantic values of a and e, which are in fact allolexes of the same lexeme (Lexeme A, the first lexeme in the lexicon). In this chapter, the production sign [=>] will sometimes mean 'is parsed as'.
Reference will be made throughout this chapter to various grammatical structures whose origins and functions will not be completely understood until they are studied in the grammar. In general, the lexicon acquires its structure from the grammar while the grammar acquires its semantic variety from the lexicon. In truth they are inseparable; neither can be deeply studied without reference to the other.
Lexeme A: Afterthought Connectives (Eks)
These are the afterthought connectives used between predicates, arguments, linked arguments, argument modifiers, and sentence modifiers. They are said to be "afterthought" because they are left-grouping, e.g., da, a de, e di => ((da a de) e di). So connectands may always be added to an ekked string without disturbing previously understood meanings.
When used between predicates, eks are prefixed with M11 by the preparser; when connecting linked arguments, the preparser inserts M1; when connecting argument modifiers, M6; and when connecting sentence modifiers—called simply "modifiers" in the grammar—M5. Only when used to connect arguments are eks unmarked.
The complete list of simple eks is:
…if and only if…
…or…but not both
In addition, any member of the PA-Lexeme may be appended to an ek, producing, for example, tensed, located or even "motivated" connectives: efa = 'and then'; evi = 'and here (at this place)'; emoi = 'and with this intention'; and so on. All such compounds are treated grammatically as members of A. Not many of these A + PA compounds have been explored, however; their semantic domain seems to transcend any found in the natural languages. The loglanist is invited to explore this new country. See Lexeme PA for a list of the PA components of A + PA compounds.
Lexeme ACI: Hyphenating Eks
These are any member of the A-Lexeme suffixed by -ci: thus aci, eci, apaci, noanoici, and so on, are all hyphenating eks. ACI connectives have an effect analogous to that of hyphen ci in a predicate string (see Lexeme CI); that is, they consolidate the two adjacent elements into a single connectand. At the moment, ACI is confined grammatically to ekking arguments and predicates; among the latter, the preparser will have prefixed M9 to the ACI word. The use of hyphenating eks could, of course, be extended to other ekked structures in the grammar, for example, to ekked modifiers. It would be grammatically costly to do so, however, since introducing ACI to an ekking structure usually involves the addition of 2 to 4 rules, and a new M-lexeme will always be required for every structure so accommodated. So for the present, and until a use for them in other contexts can be demonstrated, hyphenating eks are confined to connecting the two main ingredients of a sentence, arguments and predicates.
Lexeme AGE: Right-Grouping Eks
These are any member of the A-Lexeme suffixed by -ge: thus age, ege, apage, noanoige, and so on, are all right-grouping eks. AGE connectives have an effect analogous to the effect of the grouping operator ge in a predicate string; that is, they consolidate the entire right portion of a string of ekked elements into a single connectand. At the moment AGE words are confined grammatically to ekking arguments and predicates; among the latter, the preparser will have prefixed M11 to the AGE word. The use of right-grouping eks could, of course, be extended to other ekked structures in the grammar, for example, to ekked modifiers. It would be grammatically costly to do so, however, since introducing AGE to a new ekking structure usually involves the addition of 2 to 4 rules, and always involves the addition of another M- lexeme. So for the present, and until a use for them in other contexts can be demonstrated, right-grouping eks are confined to arguments and predicates.
Lexeme BI: Identity Operators
These are the "little word predicates" which must be kept grammatically separate from the PREDA-Lexeme because their compounds are recognized by the preparser. If it weren't for this morphological function, BI and kin would be members of PREDA. The current list of BI words is
is equal/identical to
is a member of
is part of
is less than
is greater than
and the compounds
is less than or equal to
is greater than or equal to
Other compounds may be made by prefixing no- or nu- to any of these with obvious meanings. Thus nocio makes the same claim as ciebi. No doubt mathematicians will have other allolexes to add to BI.
BI is an open lexeme; new members may added at any time.
This is special lexeme used by the lexer as a bin for words it doesn't recognize. Human users may also have such a bin for unlexed words; but unlike the machine, humans usually guess what lexeme unknown words belong to.
Lexeme CA: Predicate Word Connectives (Sheks)
These are the connectives used to join individual predicate words in an afterthought, i.e., left-grouped, mode. Morphologically, each shek is the result of inserting c before the characteristic vowel of an ek (see Lexeme A). The current list of sheks is