Saturday, October 24, 2015

The Role of Errors in the History of Loglans 6

Maxim Six: Loglan can be used as an interlanguage.

This is never listed as a goal or virtue of Loglans, but it has been a theme for many years (I suggested it around 1961 and was almost certainly not the first) and has been growing in importance in recent years.  It requires some explanation.  First, although the usual word for interlanguage is "interlingua" and that, in turn, is the name of several international auxiliary languages, this is not about a Loglan as an IAL.  Indeed, even though various users of Loglans have suggested this as a goal and several feature and maxims have made it a plausible goal (cultural neutrality, vocabulary drawn from a wider range of cultures), the official organs of Loglan and Lojban have repeated denied such an aim.  The syntactic and semantic peculiarities if Loglans has probably reinforced this position.  We will, however come back to this question from another direction later.

Rather, a interlanguage is a concept from the work of handling language data on computers.  It is a language which lays out the data supplied by various natural languages in a format more readily accessible to various manipulations: translation, abstraction, cross-referencing and the like.  The original notion come from machine translation, where it was seen first as an economy: translation between two languages require two translation programs, among three, six, four, 12 and so on, n(n-1) for n languages.  But, if there were a single language into which one could translate from any language and out of which one could translate to any language, with accuracy comparable to the one-to-one translations, then the only 2n programs would be needed.  If the new language were of a particularly simple, transparent, structure, without the details of various natural languages, the programs for a given language might be significantly simpler than any one-to-one program, which has to deal with the particularities of two languages at once,  And so, the scheme's disadvantages in number for a small group of languages might be overcome by each translation program being markedly simpler.

At the translation level, which tends to be the beginning of all processing of language data other than statistics about a particular language, the idea is that the text in one language comes in and is translated into the interlanguage.  From there, in the translation phase, version in the other languages in the system can be run off as needed.  But the interlanguage text can be dealt with in a variety of other ways: an abstract can be prepared, consequents can be inferred, categorization across a variety of fields can be made.  And so on with all the things that get done with texts.  These other uses of the text help to guide the selection of features for the interlanguage, keeping it as simple as possible yet as useful as needed.

Clearly, because of consequents drawing, formal logic is going to play a role in the mix.  But it is also there because the interlanguage version of the text sentence is supposed to show the meaning of that sentence and this is readily taken to be a formula of the currently favored logical system.  Beyond this there are clearly needs for various kinds of rhetorical information, pragmatic markers, connotation clues and the like to make translation (and the rest) accurate.  Some of these, however, may already be in some of the more advanced logical systems.  The vocabulary of the language has to be fairly fluid to allow that word-word translation does not always work well, that one language's "blue" may be another's "green" and many much more complex problems of being outside -- or at least marking separately -- all the various cultural features that might be involved  Even anaphora needs some special consideration. The sciences seem to have fewer of these sorts of problems and so have been the first areas to be dealt with along these lines.

And, of course, once you start to talk about a language into which you translate everything and from which all can be derived, someone made the leap to:  Why don't we all, at the occasions where this sort of things is to be done, just talk the interlanguage, thus eliminating at least one step -- and, for most of us actually involved, all the steps.  We talk interlanguage and so don't need the translations, though they are available for those who don't speak interlanguage (what are they doing here any how?).  There is some evidence that thoughts along this line -- as well as an ancient tradition of doing scholarship in a common language and a fondness for IALs among intellectuals -- played a role in the development of some IALs designed specifically for scientific conferences.

At every point in the above description of an interlanguage, the various maxims about the Loglans are clearly echoed: formal logic, cultural independence, freedom from ambiguity (not mentioned above but clearly necessary),   And it is already spoken by a significant number of speakers. So, a Loglan is a natural for an interlanguage or, at the worst, is a long step toward its development, needing only minor refinements to be totally adequate.  Such refinements would clearly be needed if the questions raised here about the accuracy of the various claims are borne out.  But even then this is clearly a better starting point than any natural language, since it is fully described and comes close to most of the goals.  So, let us begin with a Loglan and tweak as the best strategy to proceed.

The nearly fatal flaw with this plan is that the interlanguage is not a language in any ordinary sense. It has no phonology nor morphology, it doesn't have a syntax to lay out the linear order of words (and no words, for that matter, nor linear order).  It is a data structuring device, in which the meaning of sentences are given a conceptual form.  It is at least two-dimensional and often many more.  It involves detached and discontinuous elements.  And much more.  Now, obviously, all of these features are represented more or less well in ordinary linear languages, since that is whence they come into an interlanguage. But mainly less well, given the linear nature of language.  Even a very good Loglan -- indeed, even formal logic itself -- is always 1) going to fall short of accuracy and 2) involve a mass of irrelevancies of phonology, morphology and syntax.  So, in no way, can it substitute for an interlanguage.

But this does not mean that Loglans are useless in creating an interlanguage.  Lying, as they claim to do, somewhere between  formal logic and ordinary speech, the deviations that they make from formal logic toward speech may be as significant as the deviations the other way.  And the way these moves fail may be even more revealing.  Consider one simple case, modification pairs, such as those displayed in the iconic "pretty little girls school", here understood in the basic right-grouping sense.  Formal logic has traditionally taken constructions as simple conjunction "this is a school and for girls and little and pretty", but this is notoriously inaccurate, as "small galaxy" immediately shows.  The Loglans keep these constructions from speech, thus indicating the necessity of keeping a visible relationship between the parts.  However, they fail in that they take all such pairs as the same, where as the relationships involved may be very different for different pairs,  Even formal logic acknowledges some differences: notice the 'x is for girls" not merely 'x is girls" in the expansion above (less clear if the original were "girls'").  Clearly, the relation between "girls" and "school" is different from that between "little" and "girls school" and that difference needs to appear in the interlanguage (and, indeed, in any deep representation of the meaning of an expression). (Obviously, Loglans, like ordinary languages, can express all these difference, but, like both ordinary languages and logic, they do not do so at the most basic level, but only in elaborate periphrases.)

How would this phrase look (in a very schematic way) in an interlanguage?  It begins with an entity item attached by an instantiation connection to a nexus loosely described externally as SCHOOL, which contains threads for buildings, campuses, institutions, as well as education (and many more that are not relevant here but might be activated in different circumstances).  The addition of "girls" as a modifier activates a series of connections which gets to "student body" and this is now linked by inclusion to GIRLS.  Adding "little" looks for sized components and finds, again, building, campus and student body and the size of each of these is compared with that of the "average" SCHOOL in these areas and declared to be significantly less. (I won't go into "pretty" which behaves more or less like "little" but does pretty much eliminate institutions from the relevant connections for SCHOOL.)   The whole of the phrase is now a neural-net-like mass in semantic space: a central cluster, with radiating lines, from which yet more lines radiate to termini ut nunc.  But any of these termini -- and indeed, the central cluster -- may yet be developed as the discourse proceeds.  The school may be established as a single building, which may reflect back on the original cluster.  Or the curriculum may get spelled out.  Or a teacher introduced.  Or the girls' ethnicity may be an issue.  And so on.  To a large extent, such developments are sequential in a dialogic development, yet some may reflect back to earlier points in various ways (as in the confinement to a single building just noted).  And, of course, in the process other centers spring up and evelop and their expansions come to intersect with earlier center.  In the end, an interlanguage text is a four dimensional spreading shadow across the semantic field (unfortunately, the clearest metaphor seems to be a cancer).  No Loglan, nor otherlanguage, can represent this in its basic simplicity.  Every language can (at least in theory) say what the interlanguage represents.