Copyright, 1991, by the Logical Language Group, Inc. 2904 Beau Lane, Fairfax VA 22031-1303 USA Phone (703) 385-0273 lojbab@lojban.org All rights reserved. Permission to copy granted subject to your verification that this is the latest version of this document, that your distribution be for the promotion of Lojban, that there is no charge for the product, and that this copyright notice is included intact in the copy. Loglan and Lojban: A Linguist's Questions And An Amateur's Answers by John Cowan (ci'a la djan. kau,n.) Internet address: cowan@snark.thyrsus.com The following questions about Loglan are based on a 1969 review by Arnold M. Zwicky of James Cooke Brown's 1966 edition of Loglan 1. Although basically friendly, Zwicky's review raises a large number of linguistic objections to Loglan as it existed in 1966. The review represents the only formal notice the linguistics community has ever taken of the Loglan Project. Unfortunately, the Project has never made any reply. The answers that appear here reflect the perspective of Lojban (not Institute Loglan) as it exists in 1991. Therefore, no attempts have been made to sort out Zwicky's misunderstandings of Brown's text, Brown's misunderstandings (or mistakes in writing) about his own language, valid points as of 1969 that were later changed by Brown, and valid points as of 1969 that were changed when (or since) Lojban split from Institute Loglan. Throughout, "Loglan" refers to 1966 Loglan as seen by Zwicky, and "Lojban" to 1991 Lojban as seen by me. The word "Lojban" is derived from the same metaphor as "Loglan" ("logical language") but using Lojban words ("logji bangu"). As the title indicates, I am only an amateur (lit. "lover") of linguistics, and I may misinterpret some of Zwicky's points. The question-and-answer format used here is purely for expository convenience. Zwicky is not responsible for the form of the questions, which reflect only my interpretations of his points, except for quoted text within the questions followed by (Z), which are quotations from Zwicky's original review. That review was published in Language 45:2 (1969), pp. 444-457. 1. Lojban sentences do not have unique interpretations; how can Lojban be said to be unambiguous? The sense in which Lojban is said to be unambiguous is not a simple one, and some amplification of the fundamental claim is necessary. Ambiguity is judged on four levels: the phonological-graphical, the morphological, the syntactic, and the semantic. Lojban is audio-visually isomorphic: the writing system has a grapheme for every phoneme and vice versa, and there are no supra- segmental phonemes (such as tones or pitch) which are not represented in the writing system. Lojban's phonology contains significant pauses that affect word boundaries, and allows pauses between any two words. The optional written representation for pause is a period, although pauses can be unambiguously identified in written text from the morphological rules alone. Lojban also uses stress significantly, and again there is a written representation (capitalization of the affected vowel or syllable), which is omitted in most text, where the morphological default of penultimate stress applies. Lojban is morphologically unambiguous in two senses: a string of phonemes (including explicit pause and stress information) can be broken up into words in only one way, and each compound word can be converted to and from an equivalent phrase in only one way. The syntactic unambiguity of Lojban has been established by the use of a LALR(1) parser generator which, in cooperation with a series of simple pre-parser operations, produces a unique parse for every Lojban text. In addition, the existence of a defined 'phrase structure rule' grammar underlying the language (and tested via the parser generator) guarantees that there are no sentences where distinct deep structures generate isomorphic surface structures. On the other hand, Lojban does have transformations, although they are not explicit in the machine grammar: there are distinct surface structures which have the same semantics, and therefore reflect the same underlying deep structure. The claim for semantic unambiguity is a limited one only. Lojban contains several constructs which are explicitly ambiguous semantically. The most important of these are Lojban tanru (so-called 'metaphors') and Lojban names. Names are ambiguous in almost any language, and Lojban is no better; a name simply must be resolved in context, and the only final authority for the meaning of a name is the user of the name. tanru are further discussed in later replies. 2. If the meaning of a particular tanru cannot be completely understood from understanding the component parts, a separate dictionary entry is needed for every possible tanru, making the Lojban dictionary infinitely long. How can this be avoided? tanru are binary combinations of predicates, such that the second predicate is the 'head' and the first predicate is a modifier for that head. The meaning of the tanru is the meaning of its head, with the additional information that there is some unspecified relationship between the head and the modifier. tanru are the basis of compound words in Lojban. However, a compound word has a single defined meaning whereas the meaning of a tanru is explicitly ambiguous. Lojban tanru are not as free as English figures of speech; they are 'analytic', meaning that the components of the tanru do not themselves assume a figurative sense. Only the connection between them is unstated. Most of the constructs of Lojban are semantically unambiguous, and there are semantically unambiguous ways (such as with relative clauses) to paraphrase the meaning of any tanru. For example, "slasi mlatu" ("plastic-cat") might be paraphrased in ways that translate to "cat that is made from plastic" or "cat which eats plastic" or various other interpretations, just as in English. However, the single (compound) word derived from this tanru, "slasymlatu", has exactly one meaning from among the interpretations, which could be looked up in a dictionary (if someone had found the word useful enough to formally submit it). There is no law compelling the creation of such a word, however, and there is even an 'escape mechanism' allowing a speaker to indicate that a particular instance of a 'nonce' compound word is 'nonstandard' (has not been checked against a dictionary or other standard), and may have a meaning based on an unusual interpretation of the underlying tanru. 3. The Loglan 'primitive words' seem to have been chosen at random, without regard to any sort of semantic theory. Why was this done? Lojban content words are built up from a list of about 1300 root words (called "gismu"), which are not necessarily to be taken as semantically simple. Lojban does not claim to exhibit a complete and comprehensive semantic theory which hierarchically partitions the entire semantic space of human discourse. Rather, the 1300-odd root words blanket semantic space, in the sense that everything human beings talk about can be built up using appropriate tanru. This claim is being tested in actual usage, and root words can still be added if necessary (after careful consideration) if genuine gaps are found. For the most part, the few gaps which are now recognized (about 20 words will be added soon) reflect the completing of semantic sets. It is no longer permitted for language users to create new gismu root words (in the standard form of the language, at least); newly coined words must fall recognizably outside the highly regulated gismu morphological space (a specific and separate morphological structure is reserved for coined words - usually borrowings - and a marker is available to indicate that a word is a 'nonce' coinage rather than an established 'dictionary word'). Lojban's empirically derived word list is similar to that of Basic English, which replaces the whole English vocabulary with English-normal compounds built from about 800 root words. Lojban and Basic English both allow for the adoption of technical terms from other languages to cover things like plant and animal names, food names, and names of chemical compounds. The unfortunate terms "primitive word" and "prim" formerly used by the Loglan Project suggested the notion that Lojban's set of gismu was meant to be a list of semantic primitives. This is not the case for Lojban, and the more neutral term "root word" was adopted recently to reduce confusion. Lojban predicate words, therefore, are now divided into gismu 'root words', lujvo 'compound words' and le'avla 'borrowings' (lit. 'taken words'). (Brown did originally select some words as 'semantic primitives'; however, he later added words with no claim that the additions were 'primitive' in the same sense). 4. Some tanru seem poorly designed and not in keeping with expressed standards. Also, tanru like "nixli ckule", analogous to English "girls' school", are so open-ended in sense that there is no way to block such far-fetched interpretations as "a school intended to train girls between the ages of 6 and 10 to play the bassoon", which is patently absurd. What is the proper interpretation of tanru? In the early part of the Loglan Project, poor tanru were regrettably common. In particular, it was common for tanru to be calques on English expressions, such as "beautiful type of small" for English "pretty small". Many tanru employed the primitive for "make"' (in the sense "make from materials") where "cause" would have been more appropriate (e.g. "kill" = "dead-make"). Many years worth of effort since then have gone into removing such malglico ('derogatively English') tanru from Lojban texts. The Lojban tanru "nixli ckule" ("girl type of school") cannot mean, out of context, "school intended to train girls between 6 and 10 years of age to play the bassoon", although if such a school existed it could certainly be called a nixli ckule. This interpretation can be rejected as implausible because it involves additional restrictive information. The undefined relationship between "nixli" and "ckule" cannot drag in additional information 'by the hair', as it were. Instead, this intricate interpretation would require a larger tanru incorporating nixli ckule as one of its components, or else a non-tanru construct, probably involving a Lojban relative clause. As a comparision, such interpretations as "school containing girls", "school whose students are girls", and "school to train persons to behave like girls" are plausible with minimal context because these renderings do not involve additional restriction. 5. Lojban claims to be unambiguous, but many constructs have vague meanings, and the meanings of the primitives themselves are extremely poorly specified. On the other hand, Lojban forces precision on speakers where it is not wanted and where natural-language speakers can easily avoid it. Is this appropriate to a culturally neutral, unambiguous language? Lojban's avoidance of ambiguity does not mean an avoidance of vagueness. A Lojban aphorism states that the price of infinite precision is infinite verbosity, as indeed Wilkins' Philosophical Language illustrates. Lojban's allowable vagueness permits useful sentences to be not much longer than their natural-language counterparts. There are many ways to omit information in Lojban, and it is up to the listener to reconstruct what was meant, just as in natural languages. In each construct, there are specific required and optional components. Unlike English, omitting an optional component explicitly and unambiguously flags an ellipsis. Furthermore, the listener has a clear way of querying any of this elliptically omitted information. There are also some categories which are necessary in Lojban and not in other languages. For example, Lojban requires the speaker, whenever referring to objects, to specify whether the objects are considered as individuals, as a mass, or as a (set theoretic) set. Likewise, logical relations are made explicit: there can be no neutrality in Lojban about inclusive vs. exclusive 'or', which are no more closely related semantically than any other pair of logical connectives. These properties are a product of Lojban's fundamental design, which was chosen to emphasize a highly distinctive and non-natural syntax (that of formal first-order predicate logic) embedded in a language with the same expressive power as natural languages. Through the appearance of this one highly unusual feature, the intent of the Loglan Project has been to maximize one difference between Lojban and natural languages without compromising speakability and learnability. This difference could then be tested by considering whether the use of first-order predicate logic as a syntactic base aided fluent Lojban speakers in the use of this logic as a reasoning tool. As to the 'primitives', Lojban gismu roots are defined rather abstractly, in order to cover as large a segment of closely related semantic space as possible. These broad (but not really vague) concepts can then be restricted using tanru and other constructs to any arbitrary degree necessary for clarity. Communicating the meaning of a gismu (or any other Lojban word) is a problem of teaching and lexicography. The concepts are defined as predicate relationships among various arguments, and various experimental approaches have been explored throughout the Loglan Project to determine the best means to convey these meanings. It is believed that the current working definitions of the gismu are much more clear than the 1966 set. 6. On a more technical note, Lojban tanru involving more than two components are always left-grouping (in the absence of a marker word). Right-branching structure is "much more natural to human languages" (Z). Why was this choice made? Lojban is predominantly a left-branching language. By default, all structures are left-branching, with right-branching available when marked by a particle. Since the head of most constructs appears on the left, left-branching structures tend to favor the speaker. Nothing spoken needs to be revised to add more information. When the head is on the right, as in the case of tanru, left-branching may seem counterintuitive, as it requires the listener to retain the entire structure in mind until the head is found. However, left-branching was retained even in tanru for the sake of simplicity. Experience has shown, however, that Lojban's left-branching structure is not a major problem for language learners. Indeed, many longer English metaphors translate directly into Lojban using simple left-branching structures. 7. Loglan anaphora use a convention which is "quite precise, and also quite unlike anything in natural languages" (Z), involving counting backward from the reference to the referent. This provides unique reference, but is also difficult to understand and use. Is there nothing better that preserves the desirable property of unique reference which a logical language needs? The Lojban anaphora conventions have undergone much revision and expansion since the early days of Loglan. There now exist both the "traditional" Loglan back-counting anaphora, which refer to previous referents, and more "natural-language-like" anaphoric words which are meaningless until assigned. Assignment may be either in afterthought or forethought. These words are somewhat like natural language pronouns, but may more closely be compared to the use of regions of space in American Sign Language to refer to remote persons and things. Unassigned space regions in ASL are similarly meaningless. It is no longer a required convention that anaphora variables be assigned in a fixed order. Subscripts (as in mathematics) are allowed almost everywhere in the language, and provide for a countable infinity of variables as of many other things. Lojban also has added the capability of using individual letters and acronyms as anaphoric symbols. 8. Why does Loglan have a different and even more complex system of "personal pronouns" for speaker/listener reference? Is this level of complexity really in order for what other languages treat as a simple matter? Lojban personal pronouns have been simplified. There are now forms for I, II, III, I and II, I and III, II and III, and I and II and III. There are no separate forms (and never have been) for plurals, because number is not a mandatory grammatical category in any part of Lojban. Number is expressed, when needed, using explicit numerals (which include both precise and vague forms analogous to English 'some', 'few', 'too many', etc.) Honorifics were recently added to the language, using a general mechanism which may apply to any word or construct, not merely to pronouns. 9. Why does Loglan treat predicate connection as primary and sentence, argument, etc. connection as secondary? Whatever may have been assumed in the past for pedagogical purposes, logical connection between sentences is basic to Lojban. All other forms of logical connection may be transformed into equivalent sentence connections. 10. Why are there so many structure words, and why are many of them so similar? Wouldn't this make Loglan hard to understand at a cocktail party (or a similar noisy environment)? One of the recurrent difficulties with all forms of Loglan, including Lojban, is the tendency to fill up the available space of structure words, making words of similar function hard to distinguish in noisy environments. The phonological revisions made when Lojban split from Institute Loglan allowed for many more structure words (cmavo), but once again the list has almost entirely filled. In some cases, notably the digits 0-9, an effort has been made to separate them phonologically. The vocatives (including the words used for communication protocol, e.g. over the radio) are also maximally separated phonologically. Many other function words are based on shortened forms of corresponding gismu roots, however, and are not maximally separated. A variety of ways to say "Huh?" have been added to the language, partially alleviating the difficulty. These question words can be used to specify the type of word that was expected, or the part of the relationship that was not understood by the listener. 11. Loglan's "restrictions on stresses and pauses results in long sequences of unstressed syllables which must be pronounced without a break" (Z). This makes correct speech a "trial for a speaker of English or Russian, and not easy even for a speaker of French" (Z). Natural languages often have non-significant pauses, but in Loglan every non- required pause is forbidden. Is Loglan really speakable? Lojban allows certain flexibilities of pause and stress in the area of structure words. By default, all structure words are unstressed. However, it is possible to set off structure words with optional pauses, and even to give them optional stress, subject to a single limitation: a structure word followed by a predicate word without pause must not be stressed. Pauses are now permitted between any two words; only within a word is pause forbidden, and most words are short. gismu and cmavo are always one or two syllables long, and many lujvo compounds are only two or three syllables. 12. "A partial explanation for the existence of transformations is to be found in the necessity for providing speakers of any language with relatively acceptable variants of certain types of deep structures." (Z) Loglan has no transformations, making some sentences expressible, but far from straightforward or easy to use. Doesn't this make Loglan harder to use than typical natural languages? Lojban does have transformations, in the sense that there are several alternative surface structures that have the same semantics and therefore, presumably, the same deep structure. What it does not have is identical surface structures with differing deep structures, so a surface-structure-only grammar is sufficient to develop an adequate parsing for every text. Knowledge of transformations is required only to get the semantics right. 13. Lojban connectives cannot be used to correctly translate English "If you water it, it will grow", because material implication is too weak and the special causal connectives, which connect assertions, are too strong. What can be done instead? The English sentence "If you water it, it will grow" looks superficially like a Lojban "na.a" connection (material implication), but it actually has causal connotations not present in "na.a". Therefore, a proper translation must involve the notion of cause. Neither the Lojban coordinating causal conjunction nor the two correlative subordinating causal conjunctions (one of which subordinates the cause and the other the effect) will serve, since these require that either the cause, or the effect, or both be asserted. Instead, the correct translation of the English involves "cause" as a predicate, and might be paraphrased "The event of your watering it is a cause of the event of its future growing." 14. How can Loglan logical connectives be used in imperative sentences? Logical connectives work properly only on complete sentences, and of those, only those which actually assert something. In early versions of Loglan, imperatives were marked by a predication without a subject. In Lojban, there is a special imperative pronoun "ko". This is a second person pronoun logically equivalent to "do", the normal Lojban word for 'you', but conveying an imperative sense. Thus, an imperative can be understood as commanding the listener to make the assertion true which results when "ko" is replaced by "do". For example, "ko sisti" ('Stop!') is logically equivalent to "do sisti" ('you stop'), and pragmatically may be understood as 'Make "do sisti" true!". This allows logical connection to be used in imperatives without loss of clarity or generality; the logical connection applies to the assertion which is in effect embedded in the imperative. A minor advantage of this style of imperative is that tensed imperatives like "ko ba klama", ('Come in-the-future!') become straightforward. 15. Loglan's existential (bound) variables appear to be non-standard. Brown states that the value of an existential variable is always unknown to the speaker, rather than merely being unspecified (perhaps for reasons of privacy or germaneness). Why is this? Also, why isn't quantification over predicates provided? Why are the back-counting anaphora unable to refer to existential variables? Existential variables are now interpreted in a standard way, to refer to something unspecified, or something specified by a restrictive relative clause ("all x such that..."). There are separate sets of variables for quantifying over arguments and over predicates. In general, the back-counting anaphora (which are less important in Lojban than in Loglan) are not used to refer to other anaphoric words; this makes the counting convention a bit more complex, but leads to more generally useful results. 16. Untensed sentences ought to be neutral with respect to tense, mood, and aspect, but Brown treats untensed sentences as expressing disposition, habit, or ability -- lasting throughout all time. This is inconsistent with other parts of the language which treat ellipsized material as merely unspecified. The Lojban tense system has been greatly elaborated and clarified with respect to its Loglan predecessor. There are now specific mechanisms for stating the potentiality or actuality of a predication; in the absence of these, a predication is neutral concerning the degree of actuality expressed by it. It is no longer true that "untensed" predicates are used to express disposition or habit. They may be so used, by ellipsis, but are in fact neutral in the absence of further evidence. Lojban tense, like other incidental modifiers of a predication, tend to be contextually "sticky". When once specified in connected discourse, to whatever degree of precision seems appropriate, tense need not be respecified in each sentence. In narration, this assumption is modified to the extent that each sentence is assumed to refer to a slightly later time than the previous sentence, although with explicit tense markers it is possible to tell a story in reversed or scrambled time order. Therefore, each predication does have a tense, one that is implicit if not necessarily explicit. 17. The decisions about the degrees of predicates (the number of arguments expected for each) seem arbitrary. Color words are treated as relations of degree 2; weather predicates which have no real subject nevertheless need at least one argument; event predicates like "kiss" don't have an argument specifying the time. What theory underlies the choice of place structures? Very little. Place structures are empirically derived, like the root word list itself, and present a far more difficult problem; therefore, they will be standardized (if ever) only after everything else is complete. Many of the particular objections made above have force, and have already been accepted. There is no sufficiently complete and general case theory that allows the construction of a priori place structures for the large variety of predicates that exist in the real world. The current place structures of Lojban represent a three-way compromise: fewer places are easier to learn; more places make for more concision (arguments not represented in the place structure may be added, but must be marked with appropriate case tags); the presence of an argument in the place structure makes a metaphysical claim that it is required for the predication to be meaningful. This last point requires some explanation. For example, the predicate "klama" ("come, go") has five places: the actor, the destination, the origin, the route, and the means. Lojban therefore claims that anything not involving these five notions (whether specified in a particular sentence or not) is not an instance of "klama". The predicate "cliva" ("leave") has the same places except for the destination; it is not necesssary to be going anywhere in particular for "cliva" to hold. "litru" ("travel") has neither origin nor destination, merely, the actor, the route, and the means. The predicate "cadzu" ("walk"), involves only a walker and a means of walking (typically legs). One may walk without an origin or a destination (in circles, e.g.). For describing the act of walking from somewhere to somewhere, the tanru "cadzu klama" or the corresponding lujvo "dzukla" would be appropriate. The tanru "cadzu cliva" and "cadzu litru" may be similarly analyzed. 18. The Loglan phonological system is hard for English-speakers (to say nothing of Japanese-speakers) to use, due to the large numbers of consonant clusters and non-English diphthongs. How can a language be appropriate as an international auxiliary language when it is difficult to pronounce? Lojban phonology is much better than 1966 Loglan's was. There are now only 4 falling and 10 rising diphthongs, and the rising diphthongs are used only in names and in paralinguistic grunts representing emotions. All 25 vowel combinations are used, but they are separated by a voiceless vocalic glide written with an apostrophe, thus preventing diphthongization. English-speakers think of this glide as /h/, and even speakers of languages like French, which has no /h/, can manage this sound intervocalically. Consonant clusters are controlled more carefully as well. Only 48 selected clusters are permitted initially; some of these, such as "ml" and "mr", do not appear in English, but are still possible to English- speakers with a bit of practice. Medial consonant clusters are also restricted, to prevent mixed voiced-unvoiced clusters, consecutive stops, and other hard-to-handle combinations. The new Lojban sound /y/, IPA [@], is used to separate "bad" medial clusters wherever the morphology rules would otherwise produce them. Difficulties with the variety of permitted initial sounds are overestimated. Lojban's morphology makes pronouncing these words easier than they first appear. Initial consonant clusters occur only in content words (predicates) and names. These words seldom are spoken in isolation; rather, they are expressed in a speech stream with a rhythmic stress pattern preceded (and followed) by words that end with a vowel. The unambiguous morphology allows the words to be broken apart even if run together at a very high speech rate. Meanwhile, though, the final vowel of the preceding word serves to buffer the cluster, allowing it to be pronounced as a much easier medial cluster. Thus "le mlatu" ("the cat"), while officially pronounced /le,MLA,tu/, can be pronounced as /lem,LA,tu/ with no confusion to the listener. In addition, the buffering sound, IPA [I] (the "i" of "English "bit") is explicitly reserved for insertion at any point into a Lojban word where the speaker requires it for ease of pronunciation. The word "mlatu" may be pronounced /mIlatu/ by those who cannot manage "ml", and nothing else need be changed. This sound is "stripped" by the listener before any further linguistic processing is done. 19. Loglan words resemble their English cognates, but unsystematically so. Does this really aid learning, or does it make learning more difficult? Lojban words are less English-like than prior versions of Loglan, since they were redone using new (1985) data on numbers of speakers. English is now less important in relative terms than Mandarin Chinese, and most Lojban words are fairly equal mixtures of the two languages, with lesser influences from Spanish, Hindi, Russian, and Arabic. The other languages used in 1966 Loglan are no longer as prominent in terms of world-wide number of speakers, and were dropped from the word-making algorithm. There is no proven claim that the Lojban word-making algorithm has any meaningful correlation with learnability of the words. Brown has reported that informal 'engineering tests' were conducted early in the Loglan Project, leading to his selection of the current algorithm, but these tests have never been documented or subjected to review. The Logical Language Group has proposed formal tests of the algorithm, and is instrumenting its software used for teaching vocabulary to allow data to be gathered that will confirm or refute Brown's hypothesis. Gathering this data may incidentally provide additional insights into the vocabulary learning process, enabling Lojban to serve the additional purpose of being a test bed for research in 2nd language acquisition. In any event, the word-making algorithm used for Lojban has the clear benefit of ensuring that phonemes occur in the language in rough proportion to their occurence in the source natural languages, and in patterns and orders that are similar to those in the source languages (thus the first syllable of Lojban gismu most frequently ends in /n/, reflecting the high frequency of syllable ending /n/ in Chinese). The result is a language that is much more pleasant-sounding than, for example, randomly chosen phoneme strings, while having at least some arguable claim to being free of the European cultural bias found in the roots of most other constructed languages. 20. Loglan has an absolutely fixed word order; in some cases, changes of word order are possible, but only by the addition of marker particles. Why is this? No natural language has an absolutely fixed word order (or for that matter, an absolutely free one). Lojban's word order is by no means fixed. In fact, Lojban is only secondarily a "word order" language at all. Primarily, it is a particle language. Using a standard word order allows many of the particles to be 'elided' (dropped) in common cases. However, even the standard unmarked word order is by no means fixed; the principal requirement is that at least one argument precede the predicate, but it is perfectly all right for all of the arguments to do so, leading to an SOV word order rather than the canonical SVO (subject-verb-oject). VSO order is expressible using only 1 particle. In two-argument predicates, OSV, OVS, and VOS are also possible with only one particle, and various even more scrambled orders (when more than two-place predicates are involved) can also be achieved. 21. Loglan does not have WH-questions of the English kind (its questions are fill-in-the-blank) and does not have relative clauses. Therefore, no "unbounded" transformations (in the technical sense) exist in the language. Sentences like "I met a man that John said Mary told George to visit" can be translated only with great pain. How can such fairly common types of constructions be represented better? Lojban does have relative clauses, of the Hebrew type; the relative marker and the relative pronoun are distinct. The marker "poi" (or "noi" for non-restrictive clauses) always comes at the beginning, but the embedded clause is in normal order, using the relative pronoun "ke'a" at the appropriate location to represent whatever is being elaborated by the clause. 22. If Loglan is to be used as an international auxiliary language, it must be culturally neutral. But many of its conceptual distinctions, for example the color set, are clearly biased towards particular languages. There is a word for 'brown', which is a color not used in Chinese (although a word exists, it is rare); on the other hand, there is only one word for 'blue', although Russian-speakers convey the range of English 'blue' with two words. How can Loglan be prevented from splintering into dialects which differ in such points? To some extent, such splitting is inevitable and already exists in natural languages. Some English-speakers may use the color term 'aqua' in their idiolect, whereas others lump that color with 'blue', and still others with 'green'. Understanding is still possible, perhaps with some effort. The Lojban community will have to work out such problems for itself; there are sufficent clarifying mechanisms to resolve differences in idiolect or style between individuals. The unambiguous syntax and other constraints defined in the language prescription should make such differences much more easily resolvable than, say, the differences between two dialects of English. The prescriptive phase of Lojban is not intended to solve all problems (especially all semantic problems) but merely to provide enough structure to get a linguistic community started. After that, the language will be allowed to evolve naturally, and will probably creolize a bit in some cultures. (A recent discussion has pointed out that observing the creolization of such a highly prescribed constructed language will undoubtedly reveal much about the nature of the processes involved. 23. Loglan is supposed to be intended as a test of the Sapir-Whorf hypothesis in its negative form: "structural features of language make a difference in our awareness of the relations between ideas" (Brown). Is this simply another way of saying "Distinctions are more likely to be noticed if structurally marked" (Z)? If so, this is trivially true. A better paraphrase might be "Unmarked features are more likely to be used, and therefore will tend to constitute the backgrounded features of the language". By making the unmarked features those which are most unlike natural-language features, a new set of thought habits will be created (if Sapir-Whorf is true) which will be measurably different from those possessed by non-Lojban speakers. If Sapir-Whorf is false, which is the null hypothesis for Lojban purposes, no such distinctions in thought habits will be detectable. Further elaboration of Loglan Project thinking about Sapir-Whorf has led to another alternate formulation: "The constraints imposed by structural features of language impose corresponding constraints on thought patterns." In attempting to achieve cultural neutrality, Lojban has been designed to minimize many structural constraints found in natural languages (such as word order, and the structural distinctions between noun, verb, and adjective). If Sapir-Whorf is true, there should be measureable broadening in thought patterns (possibly showing up as increased creativity or ability to see relationships between superficially unlike concepts). Again, the null hypothesis is that no measureable distinction will exist. 24. How can "ease of thought" be measured? Measuring facility with predicate logic is not enough to establish "ease of thought" Perhaps not. However, the Sapir-Whorf hypothesis tends to be confirmed if experiments show that Lojban-speakers have a greater facility with predicate logic than non-Lojban-speakers. That would indicate that language (natural language) limits thought in ways that Lojban-speakers can bypass. This form of test is not free of its own difficulties, which have been discussed elsewhere. Summary Professor Zwicky's analysis raises several points of concern to linguists who might be interested in the potential use of Lojban for linguistic research. It is believed that sufficient planning and linguistic understanding (and occasionally serendipity) has been incorporated in the Lojban language design process to meet these concerns. Other concerns no doubt exist; it is believed they can similarly be addressed, and that Lojban will prove linguistically viable, as well as useful in our attempts to understand language. Meanwhile, as Lojban has evolved since the 1966 version of Loglan, new features, not analyzed by Zwicky, have been added to the language, further enhancing its potential value. These features, such as Lojban's expression of the several varieties of natural language negation, the system of attitudinal words for emotional expression, and the discursives used for metalinguistic manipulation and comment on the discourse in progress, raise new questions about the adequacy of Lojban's design, while providing new opportunities for exploration of the properties of natural language, as well as the correctness of the Sapir-Whorf hypothesis. In 1991, it is time for linguists to again look at Lojban, with the expectation that new questions, and new respect, will be forthcoming.