General Comments on the Design of a Language jimc 930528 Before speaking, we conceive of "relations" between "things". One category of "things" that we conceive of is physical objects; another is abstract relations between such objects; yet another is the structures of speech and thought themselves. Metaphysically defining the referents of the above assertion is beyond the scope of this document; the scope consists of describing what design features of a language are needed to make it useful to humans for speech. Specifically, I am concerned with what is called a "predicate language". Before we speak, we conceive of a relation between things. We then say words which are symbols by which a listener is supposed to be able to identify, with more or less fidelity, the relation and the things which we thought of. The relation is technically referred to as a predicate, the things related are called arguments, and the placeholders in the relation that can be occupied by arguments are called cases. You can think of the predicate as a true-false valued function with a particular number of cases. Another set of words for "cases" and "arguments" is "formal parameters" vs. "actual parameters". In Latin-derived natural languages the cases have names such as "nominative" or "dative" and comparing one predicate to another there is a certain similarity between the role in the relation of each of the cases in different words. No claim is made here that such regularity is found in every language or in all words of any particular language; nonetheless such regularities help a designer or learner powerfully. One way to define a predicate is to list every set of arguments that is thus related. The list (a set of sets) is referred to as the "meaning list" of the predicate. Some predicates (e.g. color words) have only one case; for them the meaning list will consist of a set of sets each with exactly one member (e.g. something with that color). This special case is not so special after all, because vague are the rules by which a language designer chooses which potential arguments in a relation should get cases, and which should be brought in by other means; a predicate may be designed with one case today and later be given more of them, with no major philosophical effort. Each predicate is represented by a word. A predicate word with its arguments (expressed by words) is called a "clause". Clauses have three roles in speech: 1. Connected discourse consists of a sequence of clauses, called sentences. The sentences contain all the other clauses in the speech. A sentence is the symbol set by which the speaker's meaning is presented to the listener. The predicate of the sentence is the relation which the speaker mainly wants the listener to be aware of. We say that the speaker "asserts" the sentence. Other varieties of sentences are questions and commands. 2. There are also supplementary subordinate clauses which have the same purpose as sentences, but which are buried syntactically. The speaker buries them either because they are less important, or because they have some special relation with some sub-clause of the main sentence. In this analysis, emotional and discursive statements are considered to be sentences or supplementary clauses conveying to the listener the emotion of the speaker or hints about the organization of the discourse. In other languages they are analysed as a separate kind of clause (interjection, etc.), but I think they differ only in the topic being carried to the listener, not in the fact of conveying information or in the potential ways to represent the information by words. 3. Argument clauses are specialized to represent the arguments in a clause. Generally an argument clause represents a set of arguments, called its referent set. The argument clause is built from a predicate, and the referents are those meaning list elements which fit in a particular case of that predicate, consistent with whatever other verbiage is in the argument clause. For example, the referent set of "cat food" is the set of "things" A such that for some cat N, "eat(N,A)" is true. 4. Restrictive subordinate clauses have the job of retaining only a subset of a referent set or meaning list. For example, "favorite cat food" would be the same set as above, intersected with the set of all A which is favored (by presumably the particular cat N). 5. An abstraction represents the meaning list of some predicate (consistent with additional verbiage) interpreted as the referent set of an argument. This is how we talk about events and conditions. The syntax of a language has two jobs. First, it must guide the listener to group the words into clauses; and second, it must enable him to recognize the roles of and relations between the clauses. Elements to be provided by the syntax include: 1. The role of the clause: main sentence, argument, abstraction, or supplementary or restrictive subordinate clause. Typically additional sub-categories of supplementary clause are found useful, e.g. for emotions. In many languages, commands, questions and suppositions are distinguished from assertions through machinery analogous to role selection. 2. For arguments, the case it occupies in the containing clause. The case of an argument is a sub-type of its role as an argument. 3. The export case. Arguments and restrictive clauses carry out their role through a particular case of the predicate, which must be identified through syntax. Main sentences do not necessarily have a particular export case, since they don't export a referent set, although many languages distinguish one of the cases of a sentence. While supplementary clauses are functionally like sentences, in most languages they are represented syntactically similar to restrictive clauses and hence will have an export case. Similarly, abstractions don't necessarily have an export case, but surprisingly frequently they need to have some case of the containing clause patched through to one of their own cases, which is kind of the converse of exporting. In addition in many languages, arguments bear a "determiner" which, besides syntactically "determining" the start of the argument, indicates transformations on its referent set. Typically seen transformations are: 1. In-mind vs. veridical: Do the referents really truly fit the predicate (so says the speaker), or is the predicate merely a suggestive cue to the listener? Example: "the woman was a man in drag": "woman" is in-mind whereas "man" is veridical. Variously stringent rules are seen for how to interpret in-mind arguments. 2. Sets vs. extensions: Should the referent set be interpreted as a set, or does the containing clause apply to its members one by one?