A special abstractness problem in generative grammar—that of intermediate derivational stages containing elements or sequences that are not well-formed as surface representations—is examined. Suspicious analyses employing such FALSE STEPS are cited from the phonological and syntactic literature. It is argued that false steps cannot be ruled out in general, because there is support for many false-step analyses, in both syntax and phonology; a Welsh morphophonemic case is treated in some detail.

If the grammars which linguists write are to reflect psychologically real constructs, then purely structural evidence is not sufficient proof of them; some form of psychological evidence is required. The psycholinguistic test described here was designed to show whether or not a Hindi word such as [ghõ:sla:] ‘nest’, which appears only in this form on the surface, has an underlying form with an ‘abstract’ , i.e. ghõ:sla:. The results show that some speakers have the abstract underlying form, others do not.

Many ‘physical phonetic ‘ variables are continuous, so physically there is an infinitely large range of possible sounds. The generative phonologists, unlike Bloomfield, believe in a universal level of ’systematic phonetics’, where only a fixed finite number of distinct values are possible for each phonetic variable, and hence there is only a finite set of possible segments. To justify this hypothesis it would be necessary (a) to show that facts follow from it which are arbitrary on Bloomfield's theory, and (b) to explain away any counter-evidence. It is argued here that hardly any true predictions have been shown to follow from the hypothesis, and evidence is offered against it.

Intelligibility tests between dialects give results with three characteristics that need to be taken into account in their interpretation: asymmetry, sensitivity to changes in threshold, and incompleteness. If we interpret dialects as networks of communication arrived at by optimization, we make adequate use of these properties.

A generative treatment of the early stages of development of two children suggests that the main learning process in pronunciation consists of discovering and gaining control over the articulatory features required to make sounds heard. The lexical representation differs from the phonetic output at all stages because actual pronunciations are constrained by primitive or acquired articulatory processes that have the effect of imposing a phonotactic filter on the speech output. Rules and representations depict ‘competence’ and have ‘psychological reality’ by describing the articulatory analysis made by the learner, together with the constraints imposed by the filter. The phoneme concept required is similar to Sapir's: no invariance or bi-uniqueness conditions are imposed, but phonemes are much more concrete than those of current generative phonology. The abstractness of the latter results from a methodology that enthrones economy criteria requiring assumptions about human memory and learning which are almost certainly wrong. Alternatives are discussed.

Aphasiological evidence is presented in support of various linguistic hypotheses. The transmodal performances of an aphasic subject provide clear evidence in favor of performative analysis, discourse rules, a copula-creation rule for English, the partial independence of syntax and semantics, and the function of pronouns as pure variables. The possible contribution of neurolinguistics to linguistic theory is evaluated.

The deictic verbs come and go, as well as bring and send, may be used in idioms to refer to change of state rather than to motion. In such idiomatic uses, it is proposed, the deictic center, corresponding to the goal of come, is provided by some NORMAL STATE of being, e.g. consciousness or realism as in He came round and He came down to earth. In contrast, go marks departure from a normal state, as in He went mad and He went into a daze. Come is never used to denote departure from, nor go entry into, a normal state. Besides normal-state deixis, come may be used in evaluative deixis to indicate approval of some end state, as in He came through a lot, while evaluative go is used with a non-positive meaning, as in He went through a lot. It is suggested that both classes of idioms are related to other forms of deixis, all of which derive from the basic deictic contrast between EGO and NON-EGO.

Speech performances are here considered as statistical samples drawn from a probabilistic language competence. This competence is modeled in conventional generative terms, except that optional rules are assigned application probabilities as functions of the structure of the input strings, possibly depending on the extralinguistic environment as well. We develop the mathematical background for these variable rules, which were introduced by Labov, and we apply the theory and methodology to examples from Spanish, French, and English. The data consist of relative frequencies of rule application for different types of input string, and they provide a check on the frequencies predicted by variable rules. This study supports the hypothesis that the various features in rule structural descriptions tend to act independently, in the statistical sense, on rule probabilities. The analytic framework for variable rules is easily and naturally extendable to take into account the effect of sociolinguistic and stylistic features on application frequencies.

The Finnish language in Minnesota has undergone considerable interference through four generations of contact with English, largely because of syntagmatic congruence with English sentences. The noun cases have responded in a variety of ways to this interference, demonstrating that both the degree of boundness and syntactic congruence are critical to the explanation of interference, as case-endings in highly bound environments are much more resistant to interference than those in moderately or loosely bound environments.