It is suggested that the voicing process of Verner's Law did not follow the Germanic shift of IE /p t k kw s/, but was part of it. The shift itself was a form of secondary split, the allophones that were formed with less effort becoming respectively / ð g gw z/ and those that were formed with more effort producing /f þ x xw s/: hence IE ∗dek'- > ∗teg.- > OE -tig 'decade, -ty', beside IE ∗dék- > ∗tex- > OS tehan ‘ten’. Except in initial position, the conditioning factor was the pre-Germanic primary accent. In initial position, the fixation of primary stress inhibited voicing by introducing or preserving more effort: IE ∗pḷnós > Go. fulls ‘full’. If the accent of IE ∗pḷnós had remained unchanged for centuries in Germanic, the Gothic would have been ∗bulls. Voicing was also inhibited in voiceless consonant clusters: ∗oktow > Go. ahtau ‘eight’. Unlike initial syllables, which were word-bound and received primary stress, a proclitic was phrase-bound and never fully accented: thus IE ∗kom'- > ∗gan'- > ∗ga'- > OE ge'-. The voicing process was not completely uniform in Germanic because the fixation of primary stress did not occur simultaneously in all words of all dialects. The frequent absence of medial voicing in Gothic is due to an earlier fixation of initial stress and to the very early separation of Gothic from the other Germanic dialects.

A set of rules is here presented to account for the changes of Grimm's and Verner's Laws, as well as for the Germanic consonant lengthening. It is shown that the changes which occurred were to some extent predictable from the system out of which they arose. The rules are ordered in such a way that Verner's Law and the consonant lengthening precede the simultaneous development of voiced and voiceless aspirates to spirants.

The English passive construction is here viewed essentially as a special case of the verb complement construction, the so-called passive transformation being decomposed into a few more general elementary operations (transformations). Some theoretical implications are also discussed, and general conditions on transformation as an operation are tentatively proposed.

It is claimed that what and which, both as interrogatives and as relatives, are derived from the same underlying representations, wh + some and wh + that, respectively. It is assumed that, in the basic form of relativization, the two occurrences of the noun modified by a relative clause, in the matrix and the constituent sentences, may take different determiners. Indeed, all of the four possible ways of distributing the two basic determiners in the matrix and the constituent sentences are claimed to be realized in certain types of relativization. A transformation called Definitization which transforms indefinite determiners into definite ones is substantiated in the course of the argument. Some considerations of semantic character are also added.

The traditional explanation of Proto-Romance geminate initials (as in Italian rafforzamento), in terms of a progressive assimilation subsequent to Latin, is here rejected in favor of a theory of cognation between the strikingly parallel initial morphophonemics of Romance and Celtic. To sustain this, it is necessary to redefine geminates as units and medial so-called single consonants as phonemically composite. Romance syntactic ‘gemination’ is thus replaced by syntactic ‘lenition’, as in Celtic. Lenition actually handles certain Italian features better than does gemination.

Indo-Iranian ∗-rt- is represented in Avestan sometimes as /Ř/ (written š), sometimes as rt. It is maintained that /Ř/ is the normal phonological reflex of ∗-rt-, and that rt has been restored according to Kuryłowicz's ‘Fourth Law of Analogy’ in motivated categories where there was still awareness of a morpheme boundary between the /r/ and the /t/.

A proposal for modifying the tagmemic model is here illustrated in the description of the Huixtec Tzotzil verb phrase. Pike's notion of slot-class (Longacre's function-set) is retained, but the close correlation of slot and class as properties of a unit of grammar (the tagmeme) is rejected. An attempt is made to partially specify the form of rules in the lexical hierarchy of Tzotzil, here termed the semantic component. Rules for relating its semantic, syntactic, and phonological components are also presented.