Initial h before I, n, r disappeared in ONorw. at a preliterary date but was regularly retained in Olcel. (cf. Noreen §289). In Olcel. this initial h- was often lost, or an initial h- was often added to I, n, r, through association between words having initial hl, hn, hr and words having initial l, n, r. This associative process was due to resemblance in meaning or in form aside from the initial consonants in question. After the loss or the accretion of initial h- had become established between certain words or word groups there developed a general feeling of uncertainty as to whether a word should begin with hl or l, hn or n, hr or r. Most cases of this type of general confusion are naturally found in the Late Olcel. period, yet several instances may be cited in the classical period; which shows that the process was even then well under way. For the classical period I may mention the following examples: Hniflungr (Elder Edda) for Niflungr, rār ‘damp’ (Elder Edda) for hrār, hrjā for rjā ‘wrestle’; for the Late Olcel. period hnezla (for nezla) ‘button loop’, hniᵭra (for niᵭra) ‘to lower’, reᵭr (for hreᵭr) ‘genitals’, hreifr (for reifr) ‘glad’, hrifsa (for rifsa) ‘to rob’, ringja (for hringja) ‘a round pail’.

The verbs of the Hittite hi-conjugation fall into three classes according to the final sound of the stem. In spite of more or less irregularity in each class, the following paradigms will represent the usual types of conjugation.

The infinitive form of this verb is standan. The stem of the verb must therefore have been *stađ-. Accordingly the preterite singular form stōþ must represent an earlier *stōđ; hence we should have expected a preterite plural form *stōdum(= *stōđum) instead of stōþum.

1. This paper proposes some modifications of the family-tree diagram traditionally used to indicate genetic relationships in comparative linguistics. The flaws in the traditional diagram have long been recognized; it would be difficult to improve on Bloomfield's discussion of the subject in Language §§18.9–12. Although contemporary linguists have rejected the theory on which the branching diagram was originally based, they have not taken the logical next step of trying to improve the diagram itself so as to make it compatible with the revised theory. I shall attempt to show here that it is not necessary to reject branching diagrams entirely, but that it is possible to modify the traditional diagram so that it presents a truer and clearer picture of linguistic history as it is conceived at the present time. There are probably many possible ways in which we could depict linguistic history, and I do not attempt to do justice to all possibilities here; I hope only to show some variants on the traditional diagram which may lead to a better understanding of the potential use of such devices. First, a word on the role of diagrams in historical work. The diagrams which are presented here are not intended as theoretical models, but only as visual aids. Such diagrams are at best shorthand representations of the history of a group of languages, showing in a single picture the interrelations of the whole group. The primary danger to be avoided is that of overliteral interpretation; as Bloomfield says,

The present paper attempts to analyze a portion of Spanish verb morphology, namely the morphemes which indicate person. The model within which the analysis is framed is that of Item and Arrangement. Phonemic stretches are segmented into morphs; each morph boundary coincides with a phoneme boundary, but not conversely. We reject replacive and subtractive morphs. We accept zero morphs, but not zero morphemes, i.e. morphemes with zero as their only member. We also accept portmanteau morphs, i.e. morphs which belong simultaneously to two morphemes, but (again) not morphemes with portmanteau morphs as their only member. On the basis of the Spanish data, we suggest a further restriction on the use of zero and portmanteau morphs.

This paper presents a method for the analysis of connected speech (or writing). The method is formal, depending only on the occurrence of morphemes as distinguishable elements; it does not depend upon the analyst's knowledge of the particular meaning of each morpheme. By the same token, the method does not give us any new information about the individual morphemic meanings that are being communicated in the discourse under investigation. But the fact that such new information is not obtained does not mean that we can discover nothing about the discourse but how the grammar of the language is exemplified within it. For even though we use formal procedures akin to those of descriptive linguistics, we can obtain new information about the particular text we are studying, information that goes beyond descriptive linguistics.