The present work is a collection of studies on the Tocharian languages that focuses on

the phonological properties of units larger than the word. The first chapter involves a study

of the segmental properties of external sandhi in Tocharian A couched in the framework of

Optimality Theory. A philological study involving the Tocharian br ̄ahm ̄ı alphabet reveals

that there exists a correlation between external sandhi and the orthographic strategy used

to render a word-final consonant. The second chapter provides an analysis of the prosodic

relationship between clitics and their hosts, showing that these elements form a prosodic

constituent intermediate between the word and the phrase. This is the case both in Tocharian

A as well as in Tocharian B. The final chapter addresses two different aspects of the internal

structure of Tocharian Wh-words. First, the prosodic analysis employed for clitics and

their hosts is extended to Wh-words in Tocharian B, making it possible to account for

their otherwise aberrant accentuation as well as some of their segmental properties. The

second part of the chapter investigates the nature of indefinites in Tocharian A, showing

that, contrary to the descriptions in most handbooks, the attested forms do not belong to

a single paradigm. Instead, we are dealing with two separate sets of Wh-words doubling

as indefinites, a fact that has been obscured by a phonological process that renders one set

identical to demonstrative pronouns.

In this dissertation I will attempt to answer the following question: why does Classical Armenian have three different reflexes for the Proto-Armenian epenthetic vowel word-initially before old Proto-Indo-European consonant clusters? Two of the vowels, e and a, occur in the same phonological environment, and even in doublets (e.g., Classical ełbayr beside dialectal ałbär ‘brother’).

The main constraint driving this asymmetry is the promotion of the Sonority Sequencing Principle in the grammar. Because sibilants are more sonorous than stops, the promotion of the Sonority Sequencing Principle above the Strict Layer Hypothesis causes speakers to create a semisyllable to house the sibilant extraprosodically. This extraprosodic structure is not required for old consonant-resonant clusters since they already conform to the Sonority Sequencing Principle. Because Armenian has sonority-sensitive stress, the secondary stress placed on word-initial epenthetic vowels triggers a vowel change in all words without extraprosodic structure, i.e. with the old consonant-resonant clusters. Therefore Proto-Armenian */əłbayr/ becomes Classical Armenian [èł.báyr] ‘brother,’ but Proto-Armenian */<əs>tipem/ with extraprosodic <əs> becomes [<əs>.tì.pém] ‘I rush’ because the schwa is outside the domain of stress assignment.

Scholars have long suspected that the verbal reduplication patterns reflected in the Anatolian languages encode typologically expected imperfectivity and iconic expressivism. At the same time, Indo-Europeanists have looked to the Anatolian languages for relics of archaic reduplicated categories of marked semantics and shape. This dissertation provides a comprehensive examination of the forms and functions of the various Anatolian reduplicated verbs attested in the Hittite, Luvian, and Lycian languages. Following a typological survey to establish what exactly the “expected” semantics of reduplication in a verb are, the study then provides sketches for the semantics of all known Hittite, Luvian, and Lycian reduplicated verbs of sufficient attestation to do so. I then examine the formal history of all attested Anatolian reduplicated verbs, projecting their preforms as far back as able, and divide them according to their reduplicative patterns, commenting on the relative productivity of each. In the end, typologically trivial partial copy vowel reduplication proves to be the most productive pattern of verbal reduplication in the Anatolian languages, commonly used with habitual, durative, iterative, and other universally expected imperfective nuances. The same semantics obtain in the more marginal but also productive /i/-reduplication, a formal pattern inherited from Proto-Indo-European. Similarly inherited /e/-reduplication persists in some forms, with the same semantics seen in other partial reduplicative patterns in Anatolian. The majority of “expressive” reduplicated verbs are unpaired synchronically with a base stem, and frequently show full root reduplication. There are a handful of “intensive” reduplicated verbs with full root reduplication and a “linking” /i/, which may or may not be parallel to Vedic intensives of similar shape. The evidence overwhelmingly proves that the accent was firmly on the reduplicant in every category. A number of interesting facts emerge in the exploration of the Anatolian data. We see perhaps the emergence of new reduplicative categories in Hittite asās- / ases- and Lycian ppuwe-, while the distribution of Anatolian /i/-reduplication and the phonological history of /e/-reduplication has implications for the reconstruction of the Proto-Indo-European verbal system.

This dissertation provides the first unified account for the distribution and interpretation of wh-words in Hittite as indefinites, interrogatives and relativizers. Based on cross-linguistic comparanda, Hittite wh-words display the typical behavior of indefinite polarity items, and are prosodically deficient. As such, I argue that the surface positioning of wh-words in Hittite involves the syntax-semantics interface, and the syntax-phonology interface.

The non-standard word order of indefinites in Hittite is attributed to two factors. For an existential interpretation, Hittite wh-words must remain within the [vP] and are bound by the Rule of Existential Closure. For a presuppositional interpretation, indefinites undergo Quantifier Raising to IP. The final surface position of indefinites is determined by the syntax-phonology interface to satisfy prosodic restrictions: wh-words in Hittite are subject

to Prosodic Inversion at Spell Out.

As an interrogative, I argue that the wh-form consists of a phonologically null determiner with a [+wh] feature plus the Hittite wh-word kui-. Hittite wh-in situ is triggered by an intonational Q-morpheme, and is underspecified as [Q: ], enabling it to license both yes-no questions and wh-questions, as in modern French. Based on experimental evidence from living languages, I propose that the accentual nature of wh-words in Hittite interrogatives is the outcome of

the stress assignment by the intonational Q-morpheme.

Finally, I offer the first alternative analysis to Held’s (1957) account of Hittite correlative clauses. I demonstrate that preposed “indeterminate” correlatives represent wh-conditional clauses, as in Mandarin Chinese, Serbo-Croatian, and early Latin. The accented nature of the wh-word in wh-conditionals is derived from the stress assignment by contrastive Focus. I show that the peculiar distribution of the wh-word in “determinate” preposed correlatives

resembles that of indefinites: the wh-word is bound by the rule of Existential Closure and may be subject to Prosodic Inversion at Spell Out. I show that postposed correlatives may be restrictive or non-restrictive, and that Middle Hittite and New Hittite texts also exhibit embedded relative clauses.