Output

• (2021) Mumtaz, B., Canzi, M., & Butt, M. Prosody of Case Markers in Urdu. Proc. Interspeech 2021

Abstract

This paper studies the prosody of case clitics in Urdu, for which various different claims exist in the literature. We conducted a production experiment and controlled for effects potentially arising from the phonetics of the case clitics, the syntactic function they express and clausal position. We find that case clitics are incorporated into the prosodic phrase of the noun and that they become part of the overall LH contour found on accentual phrases in Urdu/Hindi. We also find some differences across case type and position which we tie to information structural effects.

Introduction

As part of a project to develop more natural sounding speech for Text-to-Speech (TTS) for Urdu we are working on improving our basic linguistic understanding of the speech prosody of Urdu/Hindi. In this paper we focus on Urdu case markers in relation to determining boundaries of prosodic phrasing. The basic phrasal accent in Urdu/Hindi has been established to be some form of LH, where the H functions as a boundary tone for the accentual phrase [1, 2, 3, 4, 5]. We follow [4] and [5] in positioning L*H as the basic phrasal accent. In terms of prosodic phrasing, there seem to be at least two basic levels in Urdu/Hindi: i) accentual phrase (AP) and ii) intonational phrase [2, 4], but the size of the AP remains unclear, especially with respect to case markers, which have the morpho-phonological status of clitics [6]. Generally it seems that each content word receives an AP and each AP is characterized by an L*H with the H indicating the right boundary.

However, case markers have been claimed to have their own LH intonational contour [7] while still being phrased together with the AP of the noun they are attached to. If this is true, the question of how to reliably identify the right boundary of AP arises. Indeed, the data reported in the literature seems to show variation in the realization of pitch contours on the case markers [8, 9]. For example, take the ergative [ne] and the dative [ko] in the sentence ‘Rahul gave medicine to (his) mother.’ in Figure 1 from [10]. The pitch contour seems to peak with an H at the end of Rahul before the case marker [ne], while in the second NP, the H aligns with the case marker [ko] following ma. In our own work on developing an annotated speech corpus for Urdu TTS [11], we furthermore observed that the genitive NP in phrases such as ‘Rahul’s mother’ seems to carry an L*HL contour rather than the usual L*H. This raises the question whether genitives are not only syntactically different, but also prosodically different from other case marked NPs. On the other hand, the observed variation might also be due to artefacts caused by acoustic factors such as vowel height and degree of sonority (e.g., ergative [ne] vs. genitive [ki]).

We conducted a carefully controlled production experiment which varied the factors of case type, function and position and found that the AP high boundary tone aligns with the case markers, regardless of case type and NP position. This stands in contrast to the findings/observations of previous studies and indicates that the case clitics form part of the AP with the preceding noun. We did find an increased pitch excursion associated with more emphasized NPs and we also saw that the duration of the case markers varied with position in the clause. We present and discuss these findings in more detail below.