Compilation, transcription and usage of a reference speech corpus: the case of the Slovene corpus GOS

CLARIN Web Services for TEI-annotated Transcripts of Spoken Language

“…Besides Schmidt, Hedeland and Jettka (2017) and the ISO specification itself (ISO 2016), the role of TEI as a suitable basis of a standard for spoken language transcription has been discussed, among others, by Schmidt (2011) and Liégeois et al (2017). The TEI guidelines' chapter 8 on "Transcriptions of Speech" (TEI Consortium 2019) has also been used in CLARIN resources such as the GOS Corpus of Spoken Slovene (see Verdonik et al 2013) and as the basis for a CLARIN-wide format for parliamentary data. 3…”

Er ... well, it matters, right? On the role of data representations in spoken language dependency parsing

“…The SST treebank currently amounts to 29,488 tokens (3,188 utterances), which include both lexical tokens (words) and tokens signalling other types of verbal phenomena, such as filled pauses (fillers) and unfinished words, as well as some basic markers of prosody and extralinguistic speech events. The original segmentation, tokenization and spelling principles described by Verdonik et al (2013) have also been inherited by SST. Among the two types of Gos transcriptions (pronunciation-based and normalized spelling, both in lowercase only), subsequent manual annotations in SST have been performed on top of normalized transcriptions.…”

“…Segmentation: Inheriting the manual segmentation of the reference Gos corpus, sentences (utterances) in SST correspond to "semantically, syntactically and acoustically delimited units" (Verdonik et al, 2013). As such, the utterance segmentation heavily depends on subjective interpretations of what is the basic functional unit in speech, in line with the multitude of existing segmentation approaches, based on syntax, semantics, prosody, or their various combinations (Degand and Simon, 2009).…”

The Universal Dependencies Treebank for Slovenian

“…Typically, spoken language annotation denotes annotation of its representation in the form of written transcription. In the Spoken Slovenian Treebank, the spelling, tokenization and segmentation principles follow the transcription guidelines of the reference Gos corpus (Verdonik et al, 2013). The syntactic trees in the treebank span over individual utterances, manually delimited in the process of reference corpus transcription.…”