430 Germanische Sprachen; Deutsch
Refine
Year of publication
- 2009 (2) (remove)
Document Type
- Doctoral Thesis (2) (remove)
Language
- English (2)
Has Fulltext
- yes (2) (remove)
Is part of the Bibliography
- no (2)
Keywords
- Assimilation <Phonetik> (1)
- Deletion (1)
- Distinktives Merkmal (1)
- Exemplarmodelle (1)
- FUL-Modell (1)
- German (1)
- Korpusstudie (1)
- LFG (1)
- Modellvergleich (1)
- PCFG (1)
Institute
Manual development of deep linguistic resources is time-consuming and costly and therefore often described as a bottleneck for traditional rule-based NLP. In my PhD thesis I present a treebank-based method for the automatic acquisition of LFG resources for German. The method automatically creates deep and rich linguistic presentations from labelled data (treebanks) and can be applied to large data sets. My research is based on and substantially extends previous work on automatically acquiring wide-coverage, deep, constraint-based grammatical resources from the English Penn-II treebank (Cahill et al.,2002; Burke et al., 2004; Cahill, 2004). Best results for English show a dependency f-score of 82.73% (Cahill et al., 2008) against the PARC 700 dependency bank, outperforming the best hand-crafted grammar of Kaplan et al. (2004). Preliminary work has been carried out to test the approach on languages other than English, providing proof of concept for the applicability of the method (Cahill et al., 2003; Cahill, 2004; Cahill et al., 2005). While first results have been promising, a number of important research questions have been raised. The original approach presented first in Cahill et al. (2002) is strongly tailored to English and the datastructures provided by the Penn-II treebank (Marcus et al., 1993). English is configurational and rather poor in inflectional forms. German, by contrast, features semi-free word order and a much richer morphology. Furthermore, treebanks for German differ considerably from the Penn-II treebank as regards data structures and encoding schemes underlying the grammar acquisition task. In my thesis I examine the impact of language-specific properties of German as well as linguistically motivated treebank design decisions on PCFG parsing and LFG grammar acquisition. I present experiments investigating the influence of treebank design on PCFG parsing and show which type of representations are useful for the PCFG and LFG grammar acquisition tasks. Furthermore, I present a novel approach to cross-treebank comparison, measuring the effect of controlled error insertion on treebank trees and parser output from different treebanks. I complement the cross-treebank comparison by providing a human evaluation using TePaCoC, a new testsuite for testing parser performance on complex grammatical constructions. Manual evaluation on TePaCoC data provides new insights on the impact of flat vs. hierarchical annotation schemes on data-driven parsing. I present treebank-based LFG acquisition methodologies for two German treebanks. An extensive evaluation along different dimensions complements the investigation and provides valuable insights for the future development of treebanks.
Reduction in natural speech
(2009)
Natural (conversational) speech, compared to cannonical speech, is earmarked by the tremendous amount of variation that often leads to a massive change in pronunciation. Despite many attempts to explain and theorize the variability in conversational speech, its unique characteristics have not played a significant role in linguistic modeling. One of the reasons for variation in natural speech lies in a tendency of speakers to reduce speech, which may drastically alter the phonetic shape of words. Despite the massive loss of information due to reduction, listeners are often able to understand conversational speech even in the presence of background noise. This dissertation investigates two reduction processes, namely regressive place assimilation across word boundaries, and massive reduction and provides novel data from the analyses of speech corpora combined with experimental results from perception studies to reach a better understanding of how humans handle natural speech. The successes and failures of two models dealing with data from natural speech are presented: The FUL-model (Featurally Underspecified Lexicon, Lahiri & Reetz, 2002), and X-MOD (an episodic model, Johnson, 1997). Based on different assumptions, both models make different predictions for the two types of reduction processes under investigation. This dissertation explores the nature and dynamics of these processes in speech production and discusses its consequences for speech perception. More specifically, data from analyses of running speech are presented investigating the amount of reduction that occurs in naturally spoken German. Concerning production, the corpus analysis of regressive place assimilation reveals that it is not an obligatory process. At the same time, there emerges a clear asymmetry: With only very few exceptions, only [coronal] segments undergo assimilation, [labial] and [dorsal] segments usually do not. Furthermore, there seem to be cases of complete neutralization where the underlying Place of Articulation feature has undergone complete assimilation to the Place of Articulation feature of the upcoming segment. Phonetic analyses further underpin these findings. Concerning deletions and massive reductions, the results clearly indicate that phonological rules in the classical generative tradition are not able to explain the reduction patterns attested in conversational speech. Overall, the analyses of deletion and massive reduction in natural speech did not exhibit clear-cut patterns. For a more in-depth examination of reduction factors, the case of final /t/ deletion is examined by means of a new corpus constructed for this purpose. The analysis of this corpus indicates that although phonological context plays an important role on the deletion of segments (i.e. /t/), this arises in the form of tendencies, not absolute conditions. This is true for other deletion processes, too. Concerning speech perception, a crucial part for both models under investigation (X-MOD and FUL) is how listeners handle reduced speech. Five experiments investigate the way reduced speech is perceived by human listeners. Results from two experiments show that regressive place assimilations can be treated as instances of complete neutralizations by German listeners. Concerning massively reduced words, the outcome of transcription and priming experiments suggest that such words are not acceptable candidates of the intended lexical items for listeners in the absence of their proper phrasal context. Overall, the abstractionist FUL-model is found to be superior in explaining the data. While at first sight, X-MOD deals with the production data more readily, FUL provides a better fit for the perception results. Another important finding concerns the role of phonology and phonetics in general. The results presented in this dissertation make a strong case for models, such as FUL, where phonology and phonetics operate at different levels of the mental lexicon, rather than being integrated into one. The findings suggest that phonetic variation is not part of the representation in the mental lexicon.