Hans Weber. 1995. LR-inkrementelles probabilistisches Chartparsing von Worthypothesengraphen mit Unikationsgrammatiken: Eine enge Kopplung von Suche und Analyse. VerbmobilReport 52, Universitt Erlangen-Nrnberg.  Home/Search   Document Not in Database   Summary   Related Articles   Check

...., where t 0 62 Trees(G) t 0 2 Trees(G 0 ) This task was performed best using a context sensitive variant of Fujisaki s 5 unsupervised PCFG 3 . which is the formal language metaphor for lattice parsing with context free grammars. 4 Direct training was carried out in the style of [Weber 1994a] 5 [Fujisaki et al. 1991] training procedure. This training procedure is essentially exponential, but on the other hand the well known inside outside procedure could not be used since a grammar model with more than one rule of context was needed to characterize a tree set which was ....

....with pruning on an agenda in every chart vertex, which results in a beam search. Agendas in each vertex keep pairs of chart edges to be processed next. Those pairs are being scored using a linear combination of the acoustic, N Gram, and grammar model scores. It is described in more detail in [Weber 1994a] and [Gorz et al. 1996] The grammar trees which are found in that parsing step are passed to a second processing step using the original grammar G. In that step the tree found is re parsed deterministically using unification. If the approximation of G by the conjunction of G 0 and the grammar ....

[Article contains additional citation context not shown here]

Weber, Hans H. 1994a. LR-inkrementelles probabilistisches Chartparsing von Worthypothesenmengen mit Unifikationsgrammatiken: Eine enge Kopplung von Suche und Analyse. Ph.D. Thesis, Universitat Hamburg, FB Informatik, Dezember.

....loops occuring in the system. The purpose of this section is not to argue against interactivity in Verbmobil. Rather, we want to point out the complexities involved, so that algorithms fit for interactive architectures are chosen early enough in the respective development phases. See Weber [24] for a control algorithm for a loosely coupled and therefore parallel WR CP system. 12 The reason to downgrade the value of the information is that the parser, having received further word hypotheses in the meantime, might also have changed directions. 6 Conclusion The aim of this paper was ....

Hans H. Weber. LR-inkrementelles probabilistisches Chartparsing von Worthypothesenmengen mit Unifikationsgrammatiken: Eine enge Kopplung von Suche und Analyse. PhD thesis, Submitted to Universitat Hamburg, 1994.

....a mapping of hypotheses belonging to the same word copy is performed during parsing. 1 The ideas which led to this work are based, besides others, on works like Shieber 85 [19] Gorz 88 [9] Ney 93 [16] Paseler 88 [17] Fujisaki et al. 91 [8] For a more detailled description see Weber 94 [20]. Some of the work concerning the coupling with the beam decoder is a joint work with Andreas Hauenstein 1 and has already been published in Hauenstein Weber 94 [10, 11] 2 LR ACP with Beamsearch on the Agenda The use of active chart parsers and extensions of these in parsing speech has ....

....the computation of a set of predictions, based on the chart up to vertex I: 23 Definition 5.5 Computing predictions 1. Let ACTIVE be the nonempty active edges ending in vertex I , P I = 23 A broad discussion of efficiency in chart based computation of predictions can be found in Weber 94 [20]. 2. Take the n best successors s of any w = last(A.intro) A 2 ACTIVE, maximizing logP(sjw; n gram) 3. Create vertex GammaI and insert inactive edges for all s into vertex I , vertex GammaI , thus filling agenda GammaI . 4. Until below beam(agenda GammaI ) let (A,S) be pop(agenda GammaI ....

[Article contains additional citation context not shown here]

Hans H. Weber. LR-inkrementelles probabilistisches Chartparsing von Worthypothesenmengen mit Unifikationsgrammatiken: Eine enge Kopplung von Suche und Analyse. PhD thesis, Submitted to Universitat Hamburg, FB Informatik, Dezember 1994.

....r 0 in G 0 there is a corresponding rule r in G and vice versa. 3. Use an unsupervised reestimation procedure to train G 0 on B (context sensitive statistics) The syntactic parser (SynParser) is basically an incremental probabilistic search engine based on [20] for earlier versions cf. [18, 19]) it receives word hypotheses and phrase boundary hypotheses as input. The input is represented as a chart where frames correspond to chart vertices and word hypotheses are edges which map to pairs of vertices. Word boundary hypotheses (WBHs) are mapped to connected sequences of vertices which ....

....So we use an approximation of an HPSG Grammar for search such that syntactic analysis becomes more or less a second decoding step. By regarding a wider context, we even reduce the recognition gap between syntax and semantics in comparison with our previous unification based syntax parser (see [18, 19]) For practical usability the tree recognition rate must be improved. This can be achieved with a bigger training set. The dialogues we used contained only 83 utterances. Further improvement can be achieved by a larger context during training to get a better approximation of the trees built by ....

Weber, H.: LR-inkrementelles probabilistisches Chartparsing von Worthypothesenmengen mit Unifikationsgrammatiken: Eine enge Kopplung von Suche und Analyse. Ph.D. Thesis, University of Hamburg, 1995, Verbmobil Report 52.

No context found.

Hans Weber. 1995. LR-inkrementelles probabilistisches Chartparsing von Worthypothesengraphen mit Unikationsgrammatiken: Eine enge Kopplung von Suche und Analyse. VerbmobilReport 52, Universitt Erlangen-Nrnberg.

No context found.

Hans Weber. 1995. LR-inkrementelles probabilistisches Chartparsing von Worthypothesengraphen mit Unikationsgrammatiken: Eine enge Kopplung von Suche und Analyse. VerbmobilReport 52, Universitt Erlangen-Nrnberg.

No context found.

Hans Weber. 1995. LR-inkrementelles probabilistisches Chartparsing von Worthypothesengraphen mit Unikationsgrammatiken: Eine enge Kopplung von Suche und Analyse. VerbmobilReport 52, Universitt Erlangen-Nrnberg.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC