🧑‍🔬 Helmut Schmid

Helmut Schmid works in computational linguistics, on the tools that assign linguistic structure to text: part-of-speech tagging, morphological analysis and syntactic parsing. The papers here keep returning to one problem, which is scale. They deal with tagsets containing hundreds of fine-grained labels, with grammars extracted from treebanks that allow vast numbers of analyses, and with word formation that produces open-ended vocabularies. The methods combine probabilistic estimation, finite-state transducers and data structures designed for speed. German is the main test language, and Czech appears as well.

The most cited of these papers is Estimation of Conditional Probabilities With Decision Trees. It handles HMM tagging with very large tagsets by predicting each feature of a composite tag one after another, and it lets decision trees select the relevant context attributes in place of a hand-designed backoff order. SMOR describes a finite-state analyser for German that treats inflection, derivation and compounding in a single system. In it, affixes are stored as lexicon entries whose features limit the bases they may attach to. The BitPar paper presents a CKY-style parser that keeps its chart as bit arrays and joins constituents with bitwise operations. This makes heavily ambiguous treebank grammars tractable, even when every analysis is wanted rather than only the best one.

The newest paper here, from 2008, is the decision-tree tagging study. It carries the earlier interest in rich German morphology into statistical tagging, treating morphological features as the internal structure of each tag. The two 2004 papers on morphology and syntax and grammar formalisms came before it. Together the three papers span the main topics of the work, alongside part-of-speech tagging.

Schmid is based at Ludwig-Maximilians-Universität München. Recent bylines also name the Munich Center for Machine Learning and Technische Universität Darmstadt. Earlier papers came from the University of Stuttgart. Co-authors on the papers here include Florian Laws.

Helmut Schmid has 3 papers in our library, in 2004–2008, most of them on Part-of-speech tagging, Morphology, Syntax & grammar formalisms; often with Florian Laws.