This paper describes our contribution to the semantic role labeling task (SRL-only) of the CoNLL-2009 shared task in the closed challenge (Hajič et al., 2009). Our system consists of a pipeline of independent, local classifiers that identify the predicate sense, the arguments of the predicates, and the argument labels. Using these local models, we carried out a beam search to generate a pool of candidates. We then reranked the candidates using a joint learning approach that combines the local models and proposition features. To address the multilingual nature of the data, we implemented a feature selection procedure that systematically explored the feature space, yielding significant gains over a standard set of features. Our system achieved the second best semantic score overall with an average labeled semantic F1 of 80.31. It obtained the best F1 score on the Chinese and German data and the second best one on English.

We investigate different ways of learning structured perceptron models for coreference resolution when using non-local features and beam search. Our experimental results indicate that standard techniques such as early updates or Learning as Search Optimization (LaSO) perform worse than a greedy baseline that only uses local features. By modifying LaSO to delay updates until the end of each instance we obtain significant improvements over the baseline. Our model obtains the best results to date on recent shared task data for Arabic, Chinese, and English.

This paper presents the IMS contribution to the CoNLL 2017 Shared Task. In the preprocessing step we employed a CRF POS/morphological tagger and a neural tagger predicting supertags. On some languages, we also applied word segmentation with the CRF tagger and sentence segmentation with a perceptron-based parser. For parsing we took an ensemble approach by blending multiple instances of three parsers with very different architectures. Our system achieved the third place overall and the second place for the surprise languages.

Robots used in manufacturing today are tailored to their tasks by system integration based on expert knowledge concerning both production and machine control. For upcoming new generations of even more flexible robot solutions, in applications such as dexterous assembly, the robot setup and programming gets even more challenging. Reuse of solutions in terms of parameters, controls, process tuning, and of software modules in general then gets increasingly important.There has been valuable progress within reuse of automation solutions when machines comply with standards and behave according to nominal models. However, more flexible robots with sensor-based manipulation skills and cognitive functions for human interaction are far too complex to manage, and solutions are rarely reusable since knowledge is either implicit in imperative software or not captured in machine readable form.We propose techniques that build on existing knowledge by converting structured data into an RDF-based knowledge base. By enhancements of industrial control systems and available engineering tools, such knowledge can be gradually extended as part of the interaction during the definition of the robot task.

Supertagging was recently proposed to provide syntactic features for statistical dependency parsing, contrary to its traditional use as a disambiguation step. We conduct a broad range of controlled experiments to compare this specific application of supertagging with another method for providing syntactic features, namely stacking. We find that in this context supertagging is a form of stacking. We furthermore show that (i) a fast parser and a sequence labeler are equally beneficial in supertagging, (ii) supertagging/stacking improve parsing also in a cross-domain setting, and (iii) there are small gains when combining supertagging and stacking, but only if both methods use different tools. The important consideration is therefore not the method but rather the diversity of the tools involved.

We study non-deterministic oracles for training non-projective beam search parsers with swap transitions. We map out the spurious ambiguities of the transition system and present two non-deterministic oracles as well as a static oracle that minimizes the number of swaps. An evaluation on 10 treebanks reveals that the difference between static and non-deterministic oracles is generally insignificant for beam search parsers but that non-deterministic oracles can improve the accuracy of greedy parsers that use swap transitions.

Abstract:Model-based systems in control are a means to utilize efficiently human knowledge and achieve high performance. While models consisting of formalized knowledge are used during the engineering step, running systems usually do not contain a high-level, symbolic representation of the control and most of its properties, typically named numerical parameters. On a system level and beyond the plant data, there is also a need to represent the meaning of the data such that deployment and fault analysis could be augmented with partly automated inference based on the semantics of the data. To that end, we extended the formalized knowledge traditionally used in control to include the control purpose, engineering assumption, quality, involved state machines, and so on. We then represented the control semantics in a format that allows an easier extraction of information using querying and reasoning. It aims at making knowledge in control engineering reusable so that it can be shipped together with the control systems. We implemented prototypes that include automatic conversion of plant data from AutomationML into RDF triples, as well as the automated extraction of control properties, the conversion of parameters, and their storage in the same triple store. Although these techniques are standard within the semantic web community, we believe that our robotic prototypes for semantic control represent a novel approach.

We present the ICARUS Coreference Explorer, an interactive tool to browse and search coreference-annotated data. It can display coreference annotations as a tree, as an entity grid, or in a standard textbased display mode, and lets the user switch freely between the different modes. The tool can compare two different annotations on the same document, allowing system developers to evaluate errors in automatic system predictions. It features a flexible search engine, which enables the user to graphically construct search queries over sets of documents annotated with coreference.