This paper evaluates frequency and detection performance for both spelling and grammatical errors in a corpus of published Danish newspaper texts, comparing the results of three human proofreaders with those of an automatic system, DanProof. Adopting the error categorization scheme of the latter, we look at the accuracy of individual error types and their relative distribution over time, as well as the adequacy of suggested corrections. Finally, we discuss so-called artefact errors introduced by corpus processing, and the potential of DanProof as a corpus cleaning tool for identifying and correcting format conversion, OCR or other compilation errors. In the evaluation, with balanced F1-scores of 77.6 and 67.6 for 1999 texts and 2019 texts, respectively, DanProof achieved a higher recall and accuracy than the individual human annotators, and contributed the largest share of errors not detected by others (16.4% for 1999 and 23.6% for 2019). However, the human annotators had a significantly higher precision. Not counting artifacts, the overall error frequency in the corpus was low ( 0.5%), and less than half in the newer texts compared to the older ones, a change that mostly concerned orthographical errors, with a correspondingly higher relative share of grammatical errors.

This paper describes and evaluates a rule-based system implementing a novel method for quote attribution in Portuguese text, working on top of a Constraint-Grammar parse. Both direct and indirect speech are covered, as well as certain other text- embedded quote sources. In a first step, the system performs quote segmentation and identifies speech verbs, taking into account the different styles used in literature and news text. Speakers are then identified using syntactically and semantically grounded Constraint-Grammar rules. We rely on relational links and stream variables to handle anaphorical mentions and to recover the names of implied or underspecified speakers. In an evaluation including both literature and news text, the system performed well on both the segmentation and attribution tasks, achieving F-scores of 98-99% for the former and 89-94% for the latter.

This paper presents PFN-DE, a new, parsing- and annotation-oriented framenet for German, with almost 15,000 frames, covering 11,300 verb lemmas. The resource was developed in the context of a Danish/German social-media study on hate speech and has a strong focus on coverage, robustness and cross-language comparability. A simple annotation scheme for argument roles meshes directly with the output of a syntactic parser, facilitating frame disambiguation through slot-filler conditions based on valency, syntactic function and semantic noun class. We discuss design principles for the framenet and the frame tagger using it, and present statistics for frame and role distribution at both the lexicon (type) and corpus (token) levels. In an evaluation run on Twitter data, the parser-based frame annotator achieved an overall F-score for frame senses of 93.6%.

In this paper we describe and evaluate syntactic and semantic aspects of Arbobanko, a treebank for the artificial language Esperanto, as well as tools and methods used in the production of the treebank. In addition to classical morphosyntax and dependency structure, the treebank was enriched with a lexical-semantic layer covering named entities, a semantic type ontology for nouns and adjectives and a framenet-inspired semantic classification of verbs. For an under-resourced language, the quality of automatic syntactic and semantic pre-annotation is of obvious importance, and by evaluating the underlying parser and the coverage of its semantic ontologies, we try to answer the question whether the language’s extremely regular morphology and transparent semantic affixes translate into a more regular syntax and higher parsing accuracy. On the linguistic side, the treebank allows us to address and quantify typological issues such as the question of word order, auxiliary constructions, lexical transparency and semantic type ambiguity in Esperanto.

This paper presents the German Twitter section of a large (2 billion word) bilingual Social Media corpus for Hate Speech research, discussing the compilation, pseudonymization and grammatical annotation of the corpus, as well as special linguistic features and peculiarities encountered in the data. Among other things, compounding, accidental and intentional orthographic variation, gendering and the use of emoticons/emojis are addressed in a genre-specific fashion. We present the different layers of linguistic annotation (morphosyntactic, dependencies and semantic types) and explain how a general parser (GerGram) can be made to work on Social Media data, pointing out necessary adaptations and extensions. In an evaluation run on a random cross-section of tweets, the modified parser achieved F-scores of 97% for morphology (fine-grained POS) and 92% for syntax (labeled attachment score). Predictably, performance was twice as good in tweets with standard orthography than in tweets with spelling/casing irregularities or lack of sentence separation, the effect being more marked for morphology than for syntax.

This paper presents a semantic annotation scheme for Danish adjectives, focusing both on prototypical semantic content and semantic collocational restrictions on an adjective’s head noun. The core type set comprises about 110 categories ordered in a shallow hierarchy with 14 primary and 25 secondary umbrella categories. In addition, domain information and binary sentiment tags are provided, as well as VerbNet-derived frames and semantic roles for those adjectives governing arguments. The scheme has been almost fully implemented on the lexicon of the Danish VISL parser, DanGram, containing 14,000 adjectives. We discuss the annotation scheme and its applicational perspectives, and present a statistical breakdown and coverage evaluation for three Danish reference corpora.

This paper discusses the internal structure of complex Esperanto words (CWs). Using a morphological analyzer, possible affixation and compounding is checked for over 50,000 Esperanto lexemes against a list of 17,000 root words. Morpheme boundaries in the resulting analyses were then checked manually, creating a CW dictionary of 28,000 words, representing 56.4% of the lexicon, or 19.4% of corpus tokens. The error percentage of the EspGram morphological analyzer for new corpus CWs was 4.3% for types and 6.4% for tokens, with a recall of almost 100%, and wrong/spurious boundaries being more common than missing ones. For pedagogical purposes a morpheme frequency dictionary was constructed for a 16 million word corpus, confirming the importance of agglutinative derivational morphemes in the Esperanto lexicon. Finally, as a means to reduce the morphological ambiguity of CWs, we provide POS likelihoods for Esperanto suffixes.

In this paper, we describe how a Constraint Grammar with linguist-written rules can be optimized and ported to another language using a Machine Learning technique. The effects of rule movements, sorting, grammar-sectioning and systematic rule modifications are discussed and quantitatively evaluated. Statistical information is used to provide a baseline and to enhance the core of manual rules. The best-performing parameter combinations achieved part-of-speech F-scores of over 92 for a grammar ported from English to Danish, a considerable advance over both the statistical baseline (85.7), and the raw ported grammar (86.1). When the same technique was applied to an existing native Danish CG, error reduction was 10% (F=96.94).

This article describes the morphosyntactic annotation of the C-ORAL-BRASIL speech corpus, using an adapted version of the Palavras parser. In order to achieve compatibility with annotation rules designed for standard written Portuguese, transcribed words were orthographically normalized, and the parsing lexicon augmented with speech-specific material, phonetically spelled abbreviations etc. Using a two-level annotation approach, speech flow markers like overlaps, retractions and non-verbal productions were separated from running, annotatable text. In the absence of punctuation, syntactic segmentation was achieved by exploiting prosodic break markers, enhanced by a rule-based distinctions between pause and break functions. Under optimal conditions, the modified parsing system achieved correctness rates (F-scores) of 98.6% for part of speech, 95% for syntactic function and 99% for lemmatization. Especially at the syntactic level, a clear connection between accessibility of prosodic break markers and annotation performance could be documented.

This paper describes a hybrid system (FrAG) for tagging / parsing French text, and presents results from ongoing development work, corpus annotation and evaluation. The core of the system is a sentence scope Constraint Grammar (CG), with linguist-written rules. However, unlike traditional CG, the system uses hybrid techniques on both its morphological input side and its syntactic output side. Thus, FrAG draws on a pre-existing probabilistic Decision Tree Tagger (DTT) before and in parallel with its own lexical stage, and feeds its output into a Phrase Structure Grammar (PSG) that uses CG syntactic function tags rather than ordinary terminals in its rewriting rules. As an alternative architecture, dependency tree structures are also supported. In the newest version, dependencies are assigned within the CG-framework itself, and can interact with other rules. To provide semantic context, a semantic prototype ontology for nouns is used, covering a large part of the lexicon. In a recent test run on Parliamentary debate transcripts, FrAG achieved F-scores of 98.7 % for part of speech (PoS) and between 93.1 % and 96.2 % for syntactic function tags. Dependency links were correct in 95.9 %.

In this paper, we present and evaluate a new method to convert Constraint Grammar (CG) parses of running text into Constituent Treebanks. The conversion is two-step - first a grammar-based method is used to bridge the gap between raw CG annotation and full dependency structure, then phrase structure bracketing and non-terminal nodes are introduced by clustering sister dependents, effectively building one syntactic treebank on top of another. The method is compared with another approach (Bick 2003-2), where constituent structures are arrived at by employing a function-tag based Phrase Structure Grammar (PSG). Results are evaluated on a small reference corpus for both raw and revised CG input, with bracketing F-Scores of 87.5% for raw text and 97.1% for revised CG input, and a raw text edge label accuracy of 95.9% for forms and 86% for functions, or 99.7% and 99.4%, respectively, for revised CG. By applying the tools to the CG-only part of the Danish Arboretum treebank we were able to increase the size of the treebank by 86%, from 197.400 to 367.500 words.

Les corpus français librement accessibles annotés à d’autres niveaux linguistiques que morpho-syntaxique sont insuffisants à la fois quantitativement et qualitativement. Partant de ce constat, la FREEBANK – construite sur la base d’outils d’analyse automatique dont la sortie est révisée manuellement – se veut une base de corpus du français annotés à plusieurs niveaux (structurel, morphologique, syntaxique, coréférentiel) et à différents degrés de finesse linguistique qui soit libre d’accès, codée selon des schémas normalisés, intégrant des ressources existantes et ouverte à l’enrichissement progressif.