Refine
Document Type
- Conference Proceeding (4)
- Part of a Book (2)
- Article (1)
Language
- English (7)
Has Fulltext
- yes (7)
Is part of the Bibliography
- no (7)
Keywords
- Deutsch (3)
- Syntaktische Analyse (3)
- Akustische Analyse (1)
- Automatische Sprachanalyse; (1)
- Brown clustering (1)
- Cluster <Datenanalyse> (1)
- Dativ (1)
- Englisch (1)
- Französisch (1)
- German (1)
Publicationstate
Reviewstate
- Peer-Review (7)
We present the IUCL system, based on supervised learning, for the shared task on stance detection. Our official submission, the random forest model, reaches a score of 63.60, and is ranked 6th out of 19 teams. We also use gradient boosting decision trees and SVM and merge all classifiers into an ensemble method. Our analysis shows that random forest is good at retrieving minority classes and gradient boosting majority classes. The strengths of different classifiers wrt. precision and recall complement each other in the ensemble.
We investigate whether non-configurational languages, which display more word order variation than configurational ones, require more training data for a phenomenon to be parsed successfully. We perform a tightly controlled study comparing the dative alternation for English (a configurational language), German, and Russian (both non-configurational). More specifically, we compare the performance of a dependency parser when only canonical word order is present with its performance on data sets when all word orders are present. Our results show that for all languages, canonical data not only is easier to parse, but there exists no direct correspondence between the size of training sets containing free(er) word order variation and performance.
“My Curiosity was Satisfied, but not in a Good Way”: Predicting User Ratings for Online Recipes
(2014)
In this paper, we develop an approach to automatically predict user ratings for recipes at Epicurious.com, based on the recipes’ reviews. We investigate two distributional methods for feature selection, Information Gain and Bi-Normal Separation; we also compare distributionally selected features to linguistically motivated features and two types of frameworks: a one-layer system where we aggregate all reviews and predict the rating vs. a two-layer system where ratings of individual reviews are predicted and then aggregated. We obtain our best results by using the two-layer architecture, in combination with 5 000 features selected by Information Gain. This setup reaches an overall accuracy of 65.60%, given an upper bound of 82.57%.
This study investigates high vowel laxing in the Louisiana French of the Lafourche Basin. Unlike Canadian French, in which the high vowels /i, y, u/ are traditionally described as undergoing laxing (to [I, Y, U]) in word-final syllables closed by any consonant other than a voiced fricative (see Poliquin 2006), Oukada (1977) states that in the Louisiana French of Lafourche Parish, any coda consonant will trigger high vowel laxing of /i/; he excludes both /y/ and /u/ from his discussion of high vowel laxing. The current study analyzes tokens of /i, y, u/ from pre-recorded interviews with three older male speakers from Terrebonne Parish. We measured the first and second formants and duration for high vowel tokens produced in four phonetic environments, crossing syllable type (open vs. closed) by consonant type (voiced fricative vs. any consonant other than a voiced fricative). Results of the acoustic analysis show optional laxing for /i/ and /y/ and corroborate the finding that high vowels undergo laxing in word-final closed syllables, regardless of consonant type. Data for /u/ show that the results vary widely by speaker, with the dominant pattern (shown by two out of three speakers) that of lowering and backing in the vowel space of closed syllable tokens. Duration data prove inconclusive, likely due to the effects of stress. The formant data published here constitute the first acoustic description of high vowels for any variety of Louisiana French and lay the groundwork for future study on these endangered varieties.
Many applications in Natural Language Processing require a semantic analysis of sentences in terms of truth-conditional representations, often with specific desiderata in terms of which information needs to be included in the semantic analysis. However, there are only very few tools that allow such an analysis. We investigate the representations of an automatic analysis pipeline of the C&C parser and Boxer to determine whether Boxer’s analyses in form of Discourse Representation Structure can be successfully converted into a more surface oriented event semantic representation, which will serve as input for a fusion algorithm for fusing hard and soft information. We use a data set of synthetic counter intelligence messages for our investigation. We provide a basic pipeline for conversion and subsequently discuss areas in which ambiguities and differences between the semantic representations present challenges in the conversion process.
We investigate how the granularity of POS tags influences POS tagging, and furthermore, how POS tagging performance relates to parsing results. For this, we use the standard “pipeline” approach, in which a parser builds its output on previously tagged input. The experiments are performed on two German treebanks, using three POS tagsets of different granularity, and six different POS taggers, together with the Berkeley parser. Our findings show that less granularity of the POS tagset leads to better tagging results. However, both too coarse-grained and too fine-grained distinctions on POS level decrease parsing performance.
Brown clustering has been used to help increase parsing performance for morphologically rich languages. However, much of the work has focused on using clustering techniques to replace terminal nodes or as a feature for parsing. Instead, we choose to examine how effectively Brown clustering is for unlexicalized parsing by creating data-driven POS tagsets which are then used with the Berkeley parser. We investigate cluster sizes as well as on what information (e.g. words vs. lemmas) clustering will yield the best parser performance. Our results approach the current state of the art results for the German T¨uBa-D/Z treebank when using parser internal tagging.