Distinguishing affixoid formations from compounds

Proceedings of the 27th International Conference on Computational Linguistics, pages 3853–3865 Santa Fe, New Mexico, USA, August 20-26, 2018. 3853 Distinguishing affixoid formations from compounds Josef Ruppenhofer LeibnizScienceCampus LiMo Institute for German Language Mannheim ruppenhofer@ids-mannheim.de Michael Wiegand Spoken Language Systems Saarland University michael.wiegand @lsv.uni-saarland.de Rebecca Wilm Katja Markert LeibnizScienceCampus LiMo Computational Linguistics Heidelberg University {wilm,markert} @cl.uni-heidelberg.de Abstract We study German affixoids, a type of morpheme in between affixes and free stems. Several prop- erties have been associated with them – increased productivity; a bleached semantics, which is often evaluative and/or intensifying and thus of relevance to sentiment analysis; and the existence of a free morpheme counterpart – but not been validated empirically. In experiments on a new data set that we make available, we put these key assumptions from the morphological literature to the test and show that despite the fact that affixoids generate many low-frequency formations, we can classify these as affixoid or non-affixoid instances with a best F1-score of 74%. 1 Introduction In this work, we study a subset of complex German words that includes many hapaxes, namely items that have as one of their morphological components a so-called affixoid or semi-affix. Examples of affixoid formations include Gesetzeshengst ‘jobsworth, stickler for the letter of the law’ [lit. ‘law/legal stallion’] or Mentalit¨atsmonster ‘person with laser focus’ [lit. ‘mentality monster’]. The class of affixoid mor- phemes sits in between affixes and stems. While various criteria have been proposed to identify affixoids (Schmidt, 1987), at least the following three are widely taken for granted (ten Hacken, 2000; Elsen, 2009): (i) increased productivity; (ii) semantic bleaching/decreased semantic specificity; and (iii) an et- ymological and formal link to an existing free stem. The first two criteria are applied by comparing the affixoid to its corresponding free stem. For instance, in Weingott ‘deity of wine’, Gott is a free morpheme occurring with its regular meaning; in Gitarrengott ‘guitar god’, Gott is an affixoid: the whole word does not refer to a deity but to a talented human guitar player. Such evaluative or intensifying meanings are a hallmark of affixoids (Meibauer, 2013). The last of the three criteria distinguishes affixoids from affixes, which by definition occur only bound to other morphemes. For example, German -heit, whose English cognate is -hood as in falsehood, is an affix: there is no longer a related free form. By contrast, German -gott is an affixoid since there is a related free form Gott, cognate with English god. Research on affixoids has been centered on Germanic languages like German, Dutch and Swedish (Ascoop and Leuschner, 2006; Booij, 2005; Booij and H¨uning, 2014; Norde and Van Goethem, 2014). However, we believe affixoids are not an exclusive feature of these languages. They are likely to arise in other languages with productive compounding. For English, for instance, there is little to no systematic research but arguably quality (as in quality press/furniture/diamonds but not in quality management) and nut (as in health/math/trivia nut but not in pecan nut) can be considered English affixoids. Even …

Content truncated to 500 words