Structural and lexical factors in adjective placement in complex noun phrases across Romance languages

Proceedings of the 19th Conference on Computational Language Learning, pages 247–257, Beijing, China, July 30-31, 2015. c⃝2015 Association for Computational Linguistics Structural and lexical factors in adjective placement in complex noun phrases across Romance languages Kristina Gulordava University of Geneva Kristina.Gulordava@unige.ch Paola Merlo University of Geneva Paola.Merlo@unige.ch Abstract One of the most common features across all known languages is their variability in word order. We show that differences in the prenominal and postnominal place- ment of adjectives in the noun phrase across five main Romance languages is not only subject to heaviness effects, as previously reported, but also to subtler structural interactions among dependen- cies that are better explained as effects of the principle of dependency length min- imisation. These effects are almost purely structural and show lexical conditioning only in highly frequent collocations. 1 Introduction One of the most widely observed characteristics of all languages is the variability in the linear or- der of their words, both across and within a single language. In this study, we concentrate on word order alternations where one structure can be lin- earised in two different ways. Consider, for exam- ple, the case when a phrasal verb (V + particle) has a direct object (NP), in English. Two alter- native orders are possible: VP1 = V NP Prt, and VP2 = V Prt NP. If the NP is heavy, as defined in number of words or number of syllables, it will be frequently placed after the Prt, yielding the V-Prt- NP order. Compare, for instance Call me up! to Call up the customer who called yesterday. This tendency is also formulated as a Principle of End Weight, where phrases are presented in order of increasing weight (Wasow, 2002). Cases of heavy NP-shift (Stallings et al., 1998), dative alternation (Bresnan et al., 2007) and other alternation pref- erences among verbal dependents are traditionally evoked to argue in favour of the “heaviness” effect. In this work, we study the alternations in the noun-phrase domain, much less investigated in Figure 1: Percent of postnominal simple (green) and heavy (red) adjectives across seventeen lan- guages. connection with the heaviness effect. Abeill´e and Godard (2000) introduce the heaviness of adjec- tive phrases as a principle explaining their post- nominal placement compared to ‘light’ adjectives in French. Their observations have been recently confirmed in a corpus study by Thuilier (2012). Cross-linguistically, the data that we have col- lected across many languages and several families, presented in Figure 1, confirm the heaviness effect for adjectives1. By extracting relevant statistics from gold dependency annotated corpora, we can observe that heavy adjectives (adjective phrases of at least two words) appear more frequently post- nominally than simple adjectives. While the effect of size or heaviness is well- documented, this statistics is very coarse and it confounds various linguistic factors, such as types 1We use the following languages and treebanks: English, Czech, Spanish, Chinese, Catalan, German, Italian (Hajiˇc et al., 2009), Danish, Dutch, Portuguese, Swedish (Buchholz and Marsi, 2006), Latin, Ancient Greek (Haug and …

Content truncated to 500 words