A Semantic-Based Approach to Noun-Noun Compound Interpretation

A Semantic-Based Approach to Noun-Noun Compound Interpretation

In this paper, we propose a method to identify the possible readings of Chinese noun-noun compounds (NNCs).To avoid problems such as vagueness of interpretation, limited or sporadic coverage, and arbitrariness of semantic relation classification, we considereda large number ofnoun-noun compounds from our Prefix-Suffix System and identified their semantic relations with two semantic networks: FrameNet (https://framenet.icsi.berkeley.edu/fndrupal/home) and E-HowNet(http://ehownet.iis.sinica.edu.tw/ehownet.php). We found that N1 and N2 are either linked by semantic roles assigned by events (complex relations) or by static relations (simple relations) including meronymy, conjunction, and the host-attribute-value relation. Furthermore, for both types of relations, the possible readings of the resulting compoundare limited enough for computational implementation. Regarding simple relations, conjunctionhas limited productivity; meanwhile, the two components usually belong to the same semantic type, such as鐘錶zhong-biao‘clock and watch.'The limited productivity makes it possible for E-HowNet to include a large proportion of such pairs in its dictionary, while similar semantic types can be identified through its taxonomy. Likewise, the host-attribute-value combinationsarealso a type of simple relations which involves combinations of hosts, attributes, and values in certain orders, such as Value-Host (e.g.鐵桌tie-zhuo‘iron table/desk'), Host-Attribute (e.g.車速che-su‘car speed'), and Value-Attribute (e.g.法式fa-shi‘French-style'). Themeronymic relations are the third type of simple relations, which are annotated by semantic roles like‘part of'and‘whole'(Part-Whole: e.g.雙底船shuang-dichuan‘double-bottom'; Whole-Part: e.g.車輪che-lun‘car tire').