Compound or Term Features? Analyzing Salience in Predicting the Difficulty of German Noun Compounds across Domains

Abstract

Predicting the difficulty of domain-specific vocabulary is an important task towards a better understanding of a domain, and to enhance the communication between lay people and experts. We investigate German closed noun compounds and focus on the interaction of compound-based lexical features (such as frequency and productivity) and terminologybased features (contrasting domain-specific and general language) across word representations and classifiers. Our prediction experiments complement insights from classification using (a) manually designed features to characterise termhood and compound formation and (b) compound and constituent word embeddings. We find that for a broad binary distinction into easy vs. difficult general-language compound frequency is sufficient, but for a more fine-grained four-class distinction it is crucial to include contrastive termhood features and compound and constituent features.

Made of Steel? Learning Plausible Materials for Components in the Vehicle Repair Domain

“…(Domain-specific) MWEs and Compositionality Multiword expressions (MWEs) are challenging for any natural understanding system, given that MWE meanings are idiosyncratic to some degree, i.e., the meaning of an MWE is not entirely (or even not at all) predictable from the meanings of the constituents (Sag et al, 2002;Reddy et al, 2011;Salehi et al, 2014;Schulte im Walde et al, 2016;Cordeiro et al, 2019;Schulte im Walde and Smolka, 2020). Even though MWEs are ubiquitous not only in general-but also in domain-specific language (Clouet and Daille, 2014;Hätty et al, 2021), up to date only few NLP systems have exploited MWE meaning modules, as in machine translation (Cholakov and Kordoni, 2014;Weller et al, 2014). This study is faced with 98% noun compounds among our domain-specific targets, and we test the compound-head compositionality assumption (e.g., a seat heating switch led "is a type of" led but an engine valves rocker arm "is not a type of" arm) to fight the severe MWE-triggered data sparsity.…”

“…Finally, we tackle the challenge of handling a dataset with a vast majority of multiword expressions, i.e., 98% of our targets are noun compounds (such as seat heating switch led). We address the corresponding severe data sparsity by assuming that many domain-specific compounds are compositional (Clouet and Daille, 2014;Hätty et al, 2021) such that our domain-specific model may fall back to information regarding the compound's head (in the example above: the right-most simplex noun in the compound led), and thus improve on the data sparsity.…”