Development of Maximally Reusable Grammars:
Parallel Development of Hebrew and Arabic Grammars


Tali Arad Greshler, Livnat Herzig Sheinfux, Nurit Melnik & Shuly Wintner

In this talk we discuss the development of linguistic grammars of two different yet related languages. We show that such grammars can be developed and implemented in parallel, with language-independent fragments serving as shared resources, and language-specific ones defined separately for each language.

The two grammars in the focus of our research are of Modern Hebrew (MH) and Modern Standard Arabic (MSA). The basic infrastructure, or core, of the grammars is “standard” HPSG (Pollard and Sag, 1994; Sag et al., 2003). Moreover, as these two languages are related, they exhibit a number of shared phenomena which can be attributed to their Semitic roots. Nevertheless, since the languages diverged several millennia ago the end grammars are quite different and do require language-specific accounts.

More generally, we identify four types of relations that exist between the grammars of two languages: (i) The two languages share some construction or syntactic phenomenon. (ii) Some phenomenon is present in one language but is absent from the other. (iii) The two languages share some construction, but impose different constraints on its realization. (iv) Some phenomenon seems similar in the two languages, but is in fact a realization of different constructions. While the challenge is to maximize the common parts of the grammars, it is important to be cautious with seemingly similar phenomena across the two languages. In some cases, as we will show, the solution is to define a shared construction with different language-specific constraints. Conversely, other cases are best accounted for by the definition of distinct constructions.

We demonstrate how the different types of relations can be implemented in parallel grammars with maximally shared resources. The examples pertain to the MH and MSA grammars, yet similar issues and considerations are applicable to other pairs of languages that have some degree of similarity.