Interactive Analysis Of Multi-Layer Linguistic Corpora With Annis
Florian Zipser, Thomas R Krause, Arne Neumann · Zenodo (CERN European Organization for Nuclear Research) · 2015
In this poster, we present new features of the ANNIS 1 search- and visualization system for multi-layered corpora. ANNIS was developed as a platform to explore a wide range of corpora. It is not limited to a specific type of annotation or a single corpus. Instead, ANNIS abstracts the linguistic data and interprets them as a graph (with nodes representing tokens or structural elements like phrases and edges symbolizing the relationships between them). This abstraction allows ANNIS to use the same query language for vastly different corpora. Furthermore, ANNIS comes with a set of different visualizations to display corpus specific annotation layers like syntactic trees, coreference chains, rhetorical structure trees and many more. To give the user a familiar feeling when searching through the data, these visualizations are very close to the ones used in the original annotation tools. The particular power of ANNIS is the combined search and visualization of several annotation layers at a time. This empowers linguists to search comprehensively for phenomena on several layers, which becomes important with the increasing number of multi-layered corpora like TüBa-D/Z 2 , PCC 3 or the Falko 4 corpus. In addition to performance improvements, the latest ANNIS version features a new frequency analysis module and improvements of the ANNIS query language AQL. The new module allows to calculate frequencies of annotations without resorting to external tools. AQL now supports queries with logical alternatives and the possibility to search for multiple annotations with the same or different values. Furthermore, we simplified the syntax of AQL to make complex queries easier to read and write. The frequency module, in combination with these AQL improvements simplifies the interactive analysis of linguistic corpora and allows a wider range of linguistic analysis directly in ANNIS. We will show these improvements along with the revised multi-layered PCC 2.0 corpus at the live demonstration.