Background
TMV-annotator is a tool for automatic annotation of morphosyntactic tense, mood and voice for German and English. The annotation is based on a set of hand-crafted language-specific rules.
What is morphosyntactic tense, mood and voice?
Morphosyntactic tense, mood and voice refer to the tense, mood and voice defined by the morphosyntactic structure of the given verbal complex (VC). For example, the morphosyntactic tense of the English VC 'is playing' is present progressive. The mood is defined by the inflection of the finite verb. For example, the the English VC 'will play' is indicative, while the VC 'would play' is subjunctive. The voice differentiates between active (e.g. 'will play') and passive (e.g. 'will be played') actions/states.How does it work?
The annotation is a two-step process. First, the VCs are extracted from the dependency parse trees of the sentences to be annotated. Second, the VCs are assigned tense, mood and voice values according to the specific morphosyntactic features. For example, the English VC 'is playing' consists of a finite verb in present tense ('is') and a gerund ('playing'). This combination of morphosyntactic features leads to the following TMV values: tense = presProg (present progressive), mood = indicative, voice = active.Annotation accuracy
The tool has been evaluated on the output of the Mate parse trees. It reaches accuracy of ca. 75% for all languages. Most of the annotation errors are however due to errors in the parse trees which often lead to the errors in the extraction of the VCs. Erroneous morphological annotation has in addition a negative impact on the annotations for German and English.Further information
For download the tool, please visit our Github site. Further information can be found in the following paper:Anita Ramm, Sharid Loaiciga, Annemarie Friedrich and Alexander Fraser: "Annotating tense, mood and voice for English, French and German", ACL demo, Vancouver, Canada, August 2017 (to appear).
Contact
Please feel free to contact Anita Ramm for issues related to the tool.Funding
The tool has been developed as a part of the project "Models of Morphosyntax for Statistical Machine Translation" funded by the "Deutsche Forschungsgemeinschaft (DFG)".