A morphological analyser for Kven

The 3rd International Workshop for Computational Linguistics of Uralic Languages, pages 76–88, St. Petersburg, Russia, 23–24 January 2017. c⃝2017 Association for Computational Linguistics A morphological analyser for Kven Sindre Trosterud Kainun institutti Trond Trosterud Giellatekno, UiT Anna-Kaisa Räisänen Kainun institutti Leena Niiranen UiT Mervi Haavisto Kainun institutti Kaisa Maliniemi Ruija kvenmuseum Abstract We present a morphological analyser for Kven, a Finnic language spoken in Northern Norway. Apart from the overall view of the analyser, we discuss prin- ciples for whether to treat morphological processes like gemination and stem- conditioned suffix variation as lexically, morphologically or phonologically con- ditioned alternations. The choice is often governed by the grammatical mecha- nism itself, but in many instances the analysis may be carried out in more than one way and it becomes a choice of where to place the complexity. The article discusses some of the choices we made in this particular case. 1 Introduction The article presents a morphological analyser for the Kven language, i. e. a program that is an explicit model of the Kven grammar, by which the user may either analyse or generate any Kven wordform, and presents both program design choices and some of the usage possibilities. The article is structured as follows: Section 2 first gives the technical background for the technologies used to create this analyser, and then an overview of the Kven language itself. Section 3 presents how we have implemented this FST, discusses the rationale behind some of our choices, describes our test setup and mentions the ap- plications in which the analyser is used. In section 4 we evaluate the analyser, and section 5 contains a conclusion. This work is licensed under a Creative Commons Attribution–NoDerivatives 4.0 International License. License details: http://creativecommons.org/licenses/by-nd/4.0/ 1 76 2 Background 2.1 Theoretical background This analyser is modeled like a finite-state transducer. A finite-state transducer (FST) is an automaton with both an input tape and an output tape, that defines relations between pairs of strings. When made as a specialized finite-state automation (a lexical transducer) it maps lexemes and morphological specifications (in the form of tags) to corresponding inflected forms, and vice versa. The transducer modeling morphological processes for Kven actually consists of two separate transducers. The first one is a Lexicon Transducer (Morphological) that takes input A = {the lexeme and a string of tags denoting the grammatical analysis of the word form in question}, then the second one is a morphophonological transducer that uses the first transducer’s output B = {the stem + trigger and suffixes} as input, and outputs C = {a string of letters, i.e. the word form}. The transducers are composed, so that A/B ∗ B/C = A/C (or in standard notation: A : B .o. B : C = A : C). The transducer is bidirectional, and gives both analysis and generation. A = t a k k i +N +Sg +Gen B = t a k k i ^WG > n B = t a k k i ^WG > n C …

Content truncated to 500 words