Transliterated Mobile Keyboard Input via Weighted Finite-State Transducers

By Lars Hellsten, Brian Roark, Prasoon Goyal, Cyril Allauzen, Franccoise Beaufays, Tom Y. Ouyang, Michael D. Riley, David Rybach. 2017

31 citations recorded by the source corpus.

We present an extension to a mobile key-board input decoder based on ﬁnite-state transducers that provides general translit-eration support, and demonstrate its use for input of South Asian languages using a QWERTY keyboard. On-device keyboard decoders must operate under strict latency and memory constraints, and we present several transducer optimizations that allow for high accuracy decoding under such constraints. Our methods yield substantial accuracy improvements and latency reductions over an existing baseline translit-eration keyboard approach. The resulting system was launched for 22 languages in Google Gboard in the ﬁrst half of 2017.

Subfields: Artificial Intelligence

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Federated Learning for Emoji Prediction in a Mobile Keyboard (2019, background): “Low-latency is also required, since users typically expect a keyboard response within 20 ms of an input event (Hellsten et al., 2017).”
Federated Learning of N-Gram Language Models (2019, methodology): “Weighted ﬁnite-state transducers have been used successfully to decode keyboard spatial signals us-ing a combination of spatial and language models (Ouyang et al., 2017; Hellsten et al., 2017).”
Transliteration Based Approaches to Improve Code-Switched Speech Recognition Performance (2018, methodology): “In [19], the authors extend the ﬁnite-state decoding approach to transliterated keyboards and demonstrate that WFST encoding of transliteration models allows for several optimizations that yield good accuracy.”
Processing South Asian Languages Written in the Latin Script: the Dakshina Dataset (2020, methodology): “…modeling in a range of scenarios – see Wolf-Sonkin et al. (2019) for further citations – including the use of transliteration in increasingly challenging settings, such as information retrieval with mixed scripts (Gupta et al., 2014) or for text entry on mobile devices (Hellsten et al., 2017).”