English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Recently, Transformer based pretrained language models (PLMs), such as GPT2and T5, have been leveraged to build generative task-oriented dialog (TOD)systems. A drawback of existing PLM-based models is their non-Markovarchitectures across turns, i.e., the whole history is used as the conditioninginput at each turn. First, this brings inefficiencies in memory andcomputation. Furthermore, using the whole history increases model complexityand may hurt the training efficiency, especially when facing small amounts oflabeled training data (the low-resource setting). In this paper, motivated bythe observation that dialog states could be viewed as Markov states, we proposeto build Markovian Generative Architectures (MGA) over PLM backbones forefficient TOD systems. Experiments on MultiWOZ2.1 show that in therich-resource setting, the proposed Markov models reduce memory and time costswithout performance degradation; in the low-resource setting, the trainingefficiency of the Markov models is more significant.

Building Markovian Generative Architectures over Pretrained LM Backbones for Efficient Task-Oriented Dialog Systems