Confidence Measures (CMs) can be used to estimate the reliability of the words of a hypothesis generated by a machine translation system. In the Interactive-Predictive Machine Translation (IPMT) paradigm, they are used to determine which words of the generated predictions need to be corrected, reducing the total number of words typed by the user. The CMs used must be fast enough to do not affect the interaction between the user and the machine negatively. In this paper, we present several fast CMs for Interactive Neural Machine Translation: IBM Model 1 and 2, Fast Align and Hidden Markov Model. These estimators let the system to achieve a reduction in the number of words typed by getting less-quality translations. The experiments done proved that these CMs are fast enough to use them in an IPMT system, and obtained a high relative reduction on the number of words corrected while getting good-quality translations.