<p>A SentiWordNet (SWN) is a WordNet, in which the synsets are annotated with sentiment scores. The initial steps for building a new SWN in a target language are similar to those for building a new WordNet. In particular, the creator may translate existing SWNs or WordNets to the target language or expand a seed sentiment lexicon. The next step involves training classifiers to identify the sentiment of the synset members. A major issue in building a new SWN is the lack of language resources for translating existing SWNs or WordNets to the target language and creating a dataset for training sentiment classifiers. Bilingual dictionaries are reliable aids in translation, but are expensive, may not be available, and time-consuming to construct. With the rapid development of artificial neural networks, machine translation systems have become commonplace and effective for a significant number of language pairs from around the world. Creating datasets for training machine translation models is also arguably cheaper than constructing bilingual dictionaries from scratch. This paper proposes effective approaches for constructing new SWNs using neural network translation systems. We introduce strategies for selecting synset members from translation candidates and computing sense-orders for these synset members. Our approaches are able to construct a new SWN in any language if the language is supported by machine translators.</p>