SentiD is a multilingual lexicon of affect-annotated words and sentences. Each word/sentence is annotated manually with pleasure-arousal-dominance (PAD) emotion representation model. Annotation is performed for Polish and English, but could be extended to other languages.
SentiD development method utilizes the most of the previous research on dictionaries. Similarly to Mehrabian’s and Anew dictionaries, it is annotated with PAD representation model and both means and standard deviations will be reported for the words/sentences. However, the two mentioned lexicons are for English only and the SentiD is intended to be a multilingual resource, for English, Polish and perhaps more languages. Although primary focus is to create Polish dictionary, as there is no such resource, annotation of English words allows for verification of the construction method, due to the possibility of comparison with the most frequently used dictionaries.
Use of the Internet would allow to reach more respondents and provide a larger diversity of them in order to improve representativeness. However, due to the use of a website, some threats to credibility might occur, as web survey users might intentionally or unintentionally falsify the responses (the issue is present also in paper questionnaires, however the scale of misreport is lower).
Each study participant is asked to fill in metric information and one or more sets of affect annotations to words and/or sentences. A participant may withdraw from the study at any stage and an incomplete set of words or sentences is not saved. Each set consists of n words and m sentences, and the parameters are configurable. Each word or phrase to be assessed is displayed on a separate screen, and there is no possibility of going back to the previous screen. One screen is for one word or sentence in one of the languages only. The screen contains a rated word or sentence and three sliders, corresponding to the dimensions of the PAD model. Sliders are described with dimension name, multiple adverb labels on marginal values, calibrated on a scale (-10,10) with 0.5 precision and accompanied by pictograms derived from SAM questionnaire.