Majority of the studies related to sentiment lex-icon generation exploit word-to-word relationship considering resources like dictionary such as Wordnet or an extensive collection of document corpus. Sentiment lexicon is regarded as a primary resource for sentiment analysis task. However, creating such resources is an expensive operation, and may not be feasible for resource-poor languages. For syllabic and agglutinative languages, a syllable may pose discriminating characteristics capable of detecting word polarity. An advantage of considering syllabic features is its ability to achieve reasonable detection accuracy even with a much smaller dataset. In this paper, we propose various syllabic features by exploiting syllable-to-syllable relationships and investigate its effect on word polarity detection over a limited word corpus for Manipuri language (a resource-poor language spoken in Manipur, a state in North Eastern part of India). The extracted features are subjected to various classification frameworks. From various experimental observations, it is evident that syllable-to-syllable relationship based features outperform its word-to-word relationship based counterparts.