Evaluation of natural language generation (NLG) is complex and multi-dimensional. Generated text can be evaluated for fluency, coherence, factuality, or any other dimensions of interest. Most frameworks that perform such multi-dimensional evaluation require training on large manually or synthetically generated datasets. In this paper, we study the efficacy of large language models as multi-dimensional evaluators using in-context learning, obviating the need for large training datasets. Our experiments show that in-context learning-based evaluators are competitive with learned evaluation frameworks for the task of text summarization, establishing state-of-the-art on dimensions such as relevance and factual consistency. We then analyze the effects of factors such as the selection and number of in-context examples on performance. Finally, we study the efficacy of in-context learning-based evaluators in evaluating zero-shot summaries written by large language models such as GPT-3.

Intent Identification and Slot Identification aretwo important task for Natural Language Understanding(NLU). Exploration in this areahave gained significance using networks likeRNN, LSTM and GRU. However, modelscontaining the above modules are sequentialin nature, which consumes lot of resourceslike memory to train the model in cloud itself. With the advent of many voice assistantsdelivering offline solutions for manyapplications, there is a need for finding replacementfor such sequential networks. Explorationin self-attention, CNN modules hasgained pace in the recent times. Here we exploreCNN based models like Trellis and modifiedthe architecture to make it bi-directionalwith fusion techniques. In addition, we proposeCNN with Self Attention network calledNeighbor Contextual Information Projector usingMulti Head Attention (NCIPMA) architecture. These architectures beat state of the art inopen source datasets like ATIS, SNIPS.