Aspect Sentiment Triplet Extraction (ASTE) represents a sophisticated, detailed sentiment analysis endeavor focused on identifying aspects, opinions, and their associated sentiment polarities within a given sentence. While existing methods have evolved to end-to-end models based on spans, achieving significant results, they often overlook the linguistic information between aspect spans and opinion spans within sentences. This oversight leads to poor performance when capturing complex semantic contexts. To tackle this issue, we develop a Linguistically Enhanced Span-based Architecture for Aspect Sentiment Triplet Extraction (LES-ASTE) which exploits the semantic and syntactic feature at span-level, enabling precise extraction of aspect sentiment triplets. In order to access the intrinsic connections between aspects and opinions and to better couple effective aspect spans and opinion spans obtained by Top-K prune strategy, we use span-based Graph Neural Networks (GNNs) to encode the linguistic information and enhance the span representations. Additionally, we utilize an aspect-opinion bidirectional structure to decode the span pairs so that we can extract the aspect sentiment triplets more comprehensively. Extensive experiments on SemEval datasets reveal that LES-ASTE model reaches remarkable performance, with Precision, Recall, and F1 score metrics surpassing the majority of state-of-the-art models.