In this paper, we describe a new approach for text-image alignment of middle-age documents. The method is dedicated to word-to-word alignment in a segmentation-free and learning-free way. The best word-to-word matching lies on an adapted editing cost between signatures extracted on Unicode characters and on the images. The results are evaluated on the "Queste del saint Graal" (13th c.) by palaeographers through an intuitive validation interface that also offers a very fast interactive correction process. The gain of time resulting from the absence of learning stage offers the opportunity to pay more attention to the integration of different specificities and variations in middle-age handwritten documents (presence of typical abbreviations, allographs).