This thesis presents a framework for the automatic recognition of Sign Language sentences. In previous sign language recognition works, the issues of; user independent recognition, movement epenthesis modeling and automatic or weakly supervised training have not been fully addressed in a single recognition framework. This work presents three main contributions in order to address these issues. The first contribution is a technique for user independent hand posture recognition. We present a novel eigenspace Size Function feature which is implemented to perform user independent recognition of sign language hand postures. The second contribution is a framework for the classification and spotting of spatiotemporal gestures which appear in sign language. We propose a Gesture Threshold Hidden Markov Model (GT-HMM) to classify gestures and to identify movement epenthesis without the need for explicit epenthesis training. The third contribution is a framework to train the hand posture and spatiotemporal models using only the weak supervision of sign language videos and their corresponding text translations. This is achieved through our proposed Multiple Instance Learning Density Matrix algorithm which automatically extracts isolated signs from full sentences using the weak and noisy supervision of text translations. The automatically extracted isolated samples are then utilised to train our spatiotemporal gesture and hand posture classifiers. The work we present in this thesis is an important and significant contribution to the area of natural sign language recognition as we propose a robust framework for training a recognition system without the need for manual labeling.