Five experiments investigate the scope of conceptual and grammatical encoding during spoken sentence production. An online picture description task is employed in which participants generate a variety of sentences in response to an array of moving pictured objects. Experiment 1, demonstrates longer onset latencies for single clause sentences beginning with a complex phrase (e.g. The dog and the kite move above the house) than for matched single clause sentences beginning with a simple phrase (e.g. The dog moves above the kite and the house). This finding suggests that more time is dedicated to the processing of the first phrase of an utterance than the remainder prior to speech onset. Experiments 2 and 3, compare the production of single and double clause sentences. The main effect of Experiment 1 is replicated. However, the data also suggest that some time is dedicated to the processing of elements within the second clause prior to speech onset. In Experiment 4, when participants are allowed to preview pictures prior to movement and timer onset the effect of initial phrase complexity is significantly reduced indicating that the latency effects observed previously primarily reflect lemma access. Finally, Experiment 5 demonstrates that this reduction is greater for nouns within the first phrase than for nouns beyond it. We conclude from these experiments that, prior to speech onset, lemma access is completed for the first phrase of an utterance and that high level processing is initiated but not completed for the remainder of a sentence beyond the first phrase.