The findings matter because they change how we think about the building blocks of speech. Eye movements and speech timing show that speakers encode core details about both the agent and the patient early, even when the patient will appear at the end of the sentence. That pattern implies that the preverbal message contains more than a list of words to be produced next; it holds conceptual information about multiple characters in the event. For anyone studying how people plan language, learn languages, or design systems that interact with human speakers, these results point to richer early representations than many models assume.
If you care about how ideas become sentences and how that process supports clear, inclusive communication, this line of work opens important questions. How might richer early planning affect learning in children or people acquiring a second language? Could it inform tools that support people with language production difficulties or guide conversational AI to anticipate relevant referents? Follow the full article to explore how these findings connect to human potential, growth, and more inclusive communication.
Abstract
Speaking begins with the generation of a preverbal message. While a common assumption is that the scope of message-level planning (i.e., the size of message-level increments) can be more extensive than the scope of sentence-level planning, it is unclear how much information is typically encoded at the message level in advance of sentence-level planning during spontaneous production. This study assessed the scope and granularity of early message-level planning in English by tracking production of sentences with light versus heavy sentence-final NPs. Speakers produced SVO sentences to describe pictures showing an agent acting on a patient. Half of the pictures showed one-patient events, eliciting sentences with unmodified patient names (e.g., “The tailor is cutting the dress”), and half showed two-patient events with a target patient and a non-target patient. The presence of a non-target patient required production of a prenominal or postnominal modifier to uniquely identify the target patient (e.g., “The tailor is cutting the long dress” / “the dress with sleeves”). Analyses of speech onsets and eye movements before speech onset showed strong effects of the complexity of the sentence-final character, suggesting that early message-level planning does not proceed strictly word by word (or “from left to right”) but instead includes basic information about the identity of both the sentence-initial and sentence-final characters. This is consistent with theories that assume extensive message-level planning before the start of sentence-level encoding and provides new evidence about the level of conceptual detail incorporated into early message plans.