The experience self-conscious thinking in the verbose form of inner speech is a common one. Such a covert dialogue accompanies the introspection of mental life and ful lls important roles in our cognition, such as self-regulation, self-restructuring, and re-focusing on attentional resources. Although the functional underpinning and the phenomenology of inner speech are largely investigated in psychological and philosophical elds, robotic research generally does not address such a form of self-conscious behavior. Existing models of inner speech inspire computational tools to provide the robot with a form of self-consciousness. Here, the most widespread psychological models of inner speech are reviewed, and a robot architecture implementing such a capability is outlined.
Inner speech plays a central role in daily life. A person thinks over her mental
states, perspectives, emotions and external events by generating thoughts in
the form of linguistics sentences. Talking to herself enables the person to pay
attention to internal and external resources, to control and regulate her behavior,
to retrieve memorized facts, to learn and store new information and, in general,
to simplify otherwise demanding cognitive processes [
Moreover, inner speech allows restructuring the perception of the external world and the perception of self by enabling high-level cognition, including selfcontrol, self-attention, and self-regulation.
Even if second-order thoughts may not need language but, for example,
images or sensations, Bermudez [
We model such a necessary capability within a cognitive architecture for robot self-consciousness by considering the underlying cognitive processes and components of inner speech.
It should be remarked that in the present paper such processes are taken into account independently from the origin of the linguistics abilities which are supposed acquired by the robot.
In Section 2 we show a brief overview of the cognitive models underlying the proposed robot architecture, which is detailed in Section 3. Conclusions and future works about the proposed robot architecture are discussed in Section 4. 2
Inner speech cannot be directly observed, thus reducing the scope for empirical studies. However, theoretical perspectives were developed during the last decades, and some of them are recognized in di erent research communities.
Vygotsky [
Morin [
Baddeley [
Inner speech is usually conceived as the back-propagation of produced
sentences to an inner ear: thus, a person rehears the internal voice she delivers.
Steels [
In the same line, Clowes [
al.[
The proprioception module, according to Morin [
The motor module is decomposed in three sub-components: the Action module, the Covert Articulation module (CA) and the Self Action module (SA). In particular: { The Action module represents the actions the agent performs on the outside world producing modi cations to the external environment (not including the self) and the working memory. { The Covert Articulation (CA) module rehearses information from the Phonological Store (PS), i.e., the perceptual bu er for speech-based data considered as a sub-component of the short-term memory (see below). Such a module acts as the inner voice heard by the phonological store by rounding information in a loop. In this way, the inner speech links the covert articulation to the phonological store in a round loop. { The Self Action (SA) module represents the actions that the agent performs on itself, i.e., self-regulation, self-focusing, and self-analysis. 3.2
The memory structure, inspired by the Standard Model of the Mind is divided into three types of memories: the short-term memory (STM), the procedural and the declarative long-term memory (LTM), and the working memory system (WMS).
The short-term memory holds sensory information on the environment in which the robot is plunged that were previously coded and integrated with information coming by perception. As previously mentioned, the short-term memory includes the phonological store.
Information ow from perception to STM allows storing the aforementioned coded signals. In particular, information from perception to the phonological store is related to conscious thoughts from exteroception, and to self-conscious thoughts from proprioception.
The information ow from the working memory system to perception provides expectations or possible hypotheses that are employed for in uencing the attention process. In particular, the ow from the phonological store to proprioception enables the self-focus modality.
The long-term memory holds learned behaviors, semantic knowledge, and experience. In the considered case, the declarative LTM contains the linguistics information in terms of lexicon and grammatical structures, i.e., the LanguageLTM memory. The declarative linguistics information is assumed acquired, as speci ed above, and represent the grammar of the robot. Moreover, the Episodic Long-Term Memory (EBLTM) is the declarative long-term memory component which communicates to the Episodic Bu er (EB) within the working memory system, that acts as a `backup' store of long-term memory data.
The procedural LTM contains the composition rules according to which the linguistic structures are arranged for producing sentences at di erent levels of completeness and complexity. A procedure does not concern the grammatical plausibility of the structures only. Other rules concerning the regulation, the focusing and the restructuring of resources within the whole environment (including the self) are to be considered.
Finally, the working memory system holds task-speci c information `chunks' and streamlines them to the cognitive processes during the task execution, step by step according to the cognitive cycle of the Standard Model of the Mind. The working memory system deals with cognitive tasks such as mental arithmetic and problem-solving. The Central Executive (CE) sub-component manages and controls the linguistic information of the rehearsal loop by the integrating (i.e., combining) data from the phonological loop and also drawing on data held in the long-term memory.
In brief, a cognitive cycle starts with the perception that converts external signals in linguistics data and holds them into the phonological store. The central executive manages the inner thinking process by enabling the working memory system to selectively attend to some stimuli or ignore others, according to the rules stored within the LTMs, and by orchestrating the phonological loop as a slave system.
At this stage, a conscious thought emerges as a result of a single round between the phonological store and the covert articulation triggered by the phonological loop, once the central executive has retrieved the data for the process. The phonological loop enables the covert articulation which acts as a motor for the internal production, and whose output stream is heard to the phonological store. The output stream also a ects the self which is then regulated and restructured.
Once the conscious thought is elicited by inner speech, the perception of the new context could take place, repeating the cognitive cycle. 4
In this paper, a cognitive architecture for inner speech cognition is presented. It is based on the Standard Model of Mind which was decomposed for including some typical components of the inner speech's models for human beings.
The working memory system of the architecture includes the phonological loop considered by Baddeley as the main component for storing spoken and written information and for implementing the cognitive rehearsal process.
The covert dialogue is modeled as a loop in which the phonological store hears the inner voice produced by the covert articulator process. The central executive is the master system which drives the whole system.
By retrieving linguistic information from the long-term memory, the central executive contributes to creating the linguistic thought whose surface form emerges by the phonological loop.
This material is based upon work supported by the Air Force O ce of Scienti c Research under award number FA9550-17-1-0232.