When the two (or more) interacting systems are described as information systems, the interaction is usually referred to as "communication". Though this process is still carried out by certain underlying physical, chemical, or biological process, to treat it as communication, or information transferring, means the process can be meaningfully described at an abstract level, by ignoring the details of the underlying process.
As explained in Section 1.3, to describe an interaction as communication means we are only interested in how the states of the two systems involved influence each other, not the details of the interaction itself, which is abstractly described as a "sender" passes some "information" to a "receiver". The information is represented as sentences of a "language", which is a system of conventions accepted by both systems.
To the system, the function of communication is similar to that of tool usage, in that it allow the system to achieve goals beyond the scope of its own operations. Furthermore, communication provides experience that cannot be directly obtained from the sensorimotor mechanism, by allowing systems to directly share their abstract beliefs and concepts.
In a communication process, both the sender and the receiver behave according to their goals for the process, which is usually different. In NARS, there are three types of sentences: judgment, question, and goal. When received by a system as incoming tasks, their interpretation and processing have been described in the previous chapters. When they are sent as outgoing tasks, for the sender they are operations with expected effects:
In an artificial system like NARS, it is possible to directly use the (internal) representation language, Narsese, as an (external) communication language. Technically this is easy to happen: we can simply implement two (or more) versions of NARS, and connect them in certain way, define message-sending operations, and give it the knowledge that sending, or processing, certain message may lead to the achieving of certain goal. Then the systems can begin to send messages to each other, as well as to process the received messages.
The complexity of this process comes from the fact that even though all the system follow the same syntactic and semantic rules, the meaning of the terms will be more or less different in the systems, due to the difference in their parameters and experiences. As a result, the sender and the receiver often cannot "fully understand" each other, in the sense of making accurate predictions about each other's goal of the communication and the current meaning of the terms involved. Consequently, misunderstanding to certain extent becomes inevitable, though for the effectuality and efficiency, all the systems have to choose and adjust its usages of terms, for the communication to be possible at all.
When a NARS communicates with a non-NARS system, it must use a language that is different from Narsese. For example, to communicate with human beings, very often the preferred language is a human language (also called "natural language"), like English or Chinese. NARS needs to learn the vocabulary, syntax, semantics, as well as the various forms of signals representing the words. Though being able to use one or more natural languages is desired for many theoretical and practical reasons, it is not a necessary condition for a system to be intelligent, just like to have a human-like sensation channel is not a necessary condition of that.
For NARS, Narsese is the unique innate language the system has, in the sense that its syntactic and semantic rules are built-in, though its vocabulary still need to be learned, together with ways to code the language in various types of signal. This language provides the foundation for the internal representation of the system, in which the system's beliefs, tasks, and concepts are expressed. It can also be used for communication, but it is limited to be used among NARS or Narsese-using systems. On the other hand, NARS can acquire various other languages, which are mostly used for communication, and information expressed in those languages are usually translated from or to Narsese for representation.
Of course, such an innate language is not universal among all intelligent systems. Biological systems, like the human beings, need to learn their native communication languages, though it can be argued that there is still a "mental language" which comes much more from nature than from nurture. This language is much less accurately specified than Narsese, though it is arguable that Narsese provides the closest model to this mental language than the various forms of "language of thought" proposed previously, in terms of its principles both in syntax and in semantics.
For other language, the learning process is more complicated, however, the learning mechanism remains the same, which is the general learning-reasoning mechanism described in the previous chapters. NARS has no special mechanism dedicated to language learning, and what make the process different is that it is working on linguistic experience, which is an abstract description of certain types of sensorimotor experience.
As stated previously, language comprehension and production are operations carried out by the system to achieve certain goals. When the system has a need and believes it can be realized by communicate with other systems, it will first choose a language for this communication session, and then decide a message to be sent. When a system receives a message, it will also first identify the language used, then process the message according to its own goal, which usually including the recognition of the intended goal of the sender.
To carry out the above process, the system needs linguistic knowledge that is specific about a certain language, which can be further divided into:
Beside the language-specific knowledge, there are also more general linguistic knowledge that covers a certain language family, or even all languages.
In NARS, the learning of a communication language starts by having terms corresponding to words, phrases, and sentences of that language, then, linguistic knowledge are built and adjusted as beliefs that relate these terms to the other (non-linguistic) terms. Especially, there will be a "symbolizing" relation that maps the terms (concepts) in the internal representation to the items of the communication language. This mapping is usually uncertain (measured by truth-value, as the other beliefs), and many-to-many (i.e., allowing synonym and polysemy). Though the meaning of a word is largely captured by the concepts it symbolizes, the other experienced relations between the word and the other (linguistic or not) terms also contribute to its meaning, just like how the meaning of other terms are determined in NARS.
This treatment of word meaning makes natural language processing in NARS very different from the traditional AI works in this field. For example, in NARS meaning is not fully "compositional", because the meaning of a phrase or sentence may not be reducible to that of the words composing it. Also, the system is tolerant to ambiguity to a certain extent, because a word or sentence can have multiple meanings at the same time. Furthermore, what a word or a sentence means become highly context-sensitive, though not arbitrary.
According to the previous description about the communication process, this understanding of "understanding" is too rough an approximation for our purpose, though it may be good enough for some other purposes. Since communication is an goal-oriented activity for both the sender and the receiver, whether the process is successful should be judged by whether they achieved their goals, respectively.
For the sender, the direct goal of the communication is to present a task to the receiver, to be processed by the latter in an expected way. In the simplest case, the goal is indeed to add a task into the receiver's memory as a literal translation of the meaning of the message. However, very often it is not the case. Instead, the expected result is a derived task that come from the literal meaning of the message, plus certain beliefs in the mind of the receiver. This explains why an incomplete sentence or a question is often received as a declarative sentence. Therefore, for an outgoing message to be properly understood, the sender's expectation about how the message will be processed should match what actually happen in the receiver's mind, otherwise it is misunderstanding, from the sender's point of view.
The receiver's goal in the communication may be not exactly the same as the sender's, though often related to it. In the simplest case, the receive just tries to estimate the sender's goal from the message, and forms tasks accordingly, or takes certain immediate actions. As far as the receiver has a strong belief on its estimation about the goal of the sender, the receiver considers the message "understood". Similarly, misunderstanding can happen when the receiver's beliefs about the sender do not match what happened in the mind of the sender.
As soon as the experience of the sender and the receiver become complex enough, "perfect understanding" become impossible, and whether a message is "understood" becomes a matter of degree. During real-time multi-round communication (such as dialogue), some misunderstandings can be recognized and corrected by the following communication, but it will be much harder if the communication is one-way (such as broadcasting) or delayed (such as reading). However, a misunderstanding does not means a failed communication, especially from the view point of the receiver. Very often a receiver achieves important goals by process the received message in a way that is completely different from what the sender expected, though still triggered by the message. Therefore, the "intended meaning" (of the sender) is not necessarily the "obtained meaning" (of the receiver), and the former is not always "more desired" than the latter.
"Don't understand", on the other hand, is indeed a failure of communication, which is very different from "misunderstand". Here the receiver fails to achieve any goal from the received message.