In language, the data of experience are organized. The continuum of extralinguistic reality is autonomously – and therefore arbitrarily – partitioned by each language. Some languages know only three basic color terms (white, black, and red); others know twelve (Berlin & Kay, 1969). The color spectrum is, of course, always the same; what varies is the way of classifying the continuum of light vibrations. In some languages, this is divided into three sections, corresponding to colors classified as light (white), those classified as dark (black), and intermediate colors (red); in others, like Italian and European languages, it is further articulated. In Latin, but not in Italian or French, which also belong to Romance languages, descendants of Latin, the brightness feature distinguishes the meaning of the word niger(referring to shiny black) from that of ater (referring to matte black), just as candidus "bright white" is distinguished from albus "matte white." In Italian, tempo corresponds to time, weather and tense in English, depending on whether the time is Augustinean, meteorological, or grammatical; to Italian andare and English to go, German responds with two different verbs, gehen and fahren, depending on whether one is walking or using transportation. Italian selects the relationship with 'eye' as relevant to describing the extralinguistic referent of the word 'occhiali'; English glasses and German Brille select the material (glass and beryllium respectively) of which the object is or was made; French lunettes its originally crescent-shaped form; in Italian, glasses are also called 'lenti', referring to the object's physico-optical function: in the same language, distinct signs may have the same denotation but convey different descriptive values. Examples of this kind can multiply. Languages are cognitive instruments that convey not a given reality, but an interpreted reality.
The noetic continuum allows for a potentially infinite number of interpretations, which languages encode into discrete forms. Each language organizes conceptual space differently, as it interprets the world differently. Not all linguistic categories, indeed, have identical manifestations in all languages or in diachronically different stages of the same language. Consider, for example, the categories of number and tense. Some languages, like ancient Greek, distinguish the notion of one not only from that of more than one but also from that of exactly two, and they have different encodings for singular, plural, and dual; other languages, like Italian and English, instead, only distinguish between singular and plural; others, like Arabic, also present a specific form of paucal (the "few plural"), and still others, like some languages of southeastern Australia, have trial, which encodes the notion of exactly three. In Greek, the dual tends to progressively reduce its scope of application, until it disappears: the partitioning of the noetic continuum corresponding to the category of number, in modern Greek, produces two formal distinctions and not three, as in the earliest stages of the language.
Proto-Indo-European, the reconstructed language from which Italian, English, German, and other European languages descend — according to the results of the analysis of the earliest attested stages of Indo-European languages (e.g., Greek and Vedic from ancient India) — did not have grammatical encoding of tense: the verbal system was based on the opposition between process and state, without distinguishing between past, present, and future (Lazzeroni 1977, Romagno 2021). There are languages that still today do not have words, grammatical forms, constructions, or expressions that can refer to what we call "time," or to any distinction between what is happening now, what happened yesterday, and what will happen (or may happen) tomorrow. This is the case, for example, of Hopi, a Native American language from Northeastern Arizona, whose first systematic study is attributed to Benjamin L. Whorf: in Hopi, what is distant in time has the same encoding as what is distant in space, and temporal proximity is expressed in the same way as spatial proximity (Whorf 1936).
In summary, languages function as an autonomous system of classification and, therefore, of knowledge, as they encode the different ways, in different times and places, in which speakers represent, interpret, and categorize the world. The language of words is the most typical product of the human brain, "a true miracle, a revolutionary epiphany that characterizes human brain function compared to that of other animals" (Maffei 2018: 71), but it is also an arbitrary language, as it is sociostorically determined: "man, his uniqueness, and his civilization are expressed by a string of words that reason threads into the necklace of history" (Maffei 2018: 18).
As seen above, colors form a continuum, which each language arbitrarily divides, classifying under the same linguistic label a series of nuances that could also be classified differently, and in other languages they are. The chromatic gradations perceivable by the human eye exceed – it is estimated – eight million; nothing would prevent each one from having a basic name, but a lifetime would not suffice to learn them all. What applies to color names applies to every other linguistic sign (Rosch 1978): each meaningful unit classifies a category of notions ordered according to their resemblance to a notion assumed as prototypical. Evidently, this is based on a principle of economy of memory mechanisms.
The boundaries of categories are blurred: this is referred to by cognitive scientists when they talk about fuzzy categories. The more peripheral notions – those with fewer features in common with the prototype – may also be classified within contiguous categories. Commonly, we define a bird as a seagull, which flies, has feathers, and is oviparous, but also the bat, which does not have feathers and is not oviparous, but flies, and the ostrich, which has feathers and is oviparous, but does not fly. If the seagull, like the eagle or sparrow, represents the prototype of the category, the ostrich and the bat are placed on the periphery, in the blurred margin of the category, and could also belong to another category if other resemblances were assumed as pertinent: for example, in German, the bat Fledermaus is classified among mice (the "flying mouse").
In synchrony, language is a system of discrete signs on the level of form (red has more common features with orange than with green, but nothing associates the name 'red' more with the name 'orange' than with the name 'green'), but scalar on the level of content: scalar – and therefore "fuzzy" – in the sense that each sign is the symbol of a noetic category of constituents classified not based on necessary and sufficient features shared by all, to the same extent, but on features – not the same for everyone – that each constituent shares with a prototype. This principle also operates diachronically: many linguistic changes originate from the tension between discrete forms and scalar contents.
Consider, for example, the Italian imperfect indicative. Let's assume its prototypical function is to signify imperfective past (in which a viewpoint internal to the event is assumed): 'yesterday I was at the beach when I met you.' The past doesn’t belong to the hic et nunc, it's non-actual. But also what is imagined or unreal is non-actual; hence, the imperfect, in its so-called "playful" use, can signify the unreal (or the imagined): 'let's pretend you were the queen.' The trait of temporality, [+ past], is not pertinent here: the part of the queen is played, even, at a moment subsequent to the utterance. The pertinent trait, however, is that of non-actuality: "let's agree you play the part of the queen (but you are not)." The progressive substitution of the imperfect indicative for the subjunctive and conditional, observed in contemporary Italian, follows the same principle: 'se eri ammalato, non uscivi di casa' (="if you were sick, you wouldn’t leave the house"), 'se studiavi, eri promosso' (="if you studied, you were promoted"), 'volevo dell’acqua, per favore' (="I wanted some water, please"), 'non so se volevi aggiungere qualcosa' (="I don’t know if you wanted to add something"). In these cases too, temporality is not pertinent; possibility and unreality are. In Italian imperfect, therefore, the category of the indicative, whose prototypical function is the encoding of factuality, touches and overlaps with modal categories that encode non-factuality or counterfactuality.
Upon closer examination, this is the strategy – or one of the strategies – that has led to the fusion (the so-called syncretism) of cases in the Indo-European languages. Between genitive and ablative, there is no possibility of overlap in prototypical functions: 'fuga dei nemici' (flight of enemies) is different from 'fuga dal pericolo' (flight from danger). But there is a possibility of conceptual overlap between 'comunicazione del Rettorato' (communication from the Rectorate) and 'comunicazione (che viene) dal Rettorato' (communication coming from the Rectorate). This represents the crisis point, from which syncretism can arise: in Ancient Greek, in all declensions, and in Sanskrit, in the singular of all except that in -ă (not coincidentally, the most frequent), ablative and genitive are syncretized. In languages like Latin, Italian, or English, notions of place and instrument are typically encoded with locative and instrumental marks or constructs, respectively: e.g., 'sto chiuso in casa' (I am closed in the house) vs. 'taglio la carne con il coltello' (I cut the meat with the knife). However, the place can also be an instrument (and vice versa): in Italian, in fact, we can say both 'vado a Roma in treno' (I go to Rome by train) and 'vado a Roma con il treno' (I go to Rome with the train). Nevertheless, the same alternation of locative and instrumental cannot occur – or at least not yet – in the prototypes of the respective categories: we cannot say *'taglio la carne nel coltello' (I cut the meat in the knife) or 'sto chiuso con la casa' (I am closed with the house).
Linguistic systems are unstable: if – as in the cases just mentioned – two categories can merge, new categories can form and expand. A well-known case is that of the so-called "strong past" in English, those past forms that deviate from the pattern talked, walked, whatched (where ed is added to the present stem:talk – talked, walk – walked,, etc.) and instead show a change in the internal vowel: e.g., sing – sang, spring – sprang. This past form is a recessive category (meaning it tends to disappear), but it persists and – indeed – becomes productive in a group of verbs centered around a prototype (like sing – sang, spring – sprang) and linked to it by a similarity relationship: the verbs attracted to the category share one or more traits with the prototype, but not necessarily all of them (Bybee & Moder 1983).
The connectionist model applies in this and other cases to morphology: the speaker, faced with irregular forms (whether they are products of innovation or residues), reorders them by abstracting some common traits (a "schema," according to the definition of Bybee & Slobin 1982), which allow them to predict and thus produce them if not with certainty, at least with a certain degree of probability. Certainty – and therefore automatism in production – can grow over the course of linguistic history: a rule, born as probabilistic, can progressively become categorical (Ramat 1985).
The more a category expands, the more connections with the prototype weaken, until they are lost: it is sufficient that the constituents share at least one trait among them. Given, for example, a prototype AB, the category can expand to include QR and beyond, following a path like: AB – BC – CD – DE – EF – FG, and so on, where CD, DE, etc. have nothing in common with the prototype AB, but they have something in common with at least one other constituent of the category (CD has something in common with BC, DE with CD, and so on). This is the model of the so-called "family resemblance" (Rosch & Mervis 1975), which dates back to the later Wittgenstein. This model is recognized at all levels of linguistic systems and in languages even very different from each other, from the dialect of Gallipoli (Romagno 2004) to Dyirbal, an Aboriginal language of Queensland, Australia (Lakoff 1987).
So far, we have discussed categories that form and expand, as well as categories that merge. There are also cases of categories that split. Imagine a language that does not have a name – let's say – for orange, but classifies its chromatic variations into the categories of yellow and red (numerous examples in Berlin & Kay 1969). If, over time, orange were to acquire a name, a chromatic point assumed as prototypical would be isolated within the continuum between the categories of yellow and red. That point would be given a name, and other gradations of the continuum would be classified under that name.
The same process occurred in Romance languages when a point in the continuum of the noetic contents of the Latin perfect was identified, corresponding to the Italian passato prossimo: to the Latin FĒCI, Italian responds with 'feci' and 'ho fatto'. The boundaries between the two new categories are blurred; indeed, in northern Italian varieties, the passato prossimo predominates, while in southern varieties, the passato remoto does.
In conclusion: the noetic categories in which humans linguistically organize experiential data are not discrete but scalar; they are "natural categories," in Rosch's sense (1973). Language, of course, is also a tool for communicating this organized experience. But one might wonder if the end of communication is the ultimate, foundational principle governing the organization of linguistic systems. Rather, this should be identified in a supraordinate neurocognitive dimension: the economy of memory mechanisms. Speakers invariably tend to favor procedural memory, i.e., memory of rules or connections that produce automatism, and to lighten declarative memory, i.e., memory of data and information stored one by one. This principle is observed to operate both synchronically, in the organization of linguistic systems, and diachronically, in the paths of linguistic change.
In the languages of the world, there is a tendency to group words into classes. In linguistic types like Italian and Greek, which we call inflectional because words flex into different forms to encode various grammatical relations (e.g., number, tense, person), a fundamental classification is the one that produces the so-called paradigms: for example, in the verbal domain, 'amo - ami - ama - amare'; 'leggo - leggi - legge - leggere', etc. Italian speakers do not need to memorize, one by one, the paradigms of 'amare', 'leggere', 'lodare', 'scrivere', 'sentire', etc. But based on one or more common features to a class of words, they automatically produce the correct forms: they produce, that is, 'ama' like 'loda' and 'scrive' like 'legge'. If the infinitive ends in '-are', the speaker processes the entire paradigm of any verb ending in '-are' by applying a superordinate rule to the class of verbs with infinitive ending in '-are', a rule different from that applied for the production and comprehension of verbs with infinitive ending in '-ere' or '-ire'. Features such as the infinitive ending in Italian function as structural conditions of paradigms, allowing the production (and comprehension) of correct forms through the automatic application of a rule that operates not at the level of individual constituents but at that of the entire class of words. Languages like Italian or Latin, called inflected languages - those languages, that is, in which words flex into different forms to convey various grammatical information: e.g., lod-o, scriv-o = present indicative 1st singular vs. lod-a, scriv- e = present indicative 3rd singular - do not allow, in short, neither a free combination of stems (e.g., 'lod-', 'scriv-') and inflectional morphemes (e.g., '-a', '-e': third singular of the present indicative), nor a free distribution of inflectional morphemes: each stem, in fact, allows one and only one combination (the Italian 'lod-' of 'lodare' can only take the ending '-a', and not also '-e' or '-i' in the third person singular of the present indicative), and the inflectional morphemes are dependent on each other ('loda' implies 'lodano', 'lodato', 'lodavi', etc.). The inflectional system is an implicational system (Carstairs 1987). Such an arrangement of meaningful units, based on a double implication, depends on a principle of memory economy (cf. Lazzeroni 2005). If in a language every verb and every noun could freely choose an ending for every combination of tense, number, person, case, etc., the possible combinations would be very high in number, and speakers would have to know as many paradigms as there are verbs and nouns in that language: certainly - as in the case of basic color names - their entire lives would not be enough to learn them all. It is no coincidence that so-called "irregular" paradigms (e.g., Italian 'andare', 'essere'), which show the need to be memorized one by one, invariably correspond, in languages around the world, to the forms or words with the highest frequency index: frequency alone is a powerful aid to memory.
The economy of linguistic changes often manifests as an economy of memory as well. A change, when it initially affects only certain linguistic units (let's say, for example, that it affects only certain words), creates irregularity, asymmetry. The new forms are memorized one by one and not automatically produced by applying a rule (e.g., if the infinitive is in '-are', then the third person singular of the present indicative is in '-a'; if in '-ere', in '-e', and so on). But as mentioned earlier, the speaker tends to organize the units affected by the change (and, most of the time, to reorder the residual ones) into a "schema," based on the model of family resemblance. The schema does not allow for automatism, as it does not include a finite number of traits shared equally by all constituents of the category, but it facilitates lexical access through probabilistic inferences. Therefore, the schema does not eliminate the burden of memory because it does not function like a rule, but it reduces it by narrowing down the options.
Furthermore, many changes proceed symmetrically: they pass, that is, from a single categorical constituent to the superordinate taxon of the category (at the level that includes all constituents of that category). Given, for example, three elements 'x1', 'x2', x3, belonging to the category 'X', a change, if it affects 'x1', progressively tends to include 'x2' and x3 as well. The speaker, who previously applied a single operating rule to all constituents of 'X', in the phase where 'x1' (affected by the change) behaves differently from 'x2' and x3, must memorize different rules for each categorical constituent. If the new rule applies not only to the constituent 'x1' but to the entire category, declarative memory is lightened, to the advantage of procedural memory. Examples of symmetric changes are numerous and observed in languages that are very different from each other.
In conclusion: language is a memorized system, and therefore functional to memory mechanisms, aimed also at communication; but, above all, functional to the cognitive organization of experience. This is to say that we speak, first and foremost, with ourselves. Moreover, pragmatically speaking, this corresponds to a desirable reality.
