**Links to**: [[Meta-language]], [[Language models]], [[Modulations]], [[Language is music]], [[Pronoun]], [[Metalearning]], [[Neologistics]], [[Dialove]], [[Dialogics]], [[E Pointing]], [[Reference]], [[Self-reference]], [[03 Semantic noise]], [[Coming abstractions]]. ### [[Postulate]]: Besides the obvious _sonic_ aspects of natural language when produced as vocal speech, the _redundancy_ of language in general ensures that many more aspects of it can be measured/understood as frequencies. This cracks language open in an interesting way: it becomes more directly accessible as a series of patterns which can be modulated (see [[Language is music]], [[Song]], [[Music as permanent revolution]], [[03 Semantic noise]], [[Language models]] for different ways to access this statement). >... the root function of language is to control the universe by describing it. > >J. Baldwin, “Stranger in the Village.” _Notes of a native son_ (1955). Examples of what is meant by possible, currently accessible modulations can be found in: [[Modulations]]. Natural language when produced as vocalized speech can be easily understood as music: much of what we do on a daily basis, when we have a mouth that is able to produce sounds, is emphasize meanings and affective states through cadence, volume, tone, etc. Though, strangely, we tend to prioritize the representational and indexical in speech, rather than the affective-sonic. However: we are clearly easily bored or challenged when someone is “too monotone”, for example. The famous [[Robot voice]] is an entertaining example of this. **But**: even if we discard the sonic, and focus simply on the frequency of repeated words or other natural language patterns—which is what ensures that meanings are somewhat preserved—we can focus on aspects of language which can become easily modulated. Once we see the rhythm of words as that which can be measured and understood as frequencies,^[A foundational example of but one approach to this is Spärck Jones’ tf–idf, which she published in “A Statistical Interpretation of Term Specificity and Its Application in Retrieval.” Journal of Documentation. 28 (1): 11–21, 1972.] we crack language open in an interesting way: it becomes more directly accessible as a series of patterns which can be modulated (see [[Language is music]], [[Song]], [[Music as permanent revolution]], [[03 Semantic noise]], [[Language models]] for different ways to access this statement). Counting words and other types of patterns, and chunking them into different categories (i.e., patterns) gives us some of the _measures_ of language, which are fundamental constraints capable of revealing-creating more patterns. One can then, for example, count how many times one says “I” in a day (or any other pronoun). Do that for enough days and observe the rhythm: combine with other data and see what kinds of days urged the first-person pronoun more than other days or other pronouns. Count how many questions. And how many had answers, or led to more questions. Count how many colors, how many references to time, how many spatial metaphors, etc. etc. etc. Keep tallies on these things and a whole day/week/month/year/historical event can be listened to as music. Having even a basic level of access to these patterns creates a meta-perspective from which to understand what one is doing with certain aspects of language. All of the above is but one way in which language-modulating can happen. More examples in [[Modulations]] N.B.: This is not strictly, but can include, the kind of modulation Deleuze offers in the Postscript, see [[11 Post-Control Script-Societies]]. We follow Hui (2015) on a call to reappropriate the concept of modulation affirmatively: >To re-appropriate the concept of modulation is, on the one hand, to acknowledge and deepen its philosophical significance as a materialist alternative to hylomorphism, and, on the other hand, to imagine new modes of modulation that don’t simply fall into the logic that Deleuze and others have described in terms of ‘control’. What we would absolutely want to avoid would be implicitly to propose a return to hylomorphism as a mode of resistance to social control modulation. Nor would it be appropriate to let go of the concept of modulation, with all of its philosophical usefulness (particularly as deployed by Simondon), simply because of Deleuze’s polemical characterisation of modulation as a key feature of control societies. Instead, it is the final aim of this essay to suggest exploring the concept of modulation under the motif ‘modulation after control’, getting beyond the limits of what we might call ‘the modulation-control correlation’. > >Hui 2015, p. 87. ### Footnotes %% “In her often-referenced paper from 1972, Spärck Jones (2004) arguedthat almost any ‘real’ document collection will have to deal with the factthat some keywords have low specificity in the sense that they apply to alarge percentage of documents and are therefore not very good for retrievalpurposes. If a search query and a document have a very frequent word incommon in a basic coordinate indexing setup, this match would count asmuch as a match of an infrequent word, even if the latter is likely to be more‘semantically focussed’ (Manning and Schütze, 1999, p. 542). Since throw-ing out frequent terms would create problems for searchers using broadqueries, one should rather ‘exploit the good features of very frequent andnon-frequent terms, while minimizing their bad ones’ (Spärck Jones, 2004,p. 498). This could be achieved through a weighting scheme that capturesthe significance and specificity of individual keywords by situating themwithin the overall collection. Spärck Jones discusses two strategies: onecould adjust the weight given to a term by counting how often it appears in the overall collection (collection frequency), but since this may not be feasible for many indexing situations where full frequency numbers are not available, one could simply consider the number of documents (document frequency) a term appears in at least once. The following passage argues that choosing between these two strategies is not merely a matter of feasibility and shows how a simply counting decision can raise deeper epistemological concerns: >Weighting by collection frequency as opposed to document frequencyis quite different. It places greater emphasis on the value of a term as ameans of distinguishing one document from another than on its value asan indication of the content of the document itself. The relation betweenthe two forms of weighting is not obvious. In some cases a term may becommon in a document and rare in the collection, so that it would beheavily weighted in both schemes. But the reverse may also apply. Itis really that the emphasis is on different properties of terms. (Spärck Jones, 2004, p. 499) While Spärck Jones certainly makes claims about the general thrust ofeach counting method, she quickly adds that the semantic and pragmaticperformativity of each measure will, in fine, depend on the specific collectionand task it is applied to. Since collection frequency required considerablygreater logistical effort, her experiments focused on document frequencyas a means to attribute higher weight to less common search terms. Thiswork lays the foundation for the notion of ‘inverse document frequency’,where the specificity of a term is understood as a function of its rarity ina collection. If each keyword is adjusted accordingly, 12^[Inverse document frequency is normally calculated by dividing the overall number of documents by the number of documents a term appears in and taking the logarithm of the result. If a collection has 100 documents and a term appears in 5, its weight will be 1.3. If a term appears in 50 documents, its weight is 0.3.] matches for morerare or ‘specific’ keywords will thus receive a higher value in the rankingof search results. Salton later expands the idea, creating what is certainlythe most well-known term-weighting scheme in information retrieval andbeyond, tf-idf (term frequency-inverse document frequency), where boththe frequency of a term in a document and its rarity in the collection areused to calculate its weight (Salton, Wong, and Yang, 1975, p. 616). Oncethe weighting scheme is applied, all other calculations – for example, toestablish similarity – can proceed as they did before.” Rieder 2020, pp. 224-5. Further topics of investigation (from Phil of AI giusy open call 2024) Textuality and language models; Intertextuality and language models; Digital texts and AI-generated texts; Linguistics and AI; Language models in journalism; Language models and creative writing; AI research on language; Learning, creativity and AI; Meaning and AI; Creativity, machine-learning and language; Truth, post-truth and AI.