The Language–Reality Dialogue Cycle: experience → naming → classification → statement → counterargument → translation → redescription → revised experience Language is one of the fundamental media of the human world. We see objects, draw distinctions, form memories, ask for causes, name justice and injustice, write history and plan the future—and language is active in all of these. But to say that language is active is not the same as saying that language is the final creator of everything. Parallel Philosophy insists on this distinction. In this chapter, the word ‘limit’ does not mean prohibition alone; it means a horizon that gives shape. Language makes some distinctions clear and others blurred; it makes some experiences shareable while reducing the subtlety of others. Yet the limit is mobile—new words, translation, science, poetry, law, vernacular speech, multilingual dialogue, images, signs, numbers and counterarguments continually alter its line. Problem The first difficulty is linguistic determinism. If human beings could think only within the categories of their language, it would be difficult to explain new concepts, translation, scientific discovery, language change and deliberately created vocabularies. Language is powerful; it is not omnipotent. The second difficulty is the theory of the neutral label. If words merely attached names to objects already fully formed, why would categories such as ‘crime’, ‘disease’, ‘caste’, ‘border’, ‘merit’, ‘normal’, ‘development’ and ‘risk’ alter social behaviour, institutions and self-understanding? Some classifications do more than describe; they also produce social consequences. The third difficulty is translation. Two languages do not always supply exactly the same meaning, but this does not prove translation impossible. Translation is often not equivalence but a reasoned adjustment among context, purpose, use, cultural signal, and gains and losses of meaning. The fourth difficulty concerns private experience. Pain, grief, love, fear, meditation, taste, bodily fatigue or religious experience may be denser than words. Words make experience public, but the entire qualitative character of experience need not be exhausted by language. The fifth difficulty is silence. Silence may mean dissent, fear, grief, respect, strategy or lack of knowledge. If linguistic analysis attends only to what is spoken, it may miss silence produced by power and unequal opportunities to speak. The sixth difficulty is the plurality of signs. Images, maps, music, mathematical equations, gestures, architecture, clothing, ritual, interface icons and data also carry meaning. Language is a major medium, but the world of meaning is not confined to linguistic sentences. The seventh difficulty concerns the relation between category and person. Administrative categories may be useful, but a person is more complex than any category. ‘Poor’, ‘migrant’, ‘disabled’, ‘minority’, ‘rural’, ‘woman’ and ‘youth’ can reveal real inequalities, but they do not completely define the person. The eighth difficulty concerns power and vocabulary. Whose word is called ‘standard’, whose pronunciation ‘correct’, whose record ‘official’, whose memory ‘story’, whose knowledge ‘folk belief’? Institutional power can operate within such decisions. Linguistic prestige becomes linked with the prestige of knowledge. The ninth difficulty is truth. If every truth is treated only as a linguistic construction, the meaning of falsehood, error, propaganda, documentary contradiction and empirical testing becomes weak. The statement ‘the ice is melting’ is linguistic, but its truth is bound to temperature, measurement and physical processes. The tenth difficulty is semantic change. Reading a historical word in its present meaning produces anachronism. Words such as ‘nation’, ‘caste’, ‘religion’, ‘class’, ‘citizen’, ‘Śrotriya’, ‘modern’, ‘woman’ and ‘Dalit’ may have different boundaries in different periods and sources. The eleventh difficulty is digital language. Search queries, hashtags, prompts, metadata, labels, moderation classes and machine- readable fields alter real visibility and decisions. In the digital age, language is not merely communication; it is also database schema and computational instruction. The twelfth difficulty concerns artificial intelligence. A language model can produce grammatical and contextually appropriate answers, but fluent statements are not automatic proof of truth, understanding, experience or responsibility. Linguistic competence and knowledge claims require separate tests. The thirteenth difficulty concerns the multilingual person. The same person may use different conceptual resources, emotional registers, social identities and memory cues in different languages. The formula ‘one person = one linguistic world’ does not fit the reality of multilingual life. The fourteenth difficulty is disagreement. Two people may use the same word while assigning it different meanings; they may use different words while substantively agreeing. The first task in controversy is therefore not verbal sameness but clarification of claims, evidence, consequences and use. Core Proposition Parallel Philosophy treats language as an instrument of ‘mediated reality’. Reality may possess dimensions independent of language, but shared human knowledge is usually organised through concepts, signs, measurements, names, documents and narratives. Language is therefore both a condition of knowledge and an object of criticism. Language is a limit because every word draws some distinctions and cuts across some continuities. By saying ‘forest’ we identify a region, but the complexity within it—water, soil, microorganisms, animals, human uses, legal boundaries and seasonal change —is compressed into one word. A category is a useful abbreviation, not the whole of reality. Language is not a prison because people can criticise their vocabulary, create new words, change old meanings, borrow from other languages, use metaphor to understand the unfamiliar, and turn to images, numbers, examples or bodily demonstration when words fail. Linguistic humility is a rule of Parallel Philosophy: ‘My statement is not a copy of reality, but a testable presentation of it.’ This does not weaken truth; it increases the obligation to examine the evidential relation between statement and reality. Language also has an ethical dimension. Misnaming is not merely a semantic error; it can become a precondition for insult, stigma, invisibility, denial of rights or violence. Yet respectful naming should not itself become a device for stopping critical discussion. The maxim of this chapter is: ‘A word is both a window and a frame; the frame limits the view, but the window is not a sealed room—we can change our position, open another window, go outside, make a map, and alter the frame itself.’ Twenty-Five Principles of Language, Meaning and Reality 1. A word is not the object; it is a public sign directed towards the object. 2. A sign may be arbitrary, but the rules of its use are not wholly arbitrary—successful communication requires shared distinctions and practices. 3. A useful category requires boundaries; those boundaries can conceal real diversity. 4. Meaning is not merely a dictionary entry; it changes with sentence, context, speaker, listener, purpose, history and practice. 5. ‘One word = one fixed meaning’ is often a false simplicity. 6. Polysemy is not merely an error; it is one reason language can remain flexible. गजेन्द्र ठाकु रक समानान्तर दर्शन — खण्ड २ 7. Ambiguity is sometimes a defect and sometimes necessary openness. Law and science generally require greater precision; poetry and dialogue may benefit from deliberate openness. 8. Translation is not word-for-word sameness; it is an adjustment among semantic function, context, cultural weight and the purpose of the reader. 9. To say that translation is impossible is an exaggeration; to say that perfectly lossless translation is possible is also an exaggeration. 10. Changing a name does not always change the thing; but social objects and institutions can be affected by naming. 11. Language cannot create a physical fact against material evidence; it can, however, change the frame through which the fact is described or measured. 12. Silence is not meaningless; test its meaning through circumstance and relations of power, not through assumption alone. 13. An experience without a name is not therefore impossible; public recognition and policy response may nevertheless become more difficult. 14. A new name can create visibility, but a new category can also create a new stigma. 15. The history of a word is not proof of its present meaning; etymology is part of meaning, not its final judge. 16. A standard language can make administration and education easier, while suppressing the knowledge and prestige of other speech forms. 17. Multilingualism is not conceptual redundancy; it is a resource of alternative distinctions, memories, metaphors and social access. 18. Specialist vocabulary increases precision, but translation is necessary for public accountability. 19. Data do not ‘speak for themselves’; variable names, categories, units, denominators and visualisations are part of their meaning. 20. Algorithmic labelling is a linguistic and technical act with real consequences; a right to reconsider the label should be possible. 21. Abusive language should not be assessed by words alone; examine the targeted person, history, power, repetition, context and probable harm. 22. Free speech does not mean semantic immunity; criticism, counter-speech and responsibility accompany freedom of expression. 23. Language makes experience shareable; it does not exhaust experience. 24. Recognising a linguistic limit is itself the first means of crossing that limit. 25. A philosophy strengthens linguistic self-critique when it maintains a glossary of its vocabulary, records rival definitions and preserves the history of revisions. Pūrvapakṣa 1: Language Is the Final Limit of Our World A strict pūrvapakṣa may argue that thinking or speaking about anything outside language is self-contradictory; therefore talk of a ‘reality’ beyond the linguistic framework is meaningless. Response: statements may necessarily be linguistic, but their causes, resistance and success can be connected with extra- linguistic processes. Fire is not a word; a mistaken judgement about ‘fire’ can burn the body. The reading of a measuring instrument is expressed in language or numbers, but the instrument’s causal interaction does not depend upon linguistic agreement. We cannot provide an ‘undescribed view’ of extra-linguistic reality; this proves epistemic mediation, not ontological imprisonment. Pūrvapakṣa 2: Thought Is Impossible without Language In its strict form, this claim makes many cases difficult to explain—infants, animals, musicians, painters, athletes and forms of spatial reasoning. Some thought may begin without linguistic articulation; language later increases its analysis, memory and shareability. The response is not anti-language. Language is extremely important for complex propositions, counterfactuals, law, mathematical proof, historical debate and institutional rules. But it is an unnecessary extension to identify every form of cognition with sentential language. Pūrvapakṣa 3: Translation Is Always Betrayal Every translation makes selections concerning levels of respect, rhythm, cultural signal, etymology, ambiguity, gender, tense and philosophical technicality. ‘Complete sameness’ is therefore rare. But if the metaphor of betrayal becomes a claim of complete impossibility, the very criticism of ‘translation’ becomes meaningless. Better questions are: which meaning has been preserved? What has been lost? What has been gained? Who is the reader and what is the purpose? What alternative rendering is possible? A critical edition should make uncertainty explicit. Pūrvapakṣa 4: A Social Category Is Merely Constructed and Therefore Unreal ‘Constructed’ and ‘unreal’ are not synonyms. Money, borders, universities, laws, caste or administrative categories, and the institution of marriage are created through human rules and beliefs, yet they produce real consequences. The Parallel response distinguishes levels: biological or physical constraints, social rules, legal designation, self-identification and cultural narrative. Construction at one level does not erase facts at another. Pūrvapakṣa 5: Change the Words and Society Will Change Inclusive vocabulary can reduce stigma, increase visibility and give law a more respectful language. But reform of naming can remain symbolic and empty if resources, rights, opportunities, control of violence and institutional procedures do not also change. Conversely, material reform without respectful language can remain dehumanising. Language and structure are interrelated; neither should be treated as a complete substitute for the other. Words, Concepts and Evidence The first stage in testing a knowledge claim is ‘What exactly was said?’ If the boundaries of a word are unclear, the relation to evidence will also be unclear. ‘Development increased’, ‘violence declined’, ‘the language is disappearing’, ‘artificial intelligence is intelligent’, ‘the community agrees’—each requires an operational definition. The second stage is ‘What exactly was measured?’ Record the variable name separately from the measurement procedure. ‘Poverty’ may mean income, consumption, multidimensional deprivation, subjective insecurity or inclusion on an eligibility list. Do not treat one measure as the whole concept. The third stage is ‘Whose word is it?’ An administrative category and a community’s self-description may differ. In archival sources, a colonial or official label should not simply be imposed upon a later identity. The fourth stage is the ‘rival definition’. Some disputes are about facts rather than definitions; some definitional disputes conceal real conflicts of interest or power. Distinguish the two. The fifth stage is the ‘revision test’. If new evidence appears, can the word or category be changed? If a definition is designed to absorb every counterexample, the argument can become unfalsifiable. गजेन्द्र ठाकु र Indian Philosophical Dialogue Nyāya discussions of verbal testimony take the relation between language and knowledge seriously. The reliability of the speaker, sentence meaning, relations among words and the production of knowledge open a dialogue with modern discussions of testimony and communication. Parallel Philosophy does not turn this tradition into an exact modern theory; it reads it as an independent disciplinary resource for thinking about linguistic knowledge. Mīmāṃsā raises complex questions of sentence, word-power, injunction and textual interpretation. The meaning of a traditional text is not made by isolated words alone; it develops through relations among sentences, purpose, context and interpretive rules. This insight opens a useful dialogue with legal and religious hermeneutics. The Buddhist theory of apoha offers resources for thinking about the negative boundary of a class—‘this, not the other’. Parallel Philosophy does not identify it directly with modern semantics, but treats the relational character of classification as a useful comparative question. Bhartṛhari’s Vākyapadīya presents profound reflection on language, sentence and the unity of experience. It would be a historical error to create a simple identity with the modern linguistic turn; nevertheless, the philosophical seriousness of refusing to treat language as a merely external label is clear. Navya-Nyāya technical language develops extraordinary precision in semantic distinctions. Parallel Philosophy learns from this tradition that an obscure philosophical debate can become more testable by clarifying relation, qualifier, locus, absence, cognition, condition and similar elements. The traditions of alaṅkāra and dhvani show multiple levels of meaning—not merely direct statement but suggestion, resonance, rasa and context. Such literary openness has limits in public policy and science, but literary meaning should not be reduced to dictionary paraphrase. Dialogue with the Western Linguistic Turn Later Wittgenstein emphasises use, forms of life and language-games. The Parallel view does not derive from this a rigid formula that ‘meaning = use alone’; rather, it asks: in what practice does the statement function, how was the rule learned, and how is incorrect use recognised? The resistance of reality can remain an external criterion for the practice. Quine raises the problem of indeterminacy of translation and conceptual schemes; Davidson, through radical interpretation, questions extreme scheme-relativism. Parallel Philosophy learns from this debate that even where translation is underdetermined, dialogue requires a shared world, shared behaviour and conditions of charity. Gadamer emphasises tradition, fore-understanding and the fusion of historical horizons. Meaning is not each reader’s private invention; it opens in dialogue among text, tradition, question and the present horizon. Derrida exposes the instability within texts, binary hierarchies and simple metaphysics of presence. The Parallel response is that deconstruction can be a necessary critical instrument, but without reconstruction of judgement, evidence, harm and institutions it is not a final political or epistemological method. Foucault exposes relations among discourse, classification and knowledge–power. Parallel Philosophy accepts this insight, but does not turn power analysis into a shortcut for automatically rejecting truth claims; documents, causal mechanisms, counter- evidence and comparative explanation remain necessary. Searle, Grice and traditions of speech-act theory and pragmatics show that utterances do not merely state propositions—they promise, command, apologise, declare, warn, imply and perform irony. Within institutions, ‘words’ can become legal or social events, but felicity conditions and authority must be examined. Mithila and Multilingual Historical Experience Mithila’s intellectual history is not monolingual. Sanskrit, Maithili, Avahaṭṭa/Apabhraṃśa traditions, Persian and Urdu administrative influence, Hindi, Nepali, English, Tirhuta and Devanagari scripts, and local speech forms opened different routes to knowledge in different periods. It is inappropriate to impose the modern standard of one language upon the whole experience of the past. When reading manuscripts, distinguish word-forms, scripts, editorial normalisation, later copies and printed editions. Modern ‘correction’ of spelling can sometimes erase historical evidence. It is safer to provide diplomatic transcription and normalised reading as separate layers. Genealogies, Panji records, inscriptions, folk memory, songs, letters, judicial records and literary compositions are not the same type of source. Their evidential weight should vary according to linguistic form, purpose, authorship, audience and institution. Do not retroject the later fixed meaning of a title, caste-name or designation onto a medieval document. Examine separately the earliest attestation, contextual use, semantic range and institutional stabilisation of the term. The aim of Maithili philosophical writing is not merely Devanagari transliteration of Sanskrit or English terms. Wherever possible, use living and idiomatic Maithili syntax, defined technical terms, a glossary and the necessary original term, so that both precision and linguistic naturalness survive. Language, Power and Justice Whose speech is called ‘wrong’ at school? Whose accent is judged less ‘professional’ in a job interview? Whose name does not fit a form or database? Linguistic justice is not merely literary respect; it concerns access to education, employment, courts, health care and digital services. Standardisation should not be rejected altogether. Standards are useful for administration, spelling, examinations, dictionaries, search, accessibility and interoperability. The problem begins when a standard exceeds its practical function, claims social superiority and turns other forms into signs of ignorance. Linguistic rights have two levels: (1) reasonable access to expression, education and services in one’s own language or script; and (2) sufficient competence in a common public language for wider civic and economic participation. One right should not destroy the other. Context is necessary in regulating abusive language. Not every unpleasant statement is violence; but targeted harassment, credible threats, dehumanising campaigns and institutional discrimination cannot be dismissed as ‘only words’. Public institutions should make translation and interpretation policy transparent: who is served, in which languages, which translation is authoritative, how errors are corrected, and what receives priority in emergencies. The Parallel Method of Translation 1. First establish the source text, edition, page and script. 2. Record the possible semantic range of a word; do not immediately close it around one modern equivalent. 3. Understand syntax at sentence level; do not rely on word-for-word substitution. 4. Maintain a glossary for technical terms; do not use five unnecessary renderings for the same term. 5. Where ambiguity in the source is meaningful, do not give the translation a false precision. 6. Transliterate a cultural item and add a brief explanation when a local term would otherwise be lost. 7. Render proper names and titles according to historical evidence; do not infer a modern identity. 8. Distinguish quotation, narrator, editor and translator voices. 9. In poetic or literary texts, acknowledge or note losses of rhythm and voice; do not treat literalness as the only form of fidelity. 10. In legal and philosophical texts, give priority to semantic precision; do not remove conditions for the sake of elegant paraphrase. गजेन्द्र ठाकु रक समानान्तर दर्शन — खण्ड २ 11. Back-translation is only a diagnostic tool, not the final proof of quality. 12. Document disagreements among bilingual reviewers or subject specialists. 13. Machine-translation output may be a draft; it should not become an authoritative translation without human contextual review. 14. Preserve the edition history of revised translations. 15. Do not stop the work by saying ‘translation is impossible’; do not hide uncertainty by saying ‘translation is complete’. Digital Language, Search and Artificial Intelligence In a search engine, the language of the query creates visibility. Different spellings, scripts or romanisations can make the same material invisible. Digital archives therefore need variant spellings, transliteration fields, language tags and metadata crosswalks. A database category is a linguistic decision. A gender field, caste category, district name, occupation code, disability type or language code can make a real user fit or fail to fit the schema. The category ‘Other’ can be both a convenience and a form of invisibility. Prompt language changes artificial-intelligence output. A model may not maintain the same factual reliability across languages. Evaluation should separately measure language-specific benchmarks, source availability, script handling, hallucination rates and cultural bias. Generative artificial intelligence can erase distinctions among sources through fluent paraphrase. In archival and research workflows, quotation, paraphrase, inference and model-generated synthesis should be kept separate through clear provenance labels. A claim that artificial intelligence ‘understands meaning’ may have behavioural evidence, but it is not final proof of phenomenal understanding. Success at language tasks, world-grounding, error correction, memory, planning and responsibility are distinct dimensions. Deepfake and audio synthesis alter the nature of linguistic evidence. More than asking ‘whose voice is this?’, one must examine provenance, the recording chain, metadata, independent confirmation and forensic uncertainty. A Twelve-Step Cycle for Testing Linguistic Claims 1. Quote or restate the claim exactly. 2. Define the key terms. 3. Test whether meaning changes across time, place or community. 4. Identify the purpose of the speaker, author or institution. 5. Determine whether the statement is fact, value, command, inference or metaphor. 6. Test whether the supporting evidence is independent of the linguistic claim. 7. Construct a counterexample or alternative wording. 8. Note uncertainty concerning translation or edition. 9. Investigate the cause of silence or absent voices—destroyed sources, lack of access, censorship, literacy barriers or archival bias. 10. Examine the real consequences of the category—eligibility, stigma, visibility and resource allocation. 11. Ask whether revised vocabulary clarifies the claim or merely creates a euphemism. 12. Record the confidence level of the conclusion and the conditions under which it should be revised. Twenty Application Scenarios Scenario 1: a census offers only limited mother-tongue options, so people choose the nearest category. Later, the numbers are treated as complete proof of ‘real linguistic identity’. That is wrong. Examine questionnaire design, coding rules and constraints upon self-reporting. Scenario 2: in a hospital, a patient expresses pain in a local idiom while the doctor asks for a standard scale. The translation gap may cause severity to be underestimated. Add bilingual explanation, gesture and context. Scenario 3: in court, ‘yes’ may be a respectful or pressured response rather than informed consent. Examine the power relation carried by language. Scenario 4: an ethnic designation in an old manuscript is read in the modern sense of a sub-caste. Suspend the conclusion until contemporaneous documents and semantic history are examined. Scenario 5: a school punishes a student’s home speech as ‘incorrect Maithili/Hindi’. Teach standard literacy, but do not call the home speech a cognitive deficiency. Scenario 6: a government programme is described as ‘universally accessible’, while its form exists in only one language and uses complex bureaucratic terminology. Nominal availability is not real access. Scenario 7: a social-media hashtag compresses the many demands of a movement into one label. Trend count is not substantive agreement. Scenario 8: the word ‘merit’ is used to mean examination score, but moral deservingness enters the debate unnoticed. Identify the shift in definition. Scenario 9: ‘likely’ in a climate report expresses a calibrated technical probability, but a news report reads it as casual uncertainty. Translate calibrated language accurately. Scenario 10: an artificial-intelligence system labels a user ‘high risk’. Publish the label’s threshold, training target, base rate, error profile, consequences and right to reconsideration. Scenario 11: a modern translation of one word in a religious text creates controversy. Examine manuscript variants, commentary traditions, grammar, ritual context and alternative translations. Scenario 12: a farmer’s local soil name does not exactly match scientific taxonomy. Do not discard one as ‘unscientific’; create a mapping and test predictive usefulness. Scenario 13: a company uses the metaphor of ‘family’ to expect unpaid overtime. The power effect of the metaphor is open to criticism. Scenario 14: a disability community changes its preferred terminology. Respect its self-description, while separately testing objective needs relevant to service eligibility. Scenario 15: a history book chooses ‘rebellion’ rather than ‘disturbance’ for the same event. Framing alters causal and moral interpretation. A source-based rationale is required. Scenario 16: a bilingual child performs poorly on a test in one language and is declared to have low overall cognitive ability. Measure language proficiency separately from conceptual knowledge. Scenario 17: a literal translation of a poem makes the meaning intelligible but destroys poetic effect. If the purpose is literary, multiple renderings or commentary may be appropriate. Scenario 18: an old district name fails to match a new name in a database, so records appear ‘missing’. Authority-file and alias mapping are part of linguistic infrastructure. Scenario 19: in public debate, two sides use the word ‘freedom’ differently. List the concrete rights, constraints, affected actors and trade-offs involved. Scenario 20: Parallel Philosophy itself develops so much technical vocabulary that readers are excluded. Apply Chapter 96’s self- test: simplify the glossary and increase examples without destroying precision. गजेन्द्र ठाकु र Language Policy in Educational and Research Institutions Language policy in universities and journals should balance accessibility and quality. Publication only in a dominant language can increase research visibility while weakening local knowledge production. Bilingual abstracts, terminology indexes, translation grants and regional-language peer review are useful options. In archival cataloguing, preserve the original title, normalised title, transliteration, language and script tags, and alternate names in separate fields. Simplification of metadata can make searching easier, but should not erase the original evidence. In citation, make clear the relation between translated and original titles. A quotation from a secondary translation should not be presented as though it came directly from the primary source. In oral history, the interviewer’s language shapes the response. Record question wording, code-switching, translation decisions and transcript normalisation in the methodology. Language revitalisation requires more than a dictionary—children’s education, literature, digital input, speech technology, community use, intergenerational transmission and prestige are all necessary. A project of linguistic ‘purity’ can become rigid purism against documented historical variation. Standardisation can be useful, but it needs a transparent basis in corpus evidence, living usage and historical forms. Limits and Self-Critique The first limit is linguistic overreach. Not every social problem is an effect of discourse; land, wealth, weapons, health, environment, law and infrastructure involve material causation that requires separate analysis. The second limit is naïve realism. It is also an error to say ‘the object is simply there’ and deny the roles of measurement, category, observer position and conceptual scheme. The third limit is romanticism about translation. Multilingualism is not automatically inclusive; examine the prestige of dominant languages, unequal literacy, translator authority and resource gaps. The fourth limit is jargon. Unnecessary foreign terms can exclude readers in the name of philosophical precision. Where a clear Maithili equivalent exists, give it priority; where a technical distinction would be lost, preserve the original term in brackets. The fifth limit is the euphemism cycle. When a word changes, stigma can migrate to the new word; lexical reform remains short- lived if structural prejudice does not change. The sixth limit is the censorship shortcut. Harmful speech is a real problem, but suppressing disagreement or criticism as ‘wrong language’ weakens democratic dialogue. Rules should be narrow and evidence-based. The seventh limit is textualism. Do not turn oral practice, gesture, material culture, archaeology, ecology, embodiment and quantitative evidence into footnotes to text. The eighth limit is model-centric language. The fluency of AI-generated language is not a complete model of human meaning or experience; system-level analysis must include human users, institutions, datasets and interaction with the world. Chapter Conclusion ‘Language is a limit, but not a prison’ is Parallel Philosophy’s maxim of linguistic humility. Shared philosophy is difficult without language; reality does not end in language. Words give us routes towards objects, experiences and society, but every route carries selection, reduction and point of view. The philosopher’s task is therefore not to become the police of words, but to clarify limits of meaning, identify historical change, state the gains and losses of translation, attend to silence and power, and test claims against extra-linguistic or independent evidence. A completely language-independent standpoint may lie beyond human reach; yet through many languages, many methods, measurement, experience, counterargument and redescription, we can avoid becoming prisoners of a single vocabulary. Chapter 96 opens self-critique, Chapter 97 says that structure is not destiny, and Chapter 98 applies this idea to language: linguistic structure is real, but people can alter its limits through translation, neologism, education, art, science, criticism and collective use. The combined maxim of this chapter is: ‘Words show us the world; judgement gives us the capacity to change the words and see the world again.’