Pack 0 Text 2 is a live community-radio interview that must publicise a new service, explain practical procedures and negotiate potentially face-threatening questions about entitlement. Jules manages topic progression for listeners who cannot see the lockers. Samira contributes institutional knowledge, while Priya and Damo introduce access and fairness concerns. The interaction therefore combines the immediacy of broadcast conversation with the precision required for public information.

Jules establishes the live mode through the temporal deictic expression “seven forty-two” (line 1) and the sensory reference to “rain on the studio window” (line 1). His announcement of “a different arrival this Friday” (line 2) links the new service to the station setting while delaying its identification. The open interrogative “what are we getting/” (line 3) then allocates Samira a turn and frames her as the relevant expert.

Samira first makes the service concrete. The noun phrase “twelve pantry lockers” (line 4) supplies scale, and the locative clause “beside the ticket machines” (line 5) uses familiar station infrastructure as a landmark. Her adversative addition that the lockers “don't need a ticket” (line 5) plays on the railway semantic field while removing a possible access assumption. The present-tense material process in “doors unlock all night” (line 6) foregrounds availability.

The list “bread (.) rice (.) tins (.) cereal (.) long-life milk” (line 8) uses parataxis and pauses to create broadcast-friendly units. Samira follows with the inclusive instruction “take the items that are useful” (line 10) rather than defining a category of eligible person. This orientation becomes explicit when the indefinite pronoun “anyone” (line 12) answers Jules's question about permission. Three parallel negatives - “no form (.) no membership card (.) no proof of income” (line 13) - make absent barriers salient, while “no explaining your week to a stranger” (line 14) names the face threat those barriers could produce.

Priya's report that her neighbour “reads English slowly” (line 15) shifts the discussion from entitlement to interface design. Samira's response “this should feel different” (line 17) acknowledges comparison with difficult online forms without claiming universal success. The capitalised button labels “FOOD or HOUSEHOLD” (line 18) and the ordered clause “then choose a locker number” (line 18) represent the screen as a short decision sequence. This is accessible language tied to a multimodal feature rather than reassurance alone.

Damo marks his later question as delicate through the preface “a bit awkward” (line 23) and the idiom “doing it really tough” (line 24). Samira protects his positive face with “thanks Damo” and the generalisation “people do wonder that” (line 25). She then rejects gatekeeping through “no test” and the compound “deserving-enough category” (line 26). Her imperative “use it” (line 27) is direct, but its function is authorisation rather than restriction.

The challenge about one person clearing a row is not dismissed. Modal “could” in “that could happen” (line 29) concedes possibility before the coordinator explains the sign and refill process. The antithesis “an empty locker means food was useful (.) not that the service failed” (line 31) repairs the negative evaluation through semantic reframing. Jules's reformulation, “measuring supply rather than judging a person” (line 32), checks shared understanding and translates the policy for the wider audience.

Privacy is made memorable through parallel ellipsis. “STOCK tally yes (.) people tally no” (line 35) combines emphatic stress, lexical repetition and balanced opposites. The following declaratives distinguish counting objects from building a personal history (lines 36-37). Samira later treats access with similar specificity: “orange shelf labels” (line 39), forward-facing ingredient panels (line 40), “high-contrast” display settings (line 44) and “spoken instructions” (line 44) identify concrete resources across visual and auditory modes.

Reciprocity remains optional. The modal “can” in “you can add a tin” (line 48) encodes possibility, the adjective “lovely” values contribution, and the explicit clause “it is not a debt from today” (line 49) blocks a transactional inference. Jules closes with a compressed recap of location, opening time and access features (lines 50-52). Samira's assertion that “privacy isn't an extra feature” (line 53) connects dignity to functional design, while the balanced imperatives “take what helps” and “leave judgement at home” (line 54) give the segment a memorable interpersonal conclusion.

Across the text, interrogatives and reformulations organise a live exchange, while lists, explicit deixis and repeated policy contrasts serve listeners who need retrievable information. Humour and conversational particles maintain local-radio familiarity without replacing precise safety and access detail. The speakers construct community support as a practical public service in which assistance, privacy and responsibility can coexist.

Sample choice: Question 3(iii)

Language technologies do more than preserve utterances or retrieve existing writing. They segment language into recognisable units, rank some forms as more probable and offer users particular ways to repair or continue a text. These operations can widen access, yet they also make already prominent lexical, grammatical and discourse patterns easier to find, predict and reproduce. The influence is not absolute: users can reject a suggestion and designers can preserve alternatives. Nevertheless, technological convenience changes the practical cost of choosing one form over another.

Question 3(iii) Stimulus A describes autocomplete as a "quiet editor". The metaphor is useful because prediction appears while a writer is still composing, before an explicit human editor evaluates the draft. Its second clause, "it rewards patterns already made visible", links probability to prior representation. A high-frequency collocation can be offered after only a few characters, while a regional expression or an unfamiliar proper noun may require full typing and repeated correction. The interface therefore gives some graphological and lexical sequences an efficiency advantage without announcing a language rule.

Question 3(iii) Stimulus B shows a related problem at the boundary between speech and a written record. The fictional service transcribes a "formal briefing" accurately but alters "Aboriginal English grammatical patterns" in a community interview. The contrast indicates that recognition quality is socially distributed rather than simply high or low. When the altered transcript becomes searchable, later readers encounter the system's standardised reconstruction instead of the speaker's original grammar. Recording has therefore become reproduction: the technology changes which variety enters the durable record.

These effects extend beyond individual words. Predictive systems can favour conventional email openings, institutional nominalisations or familiar clause sequences because those patterns are well represented in training data. Such assistance may help a novice enter an unfamiliar genre, but it can also make one professional tenor effortless while a locally meaningful voice demands more manual repair. Transparent alternatives, editable personal dictionaries and a visible distinction between machine output and human correction can retain assistance without presenting frequency as correctness.

A verified Australian example appears in the Digital Transformation Agency's 2 September 2024 account of responsible AI in government. The article uses the institutional noun phrases "accountable officials" and "AI transparency statements" for roles and documents that agencies must be able to locate consistently. Standardised labels make responsibility searchable across the Australian Public Service. The same page identifies the simpler title "Guidance for using generative AI". The gerund-participial verb "using" states the document's practical function more directly than a densely nominalised title, improving likely recognition for non-specialist searchers.

The Australian Government Style Manual's plain-language guidance, updated 19 December 2024, demonstrates how visibility can be widened without abandoning technical precision. It introduces "large language models (LLMs)" before using the initialism. The parenthetical expansion supplies the full noun phrase at first mention, while the shorter form remains available for later cohesion and efficient searching. This design choice does not wait for a prediction system to guess what a reader knows; it deliberately creates a lexical bridge between expert and public vocabularies.

Language technologies therefore influence language through recognition, ranking, correction and storage. Question 3(iii) Stimulus A identifies the quiet pressure of prediction, while Question 3(iii) Stimulus B shows how unequal recognition can reshape a public record. The most equitable tools do not pretend to be neutral mirrors. They make automated transformations visible, support repair and keep legitimate variation retrievable alongside standard forms.