Months to Working Marain: Time to Proficiency Among 207 Contact Recruits from Other Civilisations, by First Language and Exposure Setting, 1650–1790 CE

1. Introduction

Contact has invited people from other civilisations to join it throughout the period we study (Ferrowen, 1771 CE), and they arrive with languages that share little with Marain, the language in which the Culture's citizens, drones and Minds speak to one another. In the records we use, joining was by invitation and by choice, and Marain was not a condition of it. A drone or a Mind can interpret for a newcomer in any language Contact knows. Recruits nonetheless learn Marain, because Contact's work is done in company, and a recruit who cannot follow a briefing or a disagreement at ordinary speed remains a guest at the edge of their own work (Vorrenholt, 1782 CE).

How long that takes has never been measured across the settings in which recruits learn. Contact's liaisons prepare some recruits before they arrive (Quillon-Aert, 1668 CE). Accounts of adult newcomers on Orbitals (Dovelle, 1704 CE; Ostrevane, 1729 CE) and aboard ships (Marrowen, 1757 CE) describe how learning goes for a handful of people each, seen in one setting and chosen by the author for interest. None applies one yardstick to several settings, and none asks whether the first language matters. Opinion on that question is firm and divided. Some hold that languages with rich word-building give a lasting advantage, and others that no first language outweighs hours of conversation (Cassaundre, 1745 CE).

This paper uses a register of Marain sittings for recruits who arrived at Chiark Orbital or were placed aboard ships between 1650 and 1790 CE. We write in 1800 CE, and no source we use is later than 1795 CE. We ask how many months recruits take to reach working Marain, how that time differs with the recruit's first language and with the setting of their exposure, and whether the differences remain when prior multilingualism is taken into account. We fixed the main model, with those three predictors, before we looked at outcomes. The check on proportional hazards was added afterwards.

Two terms need fixing at the start. By family we mean a coarse grouping of first languages by how they build words, which Contact's liaisons recorded at recruitment. It is the paper's own grouping and makes no claim about descent. By working Marain we mean level 3 on the scale described below, which is also our own.

2. Methods

The register joins two sources. The Marain Studies Group on Chiark Orbital kept notes on every sitting it held for Contact recruits (Chiark Orbital Hub, 1650–1794 CE), and the Hub kept an index of arrivals, placements and departures (Chiark Orbital Hub, 1650–1795 CE). Contact does not publish all of its recruitments. The register therefore covers the arrivals whose records Contact agreed to open to us, and we cannot say what share of all recruits that is. The series designation and the joined register are ours, assembled for this study from notes that the Group and the Hub had kept separately.

Altogether the register lists 226 recruits. We excluded 19: 11 whose notes document more than six months of Marain exposure before structured exposure began, and 8 who left before any sitting. That leaves 207 recruits for analysis (91.6% of those registered). The 207 arrived in 34 intakes, which we define as the groups of recruits who reached Chiark Orbital or boarded together. Intakes held between 1 and 10 recruits.

Exposure setting had three levels. Pre-arrival tuition meant that a Contact liaison taught the recruit at home before travel, with Marain as the main subject, and the recruit then came to Chiark Orbital for sittings (44 recruits). Orbital residence meant living on Chiark Orbital among its residents from the first day (99). Shipboard service meant living aboard a ship and its mixed company of residents, sitting when their ship called at Chiark Orbital or when assessors travelled aboard (64). Recruits did not draw lots for these settings. They were placed by their own wishes and by what Contact could arrange, and one of us (Hethmar) served aboard the ship that hosted 11 of the 64.

First language was coded by the recruit's liaison before any sitting, into four groups of our own devising. Agglutinative languages build words by chaining affixes (59 recruits). Inflecting languages build them by changing word forms (66). Isolating languages rely on word order and particles (49). Multimodal languages are carried by a medium other than sound alone, such as colour, gesture or touch (33). Multimodal speakers took their sittings through a voice aid or in written form, as they chose. A further binary variable recorded whether the recruit had used three or more languages before joining (86 recruits).

The outcome was the time from the first day of structured exposure to the first sitting rated at level 3 on the Working Marain Scale. The scale has five levels. Level 1 is a set of phrases for requests and greetings. Level 2 is everyday exchange that needs frequent repetition. Level 3, working Marain, is taking part in a briefing or a disagreement between Culture citizens at ordinary speed, with the register suited to the company, which is a criterion of our scale. Level 4 is fluent speech. At level 5 listeners no longer remark on the recruit's speech. A sitting was an hour's conversation with two assessors from the Group, in speech or writing as the recruit chose, offered about every three months and taken by invitation. Recruits were free to decline or to ask for a later date.

Because sittings were occasional, the date of the first level 3 sitting is later than the date on which a recruit reached the level, by an interval we cannot observe. Times are therefore upper bounds on attainment. We measured times in days and report months of 30.4 days. Follow-up ended at 48 months from the start of exposure. Recruits who left Contact or the Orbital's records before then, or who were still below level 3 at 48 months, were censored. The last arrival in the register was in 1790 CE, so every recruit had four years of possible follow-up before 1795 CE.

For reliability, a second assessor rated the recorded first sittings of 64 recruits chosen at random, without sight of the original ratings. We report quadratically weighted kappa on the five levels, with a bootstrap interval from 2,000 resamples, together with exact agreement, agreement within one level and agreement on whether the sitting reached level 3.

We estimated the time to level 3 with Kaplan–Meier curves, giving medians with log–log intervals, and modelled it with a Cox proportional-hazards regression (Breslow handling of ties) containing family, setting and multilingualism together. The reference group was agglutinative, pre-arrival tuition and fewer than three prior languages. Recruits who arrive together share teachers, companions and perhaps luck, so standard errors were clustered by intake with a correction for the 34 clusters, following the method for grouped arrivals set out by Lindqvarn (1767 CE). Tests are two-sided Wald tests. With 172 events and six coefficients, the model is not overfitted. To check proportionality we fitted the model separately to the first 10 months and to recruits still at risk after 10 months, and compared each coefficient across the two fits.

3. Results

Of the 207 recruits, 172 (83%) reached level 3 and 35 were censored: 34 before 48 months and 1 at 48 months. The overall median time to level 3 was 10.3 months (95% CI 8.7 to 11.6). By the Kaplan–Meier estimate, 59% had reached level 3 at 12 months, 87% at 24 months and 97% at 36 months. Table 1 gives the figures for each group.

[object Object]

Agglutinative and inflecting speakers had the same median, 8.2 months. Isolating speakers took 11.6 months (8.9 to 15.2), and multimodal speakers took 21.7 months (11.9 to 27.6), with an interval wide enough to reflect only 25 events. At 24 months, 94% of agglutinative and inflecting speakers had reached level 3, against 90% of isolating and 61% of multimodal speakers.

Medians fell across the settings, from 14.4 months after pre-arrival tuition to 10.3 on the Orbital and 7.6 aboard ships. The intervals overlap, and the adjusted model below is a better guide to which differences are firm. Recruits with three or more prior languages reached level 3 in 7.7 months (5.1 to 10.1), against 11.8 (9.6 to 14.9) for the others.

The family and setting groups were not balanced against each other. Cell sizes in the cross-tabulation ran from 6 recruits (isolating speakers given pre-arrival tuition, and multimodal speakers aboard ships) to 32 (agglutinative speakers on the Orbital). Of the 33 multimodal recruits, 14 lived on the Orbital, 13 had pre-arrival tuition and 6 served aboard ships.

[object Object]

Table 2 shows the adjusted results. A hazard ratio below 1 means that, at any time since the start of exposure, a recruit who had not yet reached level 3 was less likely to do so in the next interval than a comparable recruit in the reference group. Multimodal speakers reached level 3 at 0.31 times the rate of agglutinative speakers (95% CI 0.19 to 0.50), and isolating speakers at 0.52 times (0.36 to 0.76). The inflecting coefficient of 0.75 (0.54 to 1.04) did not differ detectably from the reference. Family as a whole was strongly associated with the outcome (Wald χ²(3) = 24.8, p < .001).

Setting was also associated with the outcome (Wald χ²(2) = 15.5, p < .001). Recruits living on the Orbital reached level 3 at 1.57 times the rate of those given tuition before arrival (1.10 to 2.26), and recruits aboard ships at 2.31 times (1.52 to 3.50). Recruits with three or more prior languages did so at 1.83 times the rate of the rest (1.37 to 2.44). These ratios are adjusted for each other.

Clustering changed little. Cluster-robust standard errors were between 5% and 14% smaller than the naive errors for the six coefficients, and we report the clustered values because the design requires it, although they were slightly smaller than the naive errors. In the proportionality check, 94 events fell in the first 10 months and 78 among the 90 recruits still at risk afterwards. No coefficient differed detectably between the two fits, and the smallest p-value for a difference was .33. With so few events in each half, the check could detect only large departures, and it found none.

The second assessor's ratings agreed well with the first. Weighted kappa was 0.86 (95% bootstrap CI 0.78 to 0.91). Assessors gave the same level at 35 of 64 first sittings (55%), a level within one of each other at 61 (95%), and the same verdict on whether level 3 had been reached at 55 (86%).

4. Discussion

In this register, most recruits reached working Marain within a year or two, and a substantial minority did not reach it in the first year. At 12 months 41% had yet to be recorded at level 3. A recruit sent into the field after a year of exposure should be expected to meet a good many briefings and disagreements that they cannot yet join at ordinary speed, and the median of 10.3 months is a guide to planning, not a target.

First language mattered mainly at the ends of our grouping. Agglutinative and inflecting speakers did not differ detectably. Isolating speakers were slower by about half, and multimodal speakers by about two thirds, at any point in follow-up. We cannot say why. The multimodal result may reflect distance between the languages, the demands of speaking through a voice aid or in writing, or the way our assessors judged a register that is heard. The register has no measure that could separate these.

The ordering of settings, with tuition before arrival slowest, the Orbital faster and ships fastest, fits the view held by some that hours of talk matter more than lessons (Cassaundre, 1745 CE). We did not measure hours of talk, and recruits were placed by their own wishes and by Contact's arrangements. A recruit who chose a ship may have been a person already more at ease with strangers, and one who took tuition at home may have had less time to give. The ordering is an association, and we offer the account of conversation as a hypothesis that the register could not test.

Prior multilingualism was associated with a hazard ratio of 1.83, which means that speakers of three or more languages reached level 3 at nearly twice the rate of the rest at any point in follow-up. Whether that is skill at learning, familiarity with a range of structures or simply practice at being a newcomer, we do not know.

These findings hold for recruits at the level of working Marain, and we draw nothing from them about fluency or about how recruits use the language afterwards. Level 3 was set to include suiting the register to the company, which is a criterion of our scale and not a finding about Marain itself, and we have not examined how recruits acquire registers at higher levels.

5. Limitations

The register covers only recruits whose records Contact released, and Contact keeps some matters to itself. If the withheld records differ systematically, for example by the kind of work recruits were doing, our estimates describe the open part of the record and no more. All assessments were made by the Marain Studies Group on one Orbital, and two of us belong to it. Assessors knew each recruit's setting and could see the family code, so their judgements were not blind to the predictors we report.

Our outcome is the date of the first sitting at or above level 3, which depends on when the recruit chose to sit. Recruits who were confident may have sat early, and those who were not may have waited, so the times mix attainment with willingness to be assessed. The interval between reaching level 3 and sitting is unobserved, which makes every time an upper bound. Censoring assumes that a recruit who left Contact or the records had progressed no differently from one who stayed, an assumption we cannot test. If recruits who struggled left sooner, our medians are too short. For pre-arrival tuition the clock starts at home, before travel, when no sitting is possible because sittings took place on Chiark Orbital. That lengthens recorded times in this setting and biases the tuition contrast against it, so the ratios for the other two settings may be overstated.

Scale, family grouping and multilingual variable are all ours. The agreement figures are good for the scale, but a kappa of 0.86 still left 9 of 64 sittings with different verdicts on level 3. The 64 sittings were drawn at random, not weighted toward the threshold. First ratings were spread across the levels (15, 14, 10, 12 and 13 recruits at levels 1 to 5), and only 23 sittings had one assessor at or below level 3 and the other at or above it. All 9 disagreements fall among those 23, so the error in the verdict that decides the outcome is nearer 39% for sittings close to the threshold than the 14% across all 64. The family groups are coarse, and the multimodal group in particular joins very different languages. Cells in the family and setting cross-tabulation are small, and the 34 intakes limit how well clustering can be estimated.

Finally, Hethmar served aboard the ship that hosted 11 of the 64 shipboard recruits, and the first two authors sat on the assessing panels. Hethmar supplied the arrival and placement records but did not take part in rating or coding. The first two authors did the coding and the analysis.