The theoretical foundation is the interdependence hypothesis of Cummins (1979): literacy and academic concepts developed in one language are available to a second, provided the first has been developed far enough. The practical consequence is that time spent on the mother tongue is not taken from the second language. Longitudinal comparison of programme types in the United States found that children who received longer first-language instruction did better in English in the long run than those moved early into English only (Thomas and Collier, 2002). A review of African experience reached the same conclusion about early exit, judging that a switch of medium after two or three years is too soon for most children (Alidou et al., 2006).

Evidence from settings closer to rural Nepal is consistent, with qualifications. The Lubuagan programme in the Philippines reported gains across subjects for children taught through their own language (Walter and Dekker, 2011). A randomised trial in Kenya found that mother-tongue instruction improved reading in the mother tongue while showing no added benefit for English or Kiswahili, and it drew attention to the practical obstacles of implementing the approach in ordinary schools (Piper et al., 2016). Reviews written for policy makers stress that results depend on materials, teacher language proficiency and the length of the programme (Benson, 2004).

A donor-supported multilingual education programme was piloted in a small number of schools across several districts in the late 2000s. Hough et al. (2009) describe its approach, which built teaching materials from the knowledge of community elders and treated indigenous knowledge as curriculum content. Critical studies of what followed are instructive. Phyak (2011) argued that planning for primary education remained shaped by a monolingual habit of mind and by elite preference, and his later ethnography showed teachers and parents in a multilingual school treating the local language as an obstacle to mobility (Phyak, 2013). The spread of English-medium instruction in community schools, sought by parents as a form of capital and often taught by teachers with limited English, has sharpened the difficulty (Sah and Li, 2018).

Language policy research describes implementation as layered. Ricento and Hornberger (1996) likened it to an onion, with legislation at the outside and the classroom teacher at the centre, and Hornberger and Johnson (2007) showed that local actors open or close the spaces that national policy leaves. For Nepal this means that the constitutional right is realised or lost in decisions taken by school management committees and local governments, which since federal restructuring hold responsibility for basic education. A planning procedure should therefore be addressed to that level.

Planning begins with facts that schools seldom record. For each child entering the early grades the school notes the language used at home, any other language the child understands, and how well the child follows spoken Nepali. For each teacher it notes the languages spoken and read. Ethnic surnames are an unreliable guide, since many families have shifted language, so the information should come from parents and from listening to the children. The result shows whether the school has one dominant local language, two or three sizeable groups, or many small ones, and each pattern leads to a different model.

Parents who want English for their children are making a reasonable judgement about opportunity, and a plan imposed over their objection will not last. The school should present the options plainly at an open meeting, explain what research indicates about learning a second language on the foundation of the first, and ask what parents hope and fear. Where one language predominates, it can serve as the medium in the early grades. Where several are present, the workable choices include parallel groups for literacy, a multilingual classroom in which the teacher and classroom assistants move between languages, or the mother tongue as an oral medium alongside Nepali print. Flexible use of several languages in one lesson has a research base under the name of translanguaging (García and Li Wei, 2014). The agreement reached should be minuted.

Some of Nepal's languages have established writing traditions and published primers. Others are written rarely or in competing conventions. The school should not invent a spelling system. Its task is to find out what the language community and the curriculum authorities already use, adopt it, and record the choice. Initial materials can be modest: an alphabet chart, a set of short graded stories, songs and oral texts collected locally and written down by teachers and elders, in the manner described by Hough et al. (2009). Quantity matters less at first than having something for every child to read in the language of the lesson.

A teacher who speaks the children's language has not necessarily read or written it, and may have been trained to regard it as unsuitable for school. Preparation therefore has three parts: the teacher's own literacy in the language, methods for teaching initial reading in it, and methods for teaching Nepali and English as second languages instead of assuming them. Where no teacher speaks a language that many children use, the school and local government should consider recruiting a community member as a classroom assistant. Continuing support through visits and meetings with teachers from nearby schools is more useful than a single workshop.

The plan must show, grade by grade, which language carries which subject. The research reviewed above argues against an abrupt switch and against a very early one. A defensible sequence introduces Nepali orally from the first grade, adds Nepali literacy once reading in the mother tongue is under way, teaches English as a subject by methods suited to beginners, and keeps the mother tongue as a medium for as long as teachers and materials allow. Setting this out in a timetable answers the commonest parental objection, that children will be left without the national language.

Children taught mathematics in Tharu and tested on it in Nepali are being tested on Nepali. Classroom assessment in the early grades should use the language in which the subject was taught, and reading should be assessed in each language separately. The school needs to agree this with the local education office, since reporting formats usually assume a single language.

At the end of each year the head teacher, teachers and management committee should look at attendance, at how children are reading in each language, at what teachers found hard, and at what parents now think. The purpose is to adjust the model, the materials and the staffing before the next intake, and the review then feeds a lighter repetition of the first step.