Sage Journals: Discover world-class research

Abstract

This article introduces the novel mixed language input paradigm (MiLIP), which utilizes language mixing as a pedagogical tool for language learning, and describes the methodology to conduct a longitudinal study using this paradigm. We illustrate the paradigm using the example of intrasententially mixed Greek (target language) with English (familiar language), combined with multimodal input (audio, images, videos, subtitles) over three learning phases. Participants are presented with increasingly complex Greek input as the phases progress: nouns in Phase 1, noun phrases (NPs) in Phase 2, and NPs + verbs in Phase 3. Learning in the proposed paradigm is evaluated through picture selection and grammaticality forced-choice tasks at different stages, and an expressive vocabulary task at the end point of learning. This allows for an assessment of whether learners can arrive at some knowledge of words and grammar in the target language through a scaffolding model of ordered input. Researchers and educators can apply this methodology to investigate language learning utilizing individuals’ existing language knowledge.

Keywords

Language mixing language learning longitudinal design code-switching multilingual pedagogy mixed language input paradigm

I Introduction

Learning a second language (L2) or additional language is a common learning experience for the majority of the world’s population. Research on instructed language learning has primarily focused on the type and presentation of the input the learner receives and opportunities for interaction in the new language (Gass, 1997; Krashen, 1985; VanPatten & Williams, 2014). The grounding premise of these studies suggests that input in the developing language must be presented in a unilingual mode, i.e., the modality wherein one language is presented at a time. However, this largely neglects the fact that individuals, particularly from highly linguistically diverse societies often situated in the Global South, not only communicate by naturally mixing the languages they know, but may potentially also learn new languages through such mixed input (Tsimpli et al., 2020).

It has not empirically been established whether learning through such mixed language input can indeed occur or if it is potentially more effective than learning through unilingual input. To address this gap in knowledge, we are running a larger project which aims to experimentally investigate whether mixed language input can be successfully used as a pedagogical tool for language learning. In this article, we provide a description of our novel Mixed Language Input Paradigm (MiLIP) developed for the larger project, which can be used to evaluate short- and long-term effects of learning a new language from carefully structured, mixed language input in adult learners. MiLIP serves not only as a method for vocabulary learning, but, through structured intrasentential mixing, it provides a means of scaffolding such that morphological and syntactic acquisition of a language can develop.

We provide a detailed explanation of the steps taken in stimuli and assessment development, and in doing so, this paper can inform future experimental research wishing to utilize mixed language input as a methodological and/or pedagogical approach.

1 Mixed Language Use

The fluid transition from one language to another has propelled a large body of research in the field of code-switching (CS): the practice of switching between two or more languages, and translanguaging: the pedagogical language practices of bilingual learners and teachers. From the CS literature, languages shared between interlocutors can be mixed within the same sentence/clause (intrasentential CS) or across sentences/clauses (intersentential CS) (Deuchar et al., 2007; Muysken, 2000), with mixing patterns often being structurally and/or pragmatically motivated (Auer, 2013; Stell & Yakpo, 2015). Individual words or whole phrases can be switched with large variation in frequency and pattern (Deuchar, 2020). Insertion, alternation, and congruent lexicalization are types of CS which instantiate overt language mixing, whereby elements from each language are identifiable in the respective utterances (Muysken, 2000).¹

A common analysis of insertions, alternations, and congruent lexicalizations is with reference to Myers-Scotton’s (1993) matrix language frame (MLF), which proposes a distinction between a matrix and an embedded language within each code-switched unit. Put simply, the MLF proposes that the finite verb/auxiliary in a code-switched sentence determines the matrix language. The priority of finite verbal elements is based on inflectional morphological features (tense, agreement) responsible for morphosyntactic agreement between the subject and the verb as well as for the case feature of the subject. Although debate is ongoing about the linguistic restrictions that code-switched sentences abide by and the theories that are best suited to dictate these restrictions (Deuchar et al., 2007; MacSwan, 2014; Poplack, 1980, 2008), several studies show findings compatible with the MLF (Kniaź & Zawrotna, 2021; Parafita Couto & Gullberg, 2019; Vaughan-Evans et al., 2020). MiLIP uses the MLF’s distinction between the matrix and embedded language by introducing a gradual shift in the status of the new language from embedded to matrix and the concurrent demotion of the familiar language from matrix to embedded as learning progresses.

It is worth noting that CS has been considered by some scholars as part of the broader notion of translanguaging (or vice versa). The definition, theoretical underpinnings, and application of “translanguaging” has evolved over time (Balam, 2021; Poza, 2017; Wei, 2018), with its original application in educational contexts (C. Williams, 1994, 2000) and later as a practice of social justice (e.g., García, 2009; García & Lin, 2016). While a lively debate has ensued, there is no consensus about the definition and boundaries of translanguaging, with researchers often having divergent opinions about its scope and application (for a discussion, see Balam, 2021; Treffers-Daller, 2025). For the purpose of this paper and the broad application of MiLIP, we employ pedagogical translanguaging as originally proposed by C. Williams (1994, 2000) and extended by Cenoz and Gorter (2022). Translanguaging is thus viewed as a pedagogical approach which utilizes a learner’s full linguistic repertoire in pursuit of their language improvement and content competence in educational contexts. Here, the characterization of multilingual communication advocates the use of more than one language for better understanding of academic content and learning outcomes more broadly (Cummins, 2017, 2021). Furthermore, translanguaging also defends teacher’s and learner’s production of mixed sentences, not only for learning, but for the sociolinguistic and cultural identity of languages and individuals (Makalela, 2016; Wei, 2011). Unlike translanguaging or CS frameworks that provide descriptive and/or interpretive accounts of naturalistic language use, MiLIP is expressly designed to test structured learning through intrasententially presented mixed input.

2 (Mixed) Language use in additional language teaching and learning

Prescriptive teaching models relying heavily on unilingual instruction practices often fail to mirror the pedagogical realities of multilingual settings. The standard approach to child and adult L2 learning usually endorses a one-language pedagogy (Howatt, 1984; Howatt & Smith, 2014) that (1) relies on the L2 as the exclusive language used without (or limited) recourse to the first language (L1), (2) discourages translation, and (3) enforces a rigid separation between the L2 and L1 (see Cummins, 2007). While there has been a debate disputing the effectiveness of an axiomatic unilingual principle as the sole means to successful language learning (Auerbach, 1993; Cook, 2001; Rolstad et al., 2005; Turnbull, 2001), there is limited scientific investigation of the process and outcomes of L2 learning in contexts where input is not exclusively in the L2. Even in cases of promotion of bilingual education through two-way immersion programs and programs for minority languages (Alanís, 2000; Cummins, 2019), languages are still separated in line with the prevailing principle of one subject–one language (Lambert & Tucker, 1972), largely because of the misconception that mixing languages can lead to confusion or hamper learning (Lara-Alecio et al., 2004).

Against the backdrop of the one subject–one language principle, there is little room for linguistic flexibility, such as language mixing, as a toolkit for deeper learning. Presenting definitions and explanations in another language might be a common pedagogical approach, particularly at earlier stages of language learning, yet this does not reflect the type of naturalistic language mixing that occurs as CS. This is surprising given the lack of empirical support showing that a unilingual approach (i.e., input in one language) is superior for language learning compared with a multilingual one. There is, however, some evidence that language mixing can be facilitative to content teaching and learning (see MacSwan, 2022). For instance, Reyes (2004) showed that during a science project wherein CS was permissible, competent peer interaction involving language mixing was observed in a sample of second- and fifth-grade immigrant Spanish-speaking children in a US classroom. In a mixed methods study investigating the translanguaging practices of eight English teachers in India, Anderson (2022) found that the teachers were open to translingual practices wherein learners’ and teachers’ linguistic repertoires were not restricted to English only. Such inclusive practices facilitated student learning and more broadly mirror the linguistic reality in India (also see A. Lightfoot et al., 2021). It is also evident that all languages can be drawn from in linguistically diverse social contexts (Grosjean, 2010; Wigdorowitz et al., 2022, 2023) and educational settings (Cook, 1995, 2001, 2016; Creese & Blackledge, 2010; Lin, 2013).

What we do know about language learning is that people can learn multiple languages (to varying degrees), without explicit instruction (Morgan-Short et al., 2012; J.N. Williams, 2020), whether the language is natural or artificial (Ettlinger et al., 2015), and whether they have only one or more languages in their repertoire (Bardel & Sánchez, 2020). Furthermore, it is well known that an individuals’ languages are simultaneously activated (to varying degrees) and cannot simply be switched off at will, even in explicitly unilingual environments (Sanoudaki & Thierry, 2015), though they can be pro-actively inhibited (Wu & Thierry, 2017) or integrated, via CS, to meet communicative demands and norms (Green & Abutalebi, 2013). Despite a rich and broad literature on language learning, the utility of mixed input as a pedagogical method for language learning is underexplored. Filling this gap, theoretically and empirically, will allow us to assess the linguistic costs and benefits of language mixing for language learning, and to reexamine the effectiveness of unilingual education for language learners.

Importantly, Anderson (2022) argues that strict unilingual educational policies in contexts where students and teachers share multiple languages are “outdated and counterproductive” (p. 2248). He further recommends that learner inclusion and boosting confidence and participation should be prioritized over maximal target language use. MiLIP takes these recommendations on board and attempts to address them from an experimental perspective; in this way, the present study is methodologically novel and conceptually original as it provides the first concrete way to implement mixing in a language learning context.

3 Language learning considerations

Communication and language acquisition/learning do not occur in a vacuum. Rather, multimodal (contextual) information is continuously integrated (Lazaridou et al., 2017; Spivey & Dale, 2006). In a classroom, for example, a teacher can say a word, emphasize its pronunciation, write the word down, and show an image of the object that the word refers to, all to support the learning of that specific word and its grammatical features. Students are, in turn, evaluated via written and/or oral examination, usually in a single modality at a time in such a way that they abstract away from learning from multiple modalities. The same can be done for remote, web-based learning, where images, videos, sentences, and even stories can be used to relay linguistic information. According to Mayer’s (2009) cognitive theory of multimedia learning, when humans process language across dual channels (auditory and visual), they are cognitively limited by how much information can be processed in each channel simultaneously (Sweller et al., 2011) and need to process information actively in order to adequately attend to and integrate relevant information to achieve learning. Multimodal input enhances learning when visual and verbal input are aligned and nonredundant, but overload occurs if too many competing sources are presented. It has been shown that well-designed multimodal instruction can indeed enhance retention and comprehension (Mayer, 2005). MiLIP takes this multimodality of natural interaction and learning into account in its design. The paradigm presents all input in both oral and written form, accompanied with graphics, to support the morphological and phonological properties of the language to be learned through both the auditory and the written modalities.

An additional consideration in the presented methods, inspired by classroom teaching and learning practices, is the structure of the input. Structuring language input by gradually presenting more complex grammatical structures and inviting the learner to “notice” new linguistic forms has been found to facilitate learning (Robinson, 1995; VanPatten, 1996). This ties in with the idea that language learning is a gradual and ordered process (Tsimpli, 2014). Specifically, nouns precede verbs (Gentner & Boroditsky, 2001), simple sentences precede complex ones (D. Lightfoot, 1989), and active structures are learned before passive ones (Pinker et al., 1987). This orderly development of language is accomplished naturally by a child exposed to L1 input. For the L2 learner, however, it is often helpful to structure the input so that the gradual process of learning is facilitated (VanPatten, 1996). Drawing on this body of literature, we apply such a gradual, ordered input process in MiLIP.

II Methodology

In this section, we present the methodology of a remote, web-based longitudinal language learning paradigm. In our examples, the target language to be learned is Greek, which is embedded into a familiar language, in this case, English. However, the reader can take our proposed method development guidelines and, in theory, apply them to any language combination: Greek and English are illustrative examples used in the design of MiLIP. Ethical clearance for this project was granted from the Faculty of Modern and Medieval Languages’ Research Ethics Committee at the University of Cambridge. All materials are available at https://doi.org/10.17605/OSF.IO/P7JSW.

1 Procedure

MiLIP includes eight different sessions, which are to be carried out online over a four–six-week period. Prior to the start of the main study, a declaration of interest questionnaire can be used to gauge participants’ interest, assess their eligibility, objectively evaluate their English (or relevant language) proficiency, and screen for potential bots and fraudulent cases (for a discussion of the latter, see Vogelzang et al., in preparation). The full procedure is shown in Figure 1.

Figure 1.

Procedural overview of the study. There are eight sessions in total (plus a declaration of interest questionnaire), which include a pre-study session, a post-study session, and three learning phases: (1) nouns; (2) noun phrases (NPs); and (3) NPs + verbs. Details of the stimuli in the different learning phases are provided in Section II.3.

Once eligibility has been established, the participant can be invited to the main study. In a pre-study (pre-language learning) session, they are asked to complete a (language) background questionnaire, and subsequently to complete cognitive or other tasks, as needed. The next calendar day, participants were invited to the first language learning session of Phase 1. Each phase consists of 2 learning sessions. All invites to the remaining sessions (Phase 1, Session 2; Phase 2, Session 1; Phase 2, Session 2; Phase 3, Session 1; Phase 3, Session 2) can be sent 4 days after completion of the previous session. We opted for a 4–8-day window between each language learning session to ensure the design is longitudinal and that learning takes place over an extended period, while also being cognizant of the length of participation to minimize attrition. A post-study session follows the final learning session (Phase 3, Session 2) immediately, but could be implemented with a delay to reduce the length of participation in one go and/or gauge retention.

Abilities in the target language are evaluated at different points during the study. Following the presentation of input in Phase 1 (Sessions 1 and 2) and before the start of Phase 2 (Sessions 1 and 2), participants are asked to identify the correct meaning of each newly learned Greek noun from a multiple-choice noun-based picture selection task. This allows the researcher to gauge participants’ learning (Phase 1) and retention (Phase 2) rates. At the end of Phase 2 and Phase 3 sessions, participants additionally perform a grammaticality forced-choice task (GFCT) which assesses participants’ ability to correctly identify grammatical part-learned and part-novel structures in Greek. Finally, an oral vocabulary production task and post-study questionnaire are administered at the end of Phase 3, Session 2. Participants are debriefed with a downloadable handout of examples illustrating what they have learned. The language learning sessions and subsequent tasks take approximately 30 minutes per session to complete, with Phase 3, Session 2 being a little longer due to the additional production task and post-study questionnaire. The study needs to be carried out on a laptop or desktop computer with a stable internet connection, working speakers and microphone. Participants can take notes during the language learning sessions but may not refer to them when performing the assessments. Each task is described in detail below.

All materials have been piloted on L1 English (n = 3), L1 Mandarin–L2 English (n = 2), and L1 Greek–L2 English (n = 1) speakers. The tasks were amended as needed to improve clarity, functionality, and flow, and are found to work as intended. Pilot participants were compensated for their time.

2 Participant and eligibility checks

We urge researchers to consider their participant population in-depth before data collection, as it is known that social and individual variables affect language learning. It is also important to consider how representative a sample is compared with real-world language learners.

a Language proficiency

In our example, participants will be non-native speakers of the familiar language (English) into which the language to be learned (Greek) is mixed. If participants are non-native speakers of the familiar language, proficiency should be assessed. In our example, we therefore include an English proficiency test.² In addition, the eligibility questionnaire would ask about previous knowledge of the language to be learned. This is done to ensure participants’ knowledge of and exposure to this language is minimal.

b Other eligibility criteria

We further recommend getting a comprehensive overview of the participants’ demographic and language background, as several background factors could affect their learning. Individuals with substance, language, developmental, neurological, and/or psychiatric disorders should be considered for exclusion, as all of these may affect their learning and/or (neural) processing of language. Finally, the methodology leads to the exclusion of individuals with hearing impairment, color blindness, and/or visual impairments that are not corrected with glasses/contact lenses. This is because the content is presented visually and orally.

3 Learning materials

The mixed language sentences are presented within our novel paradigm, MiLIP. In MiLIP, all input is presented in both oral and written form to support learning of the phonological properties of Greek through oral and written modalities. The language learning task is designed to manipulate the input condition that participants are taught Greek in: mixed input (sentences presented in English with inserted Greek words) or unilingual input (sentences presented in Greek and English separately). The design of the language input is based on the principles of exposure to words, phrases, and sentences in the new language on a gradual basis (Tsimpli, 2014).

The learning of Greek proceeds in three phases of development, as described in the procedure (Figure 1). Materials in each consecutive phase differ in the quantity and quality of mixing per sentence, with the quality following the order of noun (N) → N phrase (NP) → NP + verb (V). In Phase 1, the matrix language (Myers-Scotton, 1993) of the stimuli is English, and concrete Greek nouns of the three different genders (masculine, feminine, neuter) are inserted as the embedded language. In Phase 2, Greek definite and indefinite determiners are inserted together with the nouns. As the Greek nouns and determiners carry markings for gender, case (nominative or accusative), and number (singular or plural), this information has to be extracted and generalized by the learner without any explicit teaching. Verbs, specifically present and past tense action verbs, are introduced in Phase 3. These phases allow the researcher to examine whether learners can arrive at some knowledge of Greek words and grammar through a scaffolding model of ordered input in the new language. In the unilingual input condition, sentences are presented unilingually in Greek through oral and written modalities (with an English translation below), with the target words (N → NP → NP + V) highlighted in bold for each respective phase. Similarly, in the mixed input condition, sentences are presented with Greek words inserted into English sentences, also through oral and written modalities (with an English translation below), with the target words highlighted in bold for each respective phase (see Figure 2).

Figure 2.

Example of a sentence in the mixed (left) and unilingual (right) conditions implemented in Gorilla (an experiment builder platform; see https://gorilla.sc/).

In each learning session, learners are presented with the same 12 short stories, each containing six sentences (shown one at a time). The input is presented as a video clip of an L1 Greek–L2 English speaker talking, with a simultaneous written transcription at the bottom of the screen in both conditions (unilingual and mixed). Written Greek is based on a broad transliteration in Roman script supporting phonological rather than graphological properties. Finally, images accompanying the sentences, displaying characters, objects, and scenes, are presented to provide extralinguistic cues to aid learning (generated using Adobe FireFly,³ see Figure 2). In the mixed condition, the English syntactic frame will allow learners to bootstrap onto their existing knowledge of English to learn the Greek nouns and verbs. This knowledge can support acquisition through several cognitive mechanisms. Positive cross-linguistic transfer of English word order can facilitate the parsing of Greek structures (Schwartz & Sprouse, 1996), whereas determiners can provide early cues for syntactic bootstrapping of gender, number, and case (Foucart & Frenck-Mestre, 2012; Lidz et al., 2003; Wonnacott et al., 2008). In addition, cross-linguistic structural priming can strengthen shared syntactic representations across English and Greek (Schoonbaert et al., 2007). In the unilingual condition, learners largely rely on translations to access meaning.

Twelve nouns are learned in Phase 1 and Phase 2. Multiple occurrences of the same word are crucial for naturalistic and successful learning, and as such each noun is presented to participants six times in different sentential contexts. In Phase 3, we use the same nouns and sentences, now also “revealing” the verbs. Participants are not expected to learn the meaning of each verb (as they are with nouns), but to implicitly infer some of their properties (e.g., number and tense marked with certain morphemes) and agreement with the nouns in subject position.

In each session, sentences are presented in the context of 12 stories, each containing six sentences shown one at a time, as mentioned previously (Table 1 presents an example story). Each story includes two critical nouns (nouns to be learned; Table 2 provides an overview of the nouns in each story). The 12 Greek nouns to be learned were selected based on gender, animacy, and syllable count. The Greek language has three grammatical genders (masculine, feminine, and neuter), and as such, all grammatical genders are presented in the design. In terms of distribution of the grammatical genders across nouns, the neuter is considered the most frequent, followed by the feminine and finally the masculine gender (Mackridge, 1987); however, all three genders are used frequently. We considered it important to expose learners to all three genders in the same frequency in this iteration of the design. This is an aspect that can be manipulated in future research. In MiLIP, there were four of each gender: one animate (two syllables), one animate (three syllables), one inanimate (two syllables), and one inanimate (three syllables). The nouns were selected based on their suffixes, aligning with native Greek speakers’ preference for gender assignment (masculine: -as, -is; feminine: -a, -i; neuter: -o, -i) (Mastropavlou & Tsimpli, 2011). Noun frequency, as indicated by the Hellenic National Corpus,⁴ was also taken into account, trying to match nouns of different gender and the same syllable length (e.g., masculine/feminine/neuter two-syllable nouns had comparable frequencies). We tried to avoid Greek–English cognates as much as possible.

Table 1.

Example of a story (story 12) across each phase for mixed and unilingual conditions.

Sentence	Phase	Mixed condition	Unilingual condition	English translation
1	1	The museum has a pínaka by Picasso.	To musío éhi énan pínaka tu Picasso.	The museum has a painting by Picasso.
	2	The museum has énan pínaka by Picasso.	To musío éhi énan pínaka tu Picasso.	The museum has a painting by Picasso.
	3	The museum éhi énan pínaka by Picasso.	To musío éhi énan pínaka tu Picasso.	The museum has a painting by Picasso.
2	1	The pínakas costs a fortune.	O pínakas kostízi mía periusía.	The painting costs a fortune.
	2	O pínakas costs a fortune.	O pínakas kostízi mía periusía.	The painting costs a fortune.
	3	O pínakas kostízi a fortune.	O pínakas kostízi mía periusía.	The painting costs a fortune.
3	1	A fotjá broke out in the museum’s basement.	Mía fotjá kséspase sto ipóyio tu musíu.	A fire broke out in the museum’s basement.
	2	Mía fotjá broke out in the museum’s basement.	Mía fotjá kséspase sto ipóyio tu musíu.	A fire broke out in the museum’s basement.
	3	Mía fotjá kséspase in the museum’s basement.	Mía fotjá kséspase sto ipóyio tu musíu.	A fire broke out in the museum’s basement.
4	1	The guard saw the fotjá and called the fire department.	O fílakas íde ti fotjá ke kálese tin pirosvestikí.	The guard saw the fire and called the fire department.
	2	The guard saw ti fotjá and called the fire department.	O fílakas íde ti fotjá ke kálese tin pirosvestikí.	The guard saw the fire and called the fire department.
	3	The guard íde ti fotjá and called the fire department.	O fílakas íde ti fotjá ke kálese tin pirosvestikí.	The guard saw the fire and called the fire department.
5	1	The firefighters extinguished the fotjá.	I pyrosvéstes ésvisan ti fotjá.	The firefighters extinguished the fire.
	2	The firefighters extinguished ti fotjá.	I pyrosvéstes ésvisan ti fotjá.	The firefighters extinguished the fire.
	3	The firefighters ésvisan ti fotjá.	I pyrosvéstes ésvisan ti fotjá.	The firefighters extinguished the fire.
6	1	The pínakas went to another museum.	O pínakas píje se álo musío.	The painting went to another museum.
	2	O pínakas went to another museum.	O pínakas píje se álo musío.	The painting went to another museum.
	3	O pínakas píje to another museum.	O pínakas píje se álo musío.	The painting went to another museum.

Note. Phase 1 = nouns; Phase 2 = determiners + nouns; Phase 3 = determiners + nouns + verbs. Translation sentences always appear below the mixed/unilingual sentences in the input.

Table 2.

Nouns shown per story with information about their animacy, syllable length, and plurality.

Story	Noun 1(English / Greek)	Animacy	Number of syllables	Plurality (introduced in Phase 2)	Noun 2(English/Greek)	Animacy	Number of syllables	Plurality (introduced in Phase 2)
1	school / sxolío	Inanimate	3	Yes	student / mathití	Animate	3	No
2	school / sxolío	Inanimate	3	No	teacher / daskála	Animate	3	No
3	teacher / daskála	Animate	3	Yes	student / mathití	Animate	3	No
4	cousin / ksadélfi	Animate	3	No	painting / pínaka	Inanimate	3	Yes
5	cousin / ksadélfi	Animate	3	No	key / klidí	Inanimate	2	No
6	baby / moró	Animate	2	No	wound / plijí	Inanimate	2	Yes
7	man / ándras	Animate	2	No	baby / moró	Animate	2	No
8	man / ándras	Animate	2	No	wound / plijí	Inanimate	2	No
9	boy / agóri	Animate	3	No	key / klidí	Inanimate	2	Yes
10	boy / agóri	Animate	3	No	fence / fráhtis	Inanimate	2	Yes
11	fence / fráhtis	Inanimate	2	No	fire / fotjá	Inanimate	2	No
12	painting / pínaka	Inanimate	3	No	fire / fotjá	Inanimate	2	No

Each noun appears six times, spread across two stories. The first time the noun appears in each story, it is in indefinite form and the other two occurrences are in definite form. The nouns are evenly distributed in nominative and accusative case⁵ (three times each). The stories were constructed around the nouns. The verbs used are two-to-five syllables in length, in present or past form, and are all in active voice.

Each session (12 stories) repeats twice, leading to six learning sessions across three learning phases. In Phase 1 (Sessions 1 and 2), all nouns are in singular form. From Phase 2, half of the nouns are also introduced in their plural form (stories 1, 3, 4, 6, 9, 10;⁶ see Table 2). We did not do this for all nouns, so we could check participants’ ability to generalize and apply pluralization to forms they have not encountered before. All the information on nouns, determiners, and verbs that participants received is summarized in Tables 3 –5, respectively.

Table 3.

Examples of Greek case and number markings learned in Phase 1.

	Masculine	Feminine	Neuter
Singular
Nominative	mathitís / ándras	plijí / daskála	klidí / sxolío
Accusative	mathití / ándra	plijí / daskála	klidí / sxolío
Plural
Nominative	mathités / ándres	plijés / daskáles	klidjá / sxolía
Accusative	mathités / ándres	plijés / daskáles	klidjá / sxolía

Table 4.

Examples of Greek articles learned in Phase 2.

	Masculine		Feminine		Neuter
	Singular
	Def	Indef	Def	Indef	Def	Indef
Nominative	o	énas	i	mía	to	éna
Accusative	ton	énan	tin	mía	to	éna
	Plural
	Def	Indef	Def	Indef	Def	Indef
Nominative	i	∅	i	∅	ta	∅
Accusative	tus	∅	tis	∅	ta	∅

Note. ∅ = no article.

Table 5.

Examples of Greek tense and number markings on verbs learned in Phase 3.

Present		Past
3SG (end in -i)	3PL (end in -un)	3SG (end in -e)	3PL (end in -an)
méni	ménun	árhise	árhisan
pézi	pézun	évale	évalan
kostízi	kostízun	krátise	krátisan

The language input task is designed with block randomization, such that each participant will be randomly assigned to one of two equally sized, predetermined input conditions (unilingual versus mixed). Randomization is also used for the order of the 12 stories in each of the six learning sessions (the sentence order within each story remains the same, however). After each story, a multiple-choice filler comprehension question is posed, asking participants about the content of the story. Responses to comprehension questions are not included in the analysis, but are used to hold participants’ attention and ensure that they are sufficiently engaging with the task.

4 Evaluation measures

a Find the picture: picture selection task

A picture selection task is used to assess participants’ knowledge of the 12 Greek nouns presented in the language-learning part of the experiment. Picture selection tasks are widely used to assess L1 and L2 vocabulary knowledge (Huettig et al., 2011). In this task, participants are asked to select an image of a specific noun with the prompt “Which one is the [Greek noun]”⁷ from four image options presented in a quadrant (Figure 3). Each item includes images of the target noun, a semantically related distractor, an unrelated object, and one of the other target nouns that does not correspond to the correct noun (Table 6 presents an overview). In addition, there are three filler items (in English) used as attention checks (Table 7). All but eight images were obtained from the open-source Multilingual Picture Database (Duñabeitia, 2022; Duñabeitia et al., 2022), and images corresponding to the nouns that were not available on the database were sourced from Clipart Library⁸ (cousin, student, teacher, girl, grandfather, doctor, sling, school). The order of the 12 target nouns is randomized and the four image options also appear in a random configuration with each trial. The filler items are not randomized and appear after every four test items (in 5th, 10th, and 15th position). Each item is displayed on the screen for up to 6,000 ms.

Figure 3.

Example target item (left) and example filler item (right) from the picture selection task.

Table 6.

Overview of pictures chosen for target nouns in the picture selection task.

Target noun (Greek)	Target noun translation (English)	Semantically related distractor	Unrelated object	Other target noun (English)
ándras	man	woman	pacifier	school
fráhtis	fence	wall	saxophone	student
pínakas	painting	map	violin	baby
mathitís	student	grandfather	carrot	key
fotjá	fire	rain	guitar	cousin
pligí	wound	sling	giraffe	teacher
daskála	teacher	female doctor	fire extinguisher	fence
ksadélfi	cousin	sister	calculator	painting
klidí	key	lock	flower	man
moró	baby	girl	door	fire
agóri	boy	male doctor	ball	wound
sxolío	school	farm	burger	boy

Table 7.

Overview of pictures chosen for filler nouns in the picture selection task (always presented in 5th, 10th, and 15th position, respectively).

Filler noun (English)	Unrelated object	Unrelated object	Unrelated object
flower	carrot	saxophone	ball
violin	burger	fire extinguisher	giraffe
door	guitar	calculator	pacifier

As mentioned, this task is completed after Phase 1 (at the end of both Session 1 and Session 2) and before the start of Phase 2 (at the beginning of both Session 1 and Session 2) to evaluate Greek learning (Phase 1) and retention (Phase 2). To ensure participants have sufficient exposure to Greek and are indeed learning, they must obtain a score of at least 83.3% (10/12) in Phase 1 to continue to the next session. They are informed of this requirement in the instructions before the task and are given feedback about their performance at the end of the task. If a participant scores below the performance requirement, incorrectly identified words are repeated with their respective stories/materials and are retested up to a maximum of three times. There is no minimum performance requirement for Phase 2, as at this point the task tests retention of nouns learned in the previous phase and takes place before the language learning task. Measures of accuracy on target items (out of 12) and reaction times on accurate items can be calculated per session. Filler items are not included in the calculation of these scores.

b Pick the phrase: a GFCT

The participants were assessed on specific grammatical aspects of Greek through the use of a GFCT four times during the learning process. GFCTs are common assessments of grammatical features in language acquisition (Schütze & Sprouse, 2014). A GFCT is administered as the final task across both sessions of Phase 2 and Phase 3, where participants have been exposed to NPs and Vs in Greek, respectively. Importantly, this task was designed to test participants’ mastering of Greek grammatical features. Their knowledge of the meanings of these words was not assessed. In this task, participants are shown two or three options in Greek (two for case, two for number, three for gender, as appropriate) and need to select the one they think is grammatically correct. When testing knowledge of NPs (agreement with determiner in gender/case), only the Greek text (in Roman script) is used. When testing knowledge of Vs, the English translation is also included. If participants are unsure about their choice, they are encouraged to make a guess. Participants will be exposed to some of the words or word combinations (e.g., in the case of determiner and noun agreement), but will also be tested on novel words and word combinations, although they will have had exposure to all tested grammatical features. By using some novel items, we are able to check participants’ ability to generalize to unseen forms from the input.

The number of items per GFCT varies depending on the features being assessed. Specifically, the GFCT comprises 48 items in Phase 2 Session 1, 48 items in Phase 2 Session 2, 20 items in Phase 3 Session 1, and 44 items in Phase 3 Session 2 (Table 8 provides an overview). Items are randomized per session and not repeated. RTs and accuracy scores as percentages (number of correct responses out of the total number of trials per session or the total number of trials per grammatical feature) can be calculated.

Table 8.

Overview of the features tested in each GFCT, with accompanying examples. Note that Phase 2 Sessions 1 and 2 test the same grammatical features with different items (i.e., no item repetition).

	Grammatical feature tested	Learned or novel	Number of items	Example item
Phase 2 Sessions 1 and 2	Gender	Learned	24	ton fráhti tin fráhti to fráhti
	Case	Learned	16	énas fráhtis *énas fráhti
	Case	Novel N	16	énas jípas *énas jípa
	Number	Learned	20	i mathités *o mathités
	Number	Novel N	20	i hárakes *o hárakes
Phase 3 Session 1	SV agreement Number	Learned	12	A student arrived. . . Énas mathitís éftase. . . *Énas mathitís éftasan. . .
Phase 3 Session 1	SV agreement Case	Learned	8	A student arrived. . . Énas mathitís éftase. . . *Énan mathití éftase. . .
Phase 3 Session 2	SV agreement Number	Learned N(S), V but in novel sentence	12	The man is greeting his friends. O ándras heretái tus fílus tu. *I ándres heretái tus fílus tu.
	VO agreement Case	Learned V, N(O) but in novel sentence	14	The judge punished the man. O dikastís timórise ton ándra. *O dikastís timórise o ándras.
	SV agreement Number	Learned N(S,O) but novel V	18	A man cleaned the wound. Énas ándras kathárise tin plijí. *Énas ándras kathárisan tin plijí.

i GFCTs in Phase 2

In Phase 2, participants’ knowledge of agreement between determiner and noun (in gender, case, and number) is tested. For gender, items include only nouns that participants’ have been exposed to.⁹ Each of the 12 nouns is tested twice, leading to a total of 24 trials presented with a three-alternative forced choice (3-AFC) design. For case, participants are evaluated on four previously seen masculine nouns and four novel masculine nouns.¹⁰ Each noun appears in four occurrences (nominative-definite, accusative-definite, nominative-indefinite, accusative-indefinite), leading to a total of 32 trials with a 2-AFC design. For number, participants are tested on 12 already seen nouns and 12 novel nouns (which adhered to the same pluralization rules). Each masculine noun is tested in two trials: once manipulating the determiner (definite determiner in plural versus definite determiner in singular + noun in plural) and once manipulating the noun (indefinite determiner singular + noun in singular vs. noun in plural). The same logic is applied to neuter nouns. Because of the simplified spelling (Roman script), this manipulation for the feminine nouns only works with indefinite determiners in nominative case (e.g., mía daskála versus mía daskáles). Hence, for feminine nouns, there is only one trial per noun. The total trials (40 items) for nouns have a 2-AFC design and include: eight trials for seen masculine nouns, eight trials for seen neuter nouns, four trials for seen feminine nouns, eight trials for novel masculine nouns, eight trials for novel neuter nouns, and four trials for novel feminine nouns. Items are randomized and split evenly across the two sessions.

ii GFCTs in Phase 3

In Phase 3, NP + V knowledge is tested. In Session 1, knowledge of SV agreement in number and case is tested. This involves rote learning since participants have been exposed to all the structures they are asked to judge, so do not require generalization or rule application.¹¹ For SV agreement in number, participants are tested across 12 trials (one for each noun they are exposed to) following a 2-AFC design. For SV agreement in case (the subject always needs to be in the nominative case), participants are tested across eight trials (four masculine, four feminine), also following a 2-AFC design. This manipulation does not work with neuter nouns, because they are identical in surface form in nominative and accusative case. In Session 2, participants’ ability to generalize their V knowledge to new combinations and verbs is tested. SV agreement in number is tested, but this time the combinations between nouns and verbs are novel, i.e., participants have been exposed to the nouns and the verbs but not in the combinations they are asked to evaluate them in. For example, participants have seen the sentence: “To korítsi heretái to agóri” and “Énas ándras éhi,” so they should be able to infer that the singular masculine noun “ándras” should be the subject of the singular verb “heretái” in the following example trial:

“O ándras heretái tus fílus tu” vs. *“I ándres heretái tus fílus tu”

“The_sg. man_sg. greets_sg. his friends” vs. *’The_pl. men_pl. greets_sg. his friends”

There are 12 trials in total, one for each noun, following a 2-AFC design. In half of the trials, the target combination involves a singular subject (and verb), and in the other half, a plural subject (and verb). We also test VO constructions (nouns in object position need to be in accusative case in Greek), again using target nouns and verbs that participants have been exposed to but in novel combinations. In singular form, items include: two masculine definite, two masculine indefinite, and two feminine definite nouns (already seen in the input). In plural form, items include: two masculine definite and two feminine definite nouns (participants have seen the plural form in the input), and two masculine definite and two feminine definite (participants have not seen the plural form in the input; see Table 2).¹² This results in 14 trials in total, following a 2-AFC design. Finally, SV agreement in number with novel verbs is tested following the morphological properties of seen verbs. These include six trials with singular S and V and 12 trials¹³ with plural S and V, leading to a total of 18 trials with a 2-AFC design.

c Expressive vocabulary task

We designed an expressive vocabulary task to gauge participants’ active knowledge of the 12 Greek nouns from the learning tasks. We are interested in the accuracy of productions, not in terms of prosody but in terms of vocabulary. Participants are presented with an image (from those used in the picture selection task) and are asked to name what it refers to (the noun). Accuracy of the responses can be calculated, for example, with a Levenshtein or Damerau–Levenshtein distance between the produced form and the correct form (e.g., with the “stringdist” R package; Van der Loo, 2014). The Damerau–Levenshtein distance reflects the number of single character (in this case, phoneme) edits (insertions, deletions, substitutions, or transpositions) required to change one word into the other. This should then be transformed into a percentage correct score, reflecting how much of a word was produced correctly. For example, if the target response was mathitís and the participant’s response was mathatí, then one substitution (“a” instead of “i”) and one deletion (of the final “s”) leads to a distance of 2. Because there are seven phonemes in the target word, the Damerau–Levenshtein distance score would be 28.6%. Researchers could also analyze prosodic accuracy or other phonological features of interest.

d Post-study questionnaire

A short 5–10-minute questionnaire invites participants to reflect on their engagement with the study and their language learning behaviors. Specifically, it asks about participants’ learning/response strategies (guessing, intuition, memory, or rule application), confidence in task performance (not very confident, fairly confident, certain, or no observations made), and explicit grammatical rule identification (open-ended). These questions are asked for the picture selection task and GFCT. We also ask whether participants engaged in other methods of Greek learning over the duration of the study, such as through an app, reviewing any notes made, and via media channels. In addition, we ask how helpful each modality (image, audio, video, subtitle) is in facilitating learning. Answers for each modality are on a scale ranging from 0 to 100, where 0 represents “not helpful at all” and 100 represents “extremely helpful”. In addition, we ask about participants’ level of motivation both during the language learning sessions and during the assessment tasks. Answers also range from 0 to 100, where 0 represents “not motivated at all” and 100 represent “extremely motivated”. We can also get an indication of whether participants took and/or used any notes during the learning sessions, whether they felt as though they had learned any Greek, and what this learning experience compares to in terms of how much they learned in this study versus using existing learning apps (if applicable). Finally, participants can share information about their reason for participating, any technical difficulties experienced, and anything else they deem useful to add.

Responses to these questions can be used in several ways, including for clarity, exclusion (e.g., if a participant uses their notes during the assessments), or in-depth quantitative and/or qualitative analysis. In terms of quantitative analysis, we recommend calculating composite scores for (1) helpfulness ratings of each modality (images, audio, videos, subtitles) using Shannon’s entropy (e.g., with the “languageEntropy” R package; Gullifer & Titone, 2018) to get an indication of the utility of each modality on learning; and (2) motivation (story tasks, assessment tasks) by averaging the scores (0–100) across the items. Qualitatively, responses to open-ended questions can be explored.

III Stepwise methods guide

Using Figure 1 as a reference, we present a stepwise methods guide in Table 9, which researchers can use to adapt the methodology for their own research. In the table, we indicate which steps are essential (declaration of interest, phases of language learning tasks, assessments) and which optional (pre-study and post-study sessions), as well as which adaptations we recommend other researchers to consider.

Table 9.

Stepwise methods guide with recommendations to adapt the language learning methodology.

Step	Necessity	Purpose	Adaptations/recommendations
Declaration of interest	Essential	Filtering out participants before inviting them to the main study which saves resources such as time and money Check eligibility/inclusion criteria	Language proficiency test can be removed or adapted (e.g., test a different language) to fit participant population/research aims Type of information asked can be changed
Pre-study session	Optional	Get more background information that may affect language learning which is not included in the declaration of interest (e.g., socioeconomic status and language attitudes) Get any baseline measures (e.g., in case of an intervention study) Measure covariates (e.g., cognitive measures)	Type of background information asked can be changed Baseline measure of ability in the language to be learned can be added Covariate measures can be removed or added
Phases of the language learning tasks	Essential	To facilitate the learning of a novel language over time, with gradual and systematically ordered input	Design can be applied to different language combinations depending on the target and known languages of the participant population We strongly recommend keeping the input ordered (N → NP → NP + V) but researchers can include other grammatical constructions of interest within different learning phases (e.g., adjectives, negation, passive sentences, and unilingual sentences) The number of nouns, verbs, and stories can be adjusted so that researchers can shorten or extend the content and input quantity as desired The length of time between sessions can be adjusted to test different research questions (e.g., to account for sleep and memory retention) The conditions under which participants take the experiment can be adapted (e.g., supervised remote testing or in-person) The stories and images that accompany them should be adapted to each participant population to ensure that they contain relatable and culturally familiar/appropriate content
Assessments	Essential	To evaluate the learning of meaning and grammatical features of the novel language	Assessments other than picture selection and GFCTs can be developed to target different aspects of learning such as semantics/comprehension in addition to structure Production can be evaluated earlier on, and accent features can be analyzed in addition to vocabulary knowledge Assessments can be adapted to include multimodal components (e.g., visual and auditory) instead of just being presented in writing to ensure modal consistency across learning and assessment Retention of meaning and/or grammatical features could be assessed across more (or all) sessions following the first presentation of input, before the start of the next session Other assessments can be developed that target specific language learning processes (e.g., lexical decision task)
Post-study session	Optional	Get subjective information about participants’ learning experience after completing the study (e.g., reflections about learned patterns, the usefulness of input modalities, and motivation for participating)	Type of information asked can change depending on the goals of the research Post-tests can be added if researchers want to retest participants on any measure (e.g., cognitive tasks) or test retention of the language materials

IV Limitations

MiLIP, and specifically the presented Greek learning example, has certain limitations. First, the sentences within each story were developed with the nouns (and not the verbs) as the guiding structure. Given the complexities involved in creating 12 completely novel stories that meet several criteria, including that each story contains six sentences comprising two target nouns across three sentences, each sentence is between 4 and 11 words in length, the stories flow in a narrative fashion, and each sentence can have a graphical representation, we chose to develop the sentences around nouns in the first instance. The verbs included in the sentences were used to evaluate various grammatical features of Greek, but never their meaning. In future iterations of sentence stimuli, it could be worth having prespecified criteria for the development of verbs in addition to nouns so that they are more controlled and balanced in number.

A second limitation relates to the distinction between the multimodal presentation of input in the learning phases and the unimodal style of evaluation in the picture selection task and GFCTs (i.e., visual, without audio). Future assessments can be designed to incorporate more multimodal features in the assessments, similar to the presentation of language input received. For example, in the picture selection task, audio could accompany the written text, and in the GFCT, a video of a speaker saying each of the options could be played along with the written text. However, in real-world settings, assessments are primarily presented in a single modality, usually in written form, to evaluate competence (Jamieson, 2011; Yu & Xu, 2024). Our assessments therefore reflect that of real-world language evaluation. In addition, other lexical and grammatical features can be evaluated, including, for instance, the retention of verbs as well as nouns. In MiLIP, limited feedback is provided as there is an emphasis on implicit learning (rather than explicit feedback). In this way, the learning process is similar to naturalistic acquisition rather than classroom learning, and relies less on memory. This was a deliberate choice as it allows researchers to examine whether participants can make generalizations with previously unseen forms, for example with respect to case and number (e.g., in Phase 2). Future work could explore the role of feedback, however, it is important to emphasize that simulating a traditional language classroom environment was neither the aim of MiLIP, nor is it any longer the norm in language learning. Although we acknowledge that our method differs quite drastically from language learning experiences in typical classrooms, it arguably aligns closely with the kinds of informal, technology-mediated, and individual learning environments that many adult learners now engage with, making MiLIP a relevant tool for investigating contemporary language learning processes.

Finally, the audiovisual recordings and image generations were limited by budgetary constraints. The recordings were carried out by the research team in a sound-treated booth. However, to improve the quality of the recordings, a sound engineer or equivalent expert should be consulted where possible. AI was used to generate the images, which made it difficult and time-consuming to keep characters and objects consistent throughout a story. More advanced AI or images sketched by an artist could prevent these issues.

V Future applications

Our methodology seeks to establish the efficiency of mixed language input for language learning in controlled language learning contexts by using a familiar language in order to scaffold new language knowledge. Future applications of this methodological design are vast and wide-reaching. They can be applied to language learning for different language pairs as well as investigations of language learning with different and diverse populations situated in a variety of sociolinguistics contexts. For example, children (over the reading age) could be assisted by the multimodal design, and learning aptitude of older adults could be investigated. In theory, this method can be applied across typologically (dis)similar language pairs differing in, for instance, script and typological distance, as well as to speakers with different L1–L2(–Ln) linguistic profiles, and sociolinguistically diverse learners, such as bidialectal and heritage speakers. The tweaks to scaffolded language input lie in the researchers’ knowledge of the grammatical features of the languages which will need to be taken into consideration as the learning sessions progress. The paradigm can also be used to assess language learning in classroom settings compared with remote web-based settings, since learner experience (e.g., autonomy of the learner, number of learners present, predictability, and uniformity of content) between these contexts differs quite substantially. We are not claiming that MiLIP resembles classroom interactions generally, but, rather, we have drawn inspiration from how (linguistically diverse) classroom settings facilitate language learning.

This method also has the potential to transform our current thinking on how languages are learned by implementing mixed language methods inspired by effortless multilingual practices found in non-Western societies (A. Lightfoot et al., 2021). Findings from such efforts may provide alternative support, complementing pedagogical translanguaging research, to the unilingual and separationist norm currently adopted in global language teaching and learning practices that are guided by evidence for optimal learning strategies. Furthermore, our design takes into account and contributes to existing evidence showing multi-language activation when more than one language is available to learners (Soares & Grosjean, 1984; Wu & Thierry, 2017). The underlying idea of mixing languages to support learning, therefore, has the potential to be formally and more broadly applied as a beneficial pedagogical practice both in and outside the classroom. We encourage researchers and educators to adapt this methodology to their contexts and learning goals so that empirical evidence can guide effective language education.

Footnotes

Acknowledgements

We thank the pilot participants for their time in completing this study and for providing valuable feedback. Furthermore, thanks go to John Williams for giving us input about the design of the language learning conditions. We also thank the reviewers for their valuable feedback.

Data availability

All materials are available on the OSF, https://doi.org/10.17605/OSF.IO/P7JSW.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research is supported by a Cambridge Humanities Research Grant (CHRG).

Ethical approval and informed consent statements

This study received ethical approval from the Ethics Committees of the University of Cambridge. All pilot participants provided written informed consent prior to enrolment in the study. This research was conducted ethically in accordance with the World Medical Association Declaration of Helsinki.

ORCID iDs

Mandy Wigdorowitz

Chrysoula Vassiliu

Margreet Vogelzang

Notes

References

Alanís

(2000). A Texas two-way bilingual program: Its effects on linguistic and academic achievement. Bilingual Research Journal, 24(3), 225–248. http://doi.org/10.1080/15235882.2000.10162763

Anderson

(2022). The translanguaging practices of expert Indian teachers of English and their learners. Journal of Multilingual and Multicultural Development, 45(6), 2233–2251. https://doi.org/10.1080/01434632.2022.2045300

Auer

(Ed.). (2013). Code-switching in conversation: Language, interaction and identity. Routledge.

Auerbach

E.R.

(1993). Reexamining English only in the ESL classroom. TESOL Quarterly, 27(1), 9–32. https://doi.org/10.2307/3586949

Balam

(2021). Beyond differences and similarities in code-switching and translanguaging research. Belgian Journal of Linguistics, 35, 76–103. https://doi.org/10.1075/bjl.00065.bal

Bardel

Sánchez

(Eds.). (2020). Third language acquisition: Age, proficiency and multilingualism (EuroSLA Studies 3). Language Science Press.

Beatty-Martínez

A.L.

Parafita Couto

M.C.

Ameka

F.K.

Aboh

E.O.

(2025). Codeswitching. Reference Module in Social Sciences, 1–5. https://doi.org/10.1016/B978-0-323-95504-1.00503-2

Cenoz

Gorter

(2022). Pedagogical translanguaging. Cambridge University Press.

Cook

(1995). Multi-competence and learning of many languages. Language, Culture and Curriculum, 8, 93–98. https://doi.org/10.1080/07908319509525193

10.

Cook

(2001). Using the first language in the classroom. The Canadian Modern Language Review, 57(3), 402–423. http://doi.org/10.3138/cmlr.57.3.402

11.

Cook

(2016). Second language learning and language teaching (5th ed.). Routledge.

12.

Creese

Blackledge

(2010). Translanguaging in the bilingual classroom: A pedagogy for learning and teaching? The Modern Language Journal, 94, 103–115. http://doi.org/10.1111/j.1540-4781.2009.00986.x

13.

Cummins

(2007). Rethinking monolingual instructional strategies in multilingual classrooms. Canadian Journal of Applied Linguistics, 10(2), 221–240.

14.

Cummins

(2017) Teaching for transfer in multilingual educational contexts. In García

Lin

(Eds.), Bilingual education: Encyclopedia of language and education (3rd ed., pp. 103–115). Springer Science + Business Media LLC.

15.

Cummins

(2019). The emergence of translanguaging pedagogy: A dialogue between theory and practice. Journal of Multilingual Education Research, 9(13), 19–36.

16.

Cummins

(2021). Rethinking the education of multilingual learners: A critical analysis of theoretical concepts. Multilingual Matters.

17.

Deuchar

(2020). Code-switching in linguistics: A position paper. Languages, 5(2), 22. http://doi.org/10.3390/languages5020022

18.

Deuchar

Muysken

Wang

S-L.

(2007). Structured variation in codeswitching: Towards an empirically based typology of bilingual speech patterns. International Journal of Bilingual Education and Bilingualism, 10(3), 298–340. http://doi.org/10.2167/beb445.0

19.

Duñabeitia

J.A.

(2022). MultiPic: Multilingual picture database. Figshare. Dataset.

20.

Duñabeitia

J.A.

Baciero

Antoniou

Ataman

Baus

Ben-Shachar

Çağlar

O.C.

Chromý

Comesaña

Filip

Đurđević

D.F.

Dowens

M.G.

Hatzidaki

Januška

Jusoh

Kanj

Kim

S.Y.

Kırkıcı

, . . . Pliatsikas

(2022). The multilingual picture database. Scientific Data 9, 431. https://doi.org/10.1038/s41597-022-01552-7

21.

Ettlinger

Morgan-Short

Faretta-Stutenberg

Wong

P.C.M.

(2015). The relationship between artificial and second language learning. Cognitive Science, 40(4), 822–847. https://doi.org/10.1111/cogs.12257

22.

Foucart

Frenck-Mestre

(2012). Can late L2 learners acquire new grammatical features? Evidence from ERPs and eye-tracking. Journal of Memory and Language, 66(1), 226–248. https://doi.org/10.1016/j.jml.2011.07.007

23.

García

(2009). Bilingual education in the 21st century: A global perspective. Wiley-Blackwell.

24.

García

Lin

(2016). Translanguaging and bilingual education. In García

Lin

May

(Eds.), Bilingual and multilingual education, encyclopedia of language and education (pp. 117–130). Springer.

25.

Gardner-Chloros

(2009). Code-switching. Cambridge University Press.

26.

Gass

(1997). Input, interaction, and the second language learner. L. Erlbaum.

27.

Gentner

Boroditsky

(2001). Individuation, relativity, and early word learning. In Bowerman

Levinson

(Eds.), Language acquisition and conceptual development (pp. 215–256). Cambridge University Press.

28.

Green

D.W.

Abutalebi

(2013). Language control in bilinguals: The adaptive control hypothesis. Journal of Cognitive Psychology, 25(5), 515–530.

29.

Grosjean

(2010). Bilingual: Life and reality. Harvard University Press.

30.

Gullifer

J.W.

Titone

(2018). Compute language entropy with languageEntropy. https://github.com/jasongullifer/languageEntropy

31.

Howatt

A.P.R.

(1984). A history of English language teaching. Oxford University Press.

32.

Howatt

A.P.R.

Smith

(2014). The history of teaching English as a foreign language, from a British and European perspective. Language & History, 57(1), 75–95. https://doi.org/10.1179/1759753614Z.00000000028

33.

Huettig

Rommers

Meyer

A.S.

(2011). Using the visual world paradigm to study language processing: A review and critical evaluation. Acta Psychologica, 137(2), 151–171. https://doi.org/10.1016/j.actpsy.2010.11.003

34.

Jamieson

(2011). Assessment of classroom language learning. In Hinkel

(Ed.), Handbook of research in second language teaching and learning: Volume 2 (1st ed., pp. 768–785). Routledge.

35.

Kniaź

Zawrotna

(2021). Embedded English verbs in Arabic-English code-switching in Egypt. International Journal of Bilingualism, 25(3), 622–639. http://doi.org/10.1177/1367006920976909

36.

Krashen

S.D.

(1985). The input hypothesis: Issues and implications. Longman.

37.

Lambert

W.E.

Tucker

G.R.

(1972). Bilingual education of children: The St. Lambert experiment. Newbury House.

38.

Lara-Alecio

Galloway

Irby

B.J.

Rodríguez

Gómez

(2004). Two-way immersion bilingual programs in Texas. Bilingual Research Journal, 28(1), 35–54. http://doi.org/10.1080/15235882.2004.10162611

39.

Lazaridou

Marelli

Baroni

(2017). Multimodal word meaning induction from minimal exposure to natural text. Cognitive Science, 41(4), 677–705. https://doi.org/10.1111/cogs.12481

40.

Lidz

Gleitman

(2003). Understanding how input matters: Verb learning and the footprint of universal grammar. Cognition, 87(3), 151–178. https://doi.org/10.1016/S0010-0277(02)00230-5

41.

Lightfoot

Balasubramanian

Tsimpli

I.M.

Mukhopadhyay

Treffers-Daller

(2021). Measuring the multilingual reality: Lessons from classrooms in Delhi and Hyderabad. International Journal of Bilingual Education and Bilingualism, 25(6), 2208–2228. https://doi.org/10.1080/13670050.2021.1899123

42.

Lightfoot

(1989). The child’s trigger experience: Degree-0 learnability. Behavioral and Brain Sciences, 12, 321–375. http://doi.org/10.1017/S0140525X00049086

43.

Lin

(2013). Classroom code-switching: Three decades of research. Applied Linguistics Review, 4, 195–218. http://doi.org/10.1515/applirev-2013-0009

44.

Mackridge

(1987). The modern Greek language: A descriptive analysis of standard modern Greek. Oxford University Press.

45.

MacSwan

(2014). Programs and proposals in codeswitching research: Unconstraining theories of bilingual language mixing. In MacSwan

(Ed.), Grammatical theory and bilingual codeswitching (pp. 1–34). The MIT Press.

46.

MacSwan

(Ed.). (2022). Multilingual perspectives on translanguaging. Channel View Publications.

47.

Makalela

(2016). Ubuntu translanguaging: An alternative framework for complex multilingual encounters. Southern African Linguistics and Applied Language Studies, 34(3), 187–196.

48.

Mastropavlou

Tsimpli

I.M.

(2011). The role of suffixes in grammatical gender assignment in Modern Greek: A psycholinguistic study. Journal of Greek Linguistics, 11(1), 27–55. https://doi.org/10.1163/156658411X563685

49.

Mayer

R.E.

(2005). Principles of multimedia learning based on social cues: Personalization, voice, and image principles. In Mayer

R.E.

(Ed.) The Cambridge handbook of multimedia learning (pp. 201–212). Cambridge University Press.

50.

Mayer

R.E.

(2009). Multimedia learning (2nd ed.). Cambridge University Press.

51.

Morgan-Short

Steinhauer

Sanz

Ullman

M.T.

(2012). Explicit and implicit second language training differentially affect the achievement of native-like brain activation patterns. Journal of Cognitive Neuroscience, 24(4), 933–947. https://doi.org/10.1162/jocn_a_00119

52.

Muysken

(2000). Bilingual speech: A typology of CS. Cambridge University Press.

53.

Myers-Scotton

(1993). Duelling languages: Grammatical structure in codeswitching. Oxford University Press.

54.

Parafita Couto

M.C.

Bellamy

Ameka

F.K

. (2023). Theoretical linguistic approaches to multilingual code-switching. In Cabrelli

Chaouch-Orozco

González Alonso

Pereira Soares

S.M.

Puig-Mayenco

Rothman

(Eds.), The Cambridge handbook of third language acquisition (pp. 403–436). Cambridge University Press.

55.

Parafita Couto

M.C.

Gullberg

. (2019). Code-switching within the noun phrase: Evidence from three corpora. International Journal of Bilingualism, 23(2), 695–714. http://doi.org/10.1177/1367006917729543

56.

Pinker

Lebeaux

D.S.

Frost

L.A.

(1987). Productivity and constraints in the acquisition of passive. Cognition, 26(3), 195–267. http://doi.org/10.1016/S0010-0277(87)80001-X

57.

Poplack

(1980). Sometimes I’ll start a sentence in English y termino en español: Toward a typology of code-switching. Linguistics, 18(7–8), 581–618.

58.

Poplack

(2008). Code-switching. In Ammon

Dittmar

Mattheier

Trudgill

(Eds.), Sociolinguistics Volume 1 (pp. 589–596). De Gruyter Mouton.

59.

Poplack

(2018). Borrowing: Loanwords in the speech community and in the grammar. Oxford University Press.

60.

Poza

(2017). Translanguaging: Definitions, implications, and further needs in burgeoning inquiry. Berkeley Review of Education 6(2), 101–128. https://doi.org/10.5070/B86110060

61.

Reyes

(2004). Functions of code switching in schoolchildren’s conversations. Bilingual Research Journal, 28(1), 77–98. https://doi.org/10.1080/15235882.2004.10162613

62.

Robinson

(1995). Attention, memory, and the ‘noticing’ hypothesis. Language Learning, 45(2), 283–331. http://doi.org/10.1111/j.1467-1770.1995.tb00441.x

63.

Rolstad

Mahoney

Glass

G.V

. (2005). The big picture: A meta-analysis of program effectiveness research on English language learners. Educational Policy, 19(4), 572–594. http://doi.org/10.1177/0895904805278067

64.

Sanoudaki

Thierry

(2015). Language non-selective syntactic activation in early bilinguals: The role of verbal fluency. International Journal of Bilingual Education and Bilingualism, 18(5), 548–560. https://doi.org/10.1080/13670050.2015.1027143

65.

Schoonbaert

Hartsuiker

R.J.

Pickering

M.J.

(2007). The representation of lexical and syntactic information in bilinguals: Evidence from syntactic priming. Journal of Memory and Language, 56(2), 153–171. https://doi.org/10.1016/j.jml.2006.10.002

66.

Schütze

C.T.

Sprouse

(2014). Judgment data. In Podesva

R.J.

Sharma

(Eds.), Research methods in linguistics (pp. 27–50). Cambridge University Press.

67.

Schwartz

B.D.

Sprouse

R.A.

(1996). L2 cognitive states and the Full Transfer/Full Access model. Second Language Research, 12(1), 40–72. https://doi.org/10.1177/026765839601200103

68.

Soares

Grosjean

(1984). Bilinguals in a monolingual and a bilingual speech mode: The effect on lexical access. Memory & Cognition, 12, 380–386. https://doi.org/10.3758/BF03198298

69.

Spivey

M.J.

Dale

(2006). Continuous dynamics in real-time cognition. Current Directions in Psychological Science, 15(5), 207–211. https://doi.org/10.1111/j.1467-8721.2006.00437.x

70.

Stell

Yakpo

(2015). Code-switching between structural and sociolinguistic perspectives. De Gruyter.

71.

Sweller

Ayres

Kalyuga

(2011). Cognitive load theory. Springer.

72.

Treffers-Daller

(2025). Translanguaging: What is it besides smoke and mirrors? Linguistic Approaches to Bilingualism, 15(1), 1–26. https://doi.org/10.1075/lab.24015.tre

73.

Tsimpli

I.M.

(2014). Early, late or very late? Linguistic Approaches to Bilingualism, 4(3), 283–313. http://doi.org/10.1075/lab.4.3.01tsi

74.

Tsimpli

I.M.

Balasubramanian

Marinis

Panda

Mukhopadhyay

Alladi

Treffers-Daller

(2020). Research report of a four-year study of multilingualism, literacy, numeracy and cognition in Delhi, Hyderabad and Patna. University of Cambridge.

75.

Turnbull

(2001). There is a role for the L1 in second and foreign language teaching, but . . . The Canadian Modern Language Review, 57(4), 531–540. http://doi.org/10.3138/cmlr.57.4.531

76.

Van der Loo

. (2014). The stringdist package for approximate string matching. The R Journal, 6, 111–122.

77.

VanPatten

(1996). Input processing and grammar instruction in second language acquisition (Second language learning). Ablex.

78.

VanPatten

Williams

(2014). Theories in second language acquisition: An introduction. Routledge.

79.

Vaughan-Evans

Parafita Couto

M.C.

Boutonnet

Hoshino

Webb-Davies

Deuchar

Thierry

(2020). Switchmate! An electrophysiological attempt to adjudicate between competing accounts of adjective-noun code-switching. Frontiers in Psychology, 11, 549762. http://doi.org/10.3389/fpsyg.2020.549762

80.

Vogelzang

Wigdorowitz

Vassiliu

Tsimpli

I.M

. (In preparation). Fighting fraud in web-based empirical research: Identifying and preventing fraudulent submissions in linguistic and psychological research.

81.

Wei

(2011). Moment analysis and translanguaging space: Discursive construction of identities by multilingual Chinese youth in Britain. Journal of Pragmatics, 43(5), 1222–1235. https://doi.org/10.1016/j.pragma.2010.07.035

82.

Wei

(2018). Translanguaging as a practical theory of language. Applied Linguistics Review, 39(2), 9–30. https://doi.org/10.1093/applin/amx039

83.

Wigdorowitz

Pérez

A.I.

Tsimpli

I.M.

(2022). Sociolinguistic context matters: Exploring differences in contextual linguistic diversity in South Africa and England. International Multilingual Research Journal, 16(4), 345–364. https://doi.org/10.1080/19313152.2022.2069416

84.

Wigdorowitz

Pérez

A.I.

Tsimpli

I.M.

(2023). A holistic measure of contextual and individual linguistic diversity. International Journal of Multilingualism, 20(2), 469–487. https://doi.org/10.1080/14790718.2020.1835921

85.

Williams

(1994). Arfarniad o ddulliau dysgu ac addysgu yng nghyd-destun addysg uwchradd ddwyieithog [An evaluation of teaching and learning methods in the context of bilingual secondary education]. Unpublished PhD thesis, University of Wales.

86.

Williams

(2000). Welsh-medium and bilingual teaching in the further education sector. International Journal of Bilingual Education and Bilingualism, 3(2), 129–148. http://doi.org/10.1080/13670050008667703

87.

Williams

J.N.

(2020). The neuroscience of implicit learning. Language Learning, 70, 255–307. http://doi.org/10.1111/lang.12405

88.

Wonnacott

Newport

E.L.

Tanenhaus

M.K.

(2008). Acquiring and processing verb argument structure: Distributional learning in a miniature language. Cognitive Psychology, 56(3), 165–209. https://doi.org/10.1016/j.cogpsych.2007.04.002

89.

Y.J.

Thierry

(2017). Brain potentials predict language selection before speech onset in bilinguals. Brain and Language, 171, 23–30. https://doi.org/10.1016/j.bandl.2017.04.002

90.

(2024). Writing assessment literacy and its impact on the learning of writing: A netnography focusing on Duolingo English Test examinees. Language Testing in Asia, 14, 24. https://doi.org/10.1186/s40468-024-00297-x

Language mixing as a pedagogical tool for language learning: a methods development guide

Abstract

Keywords

I Introduction

1 Mixed Language Use

2 (Mixed) Language use in additional language teaching and learning

3 Language learning considerations

II Methodology

1 Procedure

2 Participant and eligibility checks

a Language proficiency

b Other eligibility criteria

3 Learning materials

4 Evaluation measures

a Find the picture: picture selection task

b Pick the phrase: a GFCT

i GFCTs in Phase 2

ii GFCTs in Phase 3

c Expressive vocabulary task

d Post-study questionnaire

III Stepwise methods guide

IV Limitations

V Future applications

Footnotes

Acknowledgements

Data availability

Funding

Ethical approval and informed consent statements

ORCID iDs

Notes

References