A Portrait of Lexical Knowledge among Adult Hebrew Heritage Speakers Dominant in American English: Evidence from Naming and Narrative Tasks

Abstract

While the field of heritage language (HL) bilingualism has grown substantially in recent years, no studies have considered heritage Hebrew speakers dominant in American English. Expanding HL studies to new language pairs is crucial to understand the generalizability of prior findings across diverse linguistic contexts. In the current study, we assess 40 adult participants (16 M, 24 F) and present an overview of their lexical abilities, as derived from a quantitative and qualitative analysis of performance on the Multilingual Naming Test (MINT) and a narrative elicitation task. We consider target accuracy, types of non-target responses, and cross-linguistic influence in the form of code-switching and calquing. Participants’ non-target responses indicated a strong grasp of Hebrew root-pattern word formation and creativity in the face of lexical gaps. Code-switching and calquing patterns in the narratives suggest that the dominant English is the clear framing language, from which speakers draw resources directly or indirectly. Although this linguistic blending leads to innovative lexical formations that would not be found in standard monolingual speech, the speakers’ overall message is still understandable. We conclude that heritage Hebrew speakers are able to clearly communicate complex thoughts in their HL while relying on their dominant language to fill lexical gaps.

1. Introduction

Heritage speakers (HSs) are early bilinguals who grow up with a heritage language at home (HL), which is distinct from the dominant societal language (SL) spoken around them (Benmamoun et al. 2013; Montrul 2016; Polinsky and Scontras 2020). As the HSs’ social lives begin to extend beyond the home, their second language, the SL, grows stronger than their first, the HL. In this way, they differ from the classically studied bilinguals who remain dominant in their first language. In recent years, interest in the phenomena of HSs’ linguistic development has grown, leading to numerous studies of distinct language pairs. Most studies focus on an HL in contact with dominant English, with an emphasis on HL-Spanish, Korean, Russian, and other languages commonly spoken in the US (Polinsky 2006; Montrul 2008; Scontras and Putnam 2020). One HL that has received little to no attention by studies of HL development is Modern Hebrew. The present work seeks to remedy this by presenting an overview of HL lexical proficiency in adult Hebrew HSs, in contact with American English.

Examining patterns exhibited by a highly understudied group can lend credence (or counterpoints) to prior findings from other language pairs. We can begin to tease apart characteristics that are common to all HSs from characteristics that are derived from attributes of the specific languages in question. On the path to a comprehensive understanding of HL linguistic systems and development, our analysis of Hebrew HSs will add a valuable data point to the growing body of the HL literature.

In this paper, we home in on the lexical knowledge of Hebrew HSs. Our interest in the lexicon is driven by several factors: first, there is relatively little work on HL lexical development as compared to other areas, such as morphosyntax (Montrul 2016; Polinsky 2018). Second, the lexicon has been shown to be more vulnerable to input factors than morphosyntax (Gharibi and Boers 2017). Finally, lexical proficiency has been found to correlate with other linguistic abilities (Fairclough and Garza 2018; Polinsky 2006; Polinsky and Kagan 2007; Montrul 2008), such that our study will provide a gateway to future works on HL-Hebrew grammars.

To establish a vocabulary baseline for Hebrew HSs, we conducted a MINT test (Gollan et al. 2012) in both the HL-Hebrew and the SL-English. We then elicited a free-form narrative in the HL, based on Mercer Mayer’s ‘Frog, Where Are You?’ (Mayer 1969), to examine vocabulary breadth and cases of cross-linguistic influence from the dominant English. Notably, we do not use a monolingual Hebrew control group to compare our HSs’ competence, as such comparisons have been contested in the field (Bayram et al. 2019; Bayram et al. 2021). Instead, we aim to give an overview of the Hebrew HSs’ performance in its own right.

1.1. Heritage Speakers and Their Lexical Abilities

When it comes to linguistic proficiency in the HL, HSs are a highly heterogeneous group, with some achieving very limited competency and others reaching a near-native level. They grow up with one or both parents, and often grandparents, speaking the HL, and in early childhood are often able to express themselves well in the HL (Polinsky 2018). While some were born to immigrant parents and encountered the SL in childhood, others were raised in a monolingual (HL) environment up until immigration and the attendance of school in the new SL, only emphasizing the claim of high HL proficiency at a young age. With the onset of schooling in the SL, HSs’ language balance shifts, with the second language becoming their more dominant one. As their social and academic/professional development, along with their everyday life, advances in the SL, their HL falls to dis- or under-use.

Consider the context of HL use. In most cases, the HL will be used only in the home, so the HS will have no need for, or experience with, HL vocabulary in other contexts, such as school. Thus, HL expression will be limited to a specific sphere of vocabulary. Furthermore, the HSs’ transition from HL to SL use does not happen in a vacuum: as HSs strengthen their SL, their parents, bilingual immigrants, go through a parallel process of assimilation, and might begin mixing their native language with the SL more often than in the HS’ youth. Thus, not only is the quantity of HL input constricted in time and context, it is no longer even exclusively HL input, as it begins to blend with the SL.

This phenomenon of limited HL vocabulary as a result of decreased input has been widely documented for HSs of different age groups (Bialystok et al. 2010; Bialystok and Luk 2012). Polinsky (2006) showed that HSs compensate for HL vocabulary gaps by code-switching and drawing from their SL. Finding systematic patterns in this blending of HL and SL could lend credence to the claim that rather than being under-proficient HL speakers, HSs are in fact native speakers of a new language variety derived from two languages in close contact (Rothman 2009), whose lexicon fundamentally differs in structure from monolinguals (Rakhilina et al. 2016). This new language variety, while structurally distinct from its home country baseline, meets the lexical and expressive needs of HSs in their daily lives.

We can begin to pinpoint specific areas of divergence from the home country baseline by considering non-target responses to a picture-naming task, where target responses are calibrated based on the most commonly produced responses from monolinguals. Categorizing HSs’ non-target responses can prove insightful about the mental representations of vocabulary within the bilingual lexicon. The limited literature available focusing on categories of HSs’ non-target responses has been inconclusive, with most studies focusing on child HSs and finding often contradictory patterns. McGregor et al. (2002) found participants to be prone to semantically related non-target responses (e.g., pot -> stove) over phonologically related ones (e.g., pot -> pop) at a rate of about 20 to 1, while Makarova and Terekhova (2020) found that phonologically related responses were considerably more common. Makarova and Terekhova (2020) also found a tendency toward hyponyms, or overly-specific terms (e.g., flower -> daisy), while Kopotev et al. (2020) and Klapicová (2018), in contrast, found a tendency toward hypernyms or generalizations (e.g., flower -> plant). Ringblom and Dobrova (2019) found similar trends in non-target response patterns between both HSs and monolinguals, a finding that was reiterated by Fridman and Meir’s (forthcoming) study of noun production in child and adult Russian HSs with two different SLs. Rakhilina et al. (2016) reported that Russian HSs in the US use considerably more calquing, or translating from a dominant language, than direct borrowing, while Kagan et al. (2021) found more borrowing among Russian HSs in the US than Russian HSs in Israel. Johannessen (2018) found that Norwegian HSs in the US tend toward borrowing on verbs more than nouns, contrasting Fridman and Meir’s (forthcoming) finding that Russian HSs dominant in English and in Hebrew both borrow considerably more on nouns than on verbs. Language-specific differences in points of challenge for different groups of HSs have been primarily demonstrated for the field of morphosyntax: for example, Russian and Korean HSs have been shown to struggle with relative clauses in the HL, while Spanish HSs have not found this construction particularly challenging (Benmamoun et al. 2013). By contrast, few studies focus on the heritage lexicon, and as demonstrated above, findings are not always corroborated across HL-SL pairs.

Thus, the present study provides novelty in several ways. In addition to profiling a previously unstudied group of HSs, we hope to bring some clarity to an otherwise sparse area of research, the HL lexicon. A careful examination of non-target response patterns can teach us how linguistic information is stored and represented in the brain. For example, associative and co-hyponym responses often indicate the presence of a concept, rather than an exact word, in the mental lexicon (Altman et al. 2017). Phonological responses may suggest a strong semantic representation of the word in the lexicon but difficulty in producing it (McGregor et al. 2002). These responses, which indicate a familiarity with a concept but a weak phonological representation for it, can be attributed to performance rather than competence (Foygel and Dell 2000). Hypernyms, holonyms, hyponyms, and metonyms demonstrate an understanding of general semantic groups and their members, while explanations point to an acute understanding of the target, and the use of creativity and broader knowledge to fill a specific lexical gap. Similarly, cross-linguistic responses indicate that an individual fully understands the target and draws on dominant, more readily-available, language systems to produce it.

While no studies have focused on lexical production patterns in heritage Hebrew, nearly all of the studies cited above focus on a heritage language in contact with English. Thus, an additional novel aspect of this work is an exploration of whether findings for other HLs in contact with English will be corroborated for Hebrew as well. If this is the case, we have a stronger claim for the universal properties of HL lexical development. If not, we will need to consider potential properties of Hebrew as a heritage language, and its speakers in the United States, that could lead to divergence.

1.2. Hebrew Typology

Hebrew and English are typologically different languages: Hebrew is a Semitic language, while English is a Germanic language. They are distinct at multiple levels. The orthographies of the two languages are entirely distinct, using different writing systems and even writing directions. Phonologically, most consonants are quite similar between the two languages (see Cohen 2009 for an overview), with a few key distinctions, while the two languages vary with respect to vowels: 5 vowels in Hebrew (Laufer 1990) and 16 vowels in English (13 monophthongs and 3 diphthongs) (Wells 1982). Considering morphology, most Hebrew words are formed by combining a three-or-four-letter root and a vocalic pattern. Words are constructed by applying vocalic patterns to a consonant-based root (e.g., CeCeC, where C-C-C is the root) or by appending prefixes and/or suffixes for a more morphologically complex formation (e.g., CaCCan, maCCeCa), with words under three syllables generally following the former format and longer words, the latter (Altman et al. 2017; Ravid 1990). For example, the root ‘D-B-K’ yields the two-syllable word devek ‘glue’ and also the three-syllable word madbeka ‘sticker’. As such, all Hebrew verbs and most Hebrew nouns are non-concatenative, although nouns are more varied (Laks 2019; Deutsch et al. 1998). This is in stark contrast with English, which is fundamentally concatenative, with only a few exceptions (Fullwood and O’Donnell 2013). Borrowings from other languages are adopted into Hebrew through a process of root extraction and conjugation into one of the existing vocalic templates, a phenomenon documented for Hebrew speakers as young as 3–4 (Berman 2003).

At the lexical level, the source languages from which Modern Hebrew draws are also quite distinct from those of English, with the former drawing primarily from Aramaic, Arabic, English, and Russian (Cohen 2019), and the latter drawing largely from Latin, French, German, and Greek (Durkin 2014). Thus, there are relatively few cognates between the two languages. In an overview of dominant Hebrew’s interaction with several minority languages in Israel, Meir et al. (2021) found that Hebrew vocabulary infiltrates even language communities that resist integration into Hebrew-speaking society, such as the ultra-orthodox speakers of Yiddish. While this is undoubtedly a phenomenon characteristic of the dominant–minority language dynamic, it is possible that it is also connected to the properties of Hebrew itself, as it may highlight Hebrew’s tenacity in situations of language contact. In order to explore this possibility further, we consider the status of Hebrew as an HL, and of the greater Hebrew-speaking community, in the United States.

1.3. Hebrew Speakers in the United States

The present study focuses on Israeli Jews who grew up in the United States1, and as such are part of two diasporas: the greater Jewish diaspora and the much newer Israeli diaspora. An extension of this dual identity, Hebrew differs from many other HLs (such as Spanish, Russian, etc.) in that it is both the national language of Israel and has also been the historic and religious language of an ethnic group for centuries. As such, many American Jews without family in Israel use Hebrew as a means of connecting to their culture and heritage, and Hebrew, together with Yiddish, serves as a key contact language for Jewish English dialects (Hary and Benor 2018, p. 422). It thus follows that heritage Hebrew as spoken in the homes of Israeli emigrants will become closely entangled with Hebrew elements of the Jewish English that pervades the communities in which these emigrants live.

As of 2000, about 1% of Israeli citizens resided outside of Israel, with the majority of them in the United States (Gold 2005 viii, p. 24). Those in the United States primarily congregate around New York City, Los Angeles, and Miami (Rebhun and Lev-Ari 2013), although in recent years communities have grown around most major metropolitan areas in the country (IAC Impact Report 2021). By most measures (education, income, etc.), Israelis can be considered one of the most successful migrant groups in the US (Gold 2005, p. 10). However, in spite of their positive acclimation into American society, they maintain a strong connection with Israel and their Israeli identities. They remain a distinct enclave within American Jewry, although highly interactive and involved with their non-Israeli Jewish peers and local communities (Gold 2005). For example, they tend to be considerably more secular, eat different foods, and subscribe to slightly different cultural norms than their American Jewish counterparts, and, of course, they usually speak Hebrew at home. The unique status of Hebrew in both Israeli and Jewish identity, in conjunction with Israelis’ known tendency to maintain strong cultural ties with their country of origin even when well-adjusted in the United States, naturally leads to questions of how Hebrew HSs preserve and express their language as adults. In the present work, we take key steps to uncovering the answers to these questions.

1.4. Research Questions and Hypothesese

The aim of this paper is to examine Hebrew HSs’ vocabulary in their heritage and dominant languages, with a focus on SL influence on the HL. We posed the following questions:

• How do Hebrew HSs compare on HL and SL vocabulary production?
• What HL non-target responses do Hebrew HSs produce and do they follow a pattern?
• What characterizes their HL narrative production?

We wanted to understand patterns of vocabulary production among Hebrew HSs both on a prescribed picture-naming task and on a free-form narrative task. Specifically, regarding RQ 1, we wanted to know how many words Hebrew HSs could produce accurately in their HL compared to their SL. We predicted that accuracy would be substantially higher in the SL. Such a result would be in line with recent studies comparing HL and SL performance, such as Macbeth et al. (2022), who found that HSs of different languages performed almost twice as accurately on MINT in their SL-English than in their HLs. Approximately the same ratio was also reported in Tao et al. (2021) for HL-Spanish and HL-Mandarin speakers with SL-English.

For RQ 2, we predicted that the most common non-target response type would be ‘Unknown’ followed by ‘Semantic’. Within ‘Semantic’, we predicted ‘Associative,’ ‘Hypernym,’ and ‘Explanation’ to take the lead as non-target response subtypes, as these were found consistently for adult HSs of Russian in the US and in Israel (Fridman and Meir, forthcoming). We thus suspected that they may be indicative of a generalized pattern in adult HSs, regardless of language pair.

For the narrative task, which we conducted only in the HL, we wanted to evaluate the average vocabulary size and type–token ratio (TTR), as well as to note cases of code-switching and calquing, which would indicate a shift from the baseline language as a result of contact with the SL. Thus, RQ 3 aimed to encapsulate the trends herewithin.

2. Materials and Methods

2.1. Participants

Forty Hebrew HSs dominant in American English participated in this study. The participants (16 men and 24 women) ranged in age from 18–44 (n = 26.3) and their ages of onset of English ranged from 0–10 (n = 2.8) years, with 23 participants having been born in Israel. Three of the participants had only 1 Hebrew-speaking parent and all were of medium to high socioeconomic status. Thirty-nine participants rated their SL knowledge at 5 out of 5, and one at 4 out of 5. By contrast, the average rating for HL knowledge was 3.7 out of 5 (SD = 0.78, range of 0–5). We controlled for language exposure such that participants’ only home languages growing up would be English or Hebrew: that is, individuals whose parents had immigrated from a non-English-speaking country to Israel, and who were therefore exposed to more than just one HL in the home setting, were not accepted for this study, to avoid confounding variables.

2.2. Procedure

Participants were recruited through personal connections, word-of-mouth, and an Israeli-American mailing list, using a snowballing recruiting technique. Nearly all participants were at some point involved in Israeli organizations such as Tzofim (Israeli Scouts), university groups, or professional networking circles, and were thus able to recommend other participants whom they knew through these activities. Participants were asked to complete a brief background questionnaire and then join a Zoom call during which they completed the two experimental tasks: Participants completed the MINT task (Gollan et al. 2012) first in their dominant English and then in Hebrew, with no additional prompting or hints. Then they were asked to tell a story in Hebrew based on Mercer Mayer’s ‘Frog, Where Are You?’ (Mayer 1969).

2.3. Coding Schemata: MINT

MINT data were first coded binarily as target (1) or non-target (0). Next, we divided the non-target responses according to a system based on previous work by Fridman and Meir (forthcoming), Foygel and Dell (2000), Altman et al. (2017), and Ramsay et al. (1999). Each non-target response was coded with a type, including Semantic, Morpho-Phonological, Both Semantic and Morpho-Phonological, Cross-Linguistic, Perceptual, and Unknown. Semantic non-target responses were further subdivided into Associative, Co-Hyponym, Explanation, Holonym, Hypernym, Metonym, and Synonym. Cross-Linguistic non-target responses were further subdivided into Borrowing, Calque, Mixing, Culturalism, L3, and Phonetic. A full list of this coding system with examples and definitions can be found in

Table 1. Coding Schemata.

Non-Target Response Type	Subtype	Definition	Example (Target–Non-Target)
Semantic	Associative	A word related to the target by some association	kluv ‘cage’–kele ‘jail’
	Co-Hyponym	A member of the same category as the target	tapuax ‘apple’–tapuz ‘orange’
	Explanation	Using a string of words to describe the target	megahetz ‘iron’–mashu xam le bgadim ‘something hot for clothes’
	Holonym	A larger whole of the target part	tris ‘blinds’–xalon ‘window’
	Hypernym	The wider category containing the target	nura ‘lightbulb’–or ‘light’
	Metonym	Contents of the target	ken ‘nest’–beitzim ‘eggs’
	Synonym	An equally acceptable term for the target	teka ‘plug’–shteker ‘plug’ (older term)
Morpho-Phonological		A word or non-word morpho-phonologically similar to the target	xilazon ‘snail’–*zilaxon ‘snail’
Both Semantic and Morpho-Phonological		A response that is both semantically and morpho-phonologically similar to the target OR a morpho-phonological error on a word semantically related to the target	nura ‘lightbulb’–menora ‘lamp’ boreg ‘screw’–*mamser ‘nail’ (masmer)
Cross-Linguistic	Borrowing	Responding in the SL	tavas ‘peacock’–peacock
	Calque	Translating an SL term into the HL	mad laxatz ‘pressure gauge’–sha’on hafsaka ‘stop watch’
	Mixing	Combining a root from the SL with HL morphosyntax	magrefa ‘rake’–merokek ‘agent that rakes’, from the English root ‘rake’’
	Culturalism	Invocation of a cultural or religious term	pe’a ‘wig’–sheitel ‘wig’ (Yiddish); a term for the head covering worn by Orthodox Jewish women
	L3	Responding in a language other than the SL or HL	matos ‘airplane’–avion ‘airplane’ (French)
	Phonetic	Producing an HL word based on the sound of the SL translation of the target	xilazon ‘snail’–snai ‘squirrel’ (sounds like hebraicized ‘snail’)
Perceptual		A miscomprehension of the picture prompt	tzaif ‘scarf’–magevet ‘towel’
Unknown		No response	-

.

2.4. Coding Schemata: Narrative

Each Hebrew narrative was transcribed in Latin letters. Prepositions and articles, attached in the written Hebrew to the root lexeme, were written as separate words so that they would be counted as a single unique token. We also counted the number of different types. For example, ha- davar ‘the thing’ and ha- yeled ‘the boy’ would be counted as three types: ha-, davar, and yeled. Instances of code-switching from English were tagged with (cs). We did not differentiate between discourse markers or ’thinking out loud’ and contextual word insertions, reasoning that both equally indicate a mental/verbal shift between the framing and inserted languages.

Once the narrative was fully transcribed, we used a program written in Python to count total tokens, target (Hebrew) tokens, unique tokens, and unique target (Hebrew) tokens. Words in English were totaled without accounting for uniqueness, as we were investigating frequency of code-switching rather than English vocabulary breadth. In addition to this number of English words, we counted the number of switches from Hebrew to English. This switch could be one inserted word (i.e., hu roeh et ha- JAR ‘he sees the jar.ENG’) or a string of words (i.e., ‘I don’t remember how to say tree’), each counted as a single instance. Adding this metric in addition to the word-by-word count allowed us to identify specific moments of shifting between languages, and provided us a more apt comparison for our next metric: calques. We defined calques as multiword phrases considered ungrammatical in standard Hebrew but clearly translated verbatim from valid grammatical phrases in English. For example, one participant noted that the story characters searched mitaxat *ha- mita ‘under the (lit.) bed’ instead of the accurate mitaxat LA- mita ‘under to the bed’.

Our goal with this analysis was to consider two perspectives. First, by calculating the TTR and target TTR based on individual words in English, we could understand at the word level the proportion of narrative expressed in the HL vs. the SL. Second, by considering switching instances as transitions from a one-or-more word string of Hebrew to a one-or-more word string of English, and comparing these to instances of one-or-more word strings of calques, we could use this parallel metric to assess strategies for SL reliance.

3. Results

3.1. Quantitative Vocabulary

On the SL-English MINT test, the average participant score was 92% (SD = 5%), with the lowest score at 79% and the highest at ceiling. By contrast, on the HL-Hebrew MINT test, the average score was 56% (SD = 16%), with scores ranging from 15% at the lowest to 82% at the highest. The scores for each of the two languages differ significantly from one another at t(39) = 16.1, p < 0.0001. Notably, even with the wide variability, the median values for each language were quite similar- 93% for English and 56% for Hebrew. The mode value for English accuracy was 93%, and 46% for Hebrew accuracy.

Figure 1 clearly illustrates the emphatic difference between HL and SL performance. We first observe that the score range for HL-Hebrew is both much larger and much lower than that for SL-English. The lines connecting the two box MINT boxplots represent each individual participant. We observe that the few lowest performers on the Hebrew MINT also performed more poorly on the English MINT. However, this pattern is not held for the top performers of the Hebrew MINT, who do not significantly outperform their peers on the English MINT. Rather, we see participants from all across the Hebrew score range converging near ceiling level on the English MINT.

3.2. Non-Target Responses

Figure 2 below illustrates the number of non-target responses produced in the HL by type. Most starkly, the most common non-target response type was ‘Unknown’, produced over four times more than the next-most-common type- ‘Semantic’, which in turn was produced over five times more than the third-most-common type- ‘Morpho-Phonological’.

Next, we dive into non-target response subtypes of the Semantic type (Figure 3). Participants’ most commonly produced subtype was ‘Hypernym,’ the category of which the target is part. The two most common target words eliciting hypernyms were ashnav ‘porthole’ (participants responded with xalon ‘window’) and nura ‘lightbulb’ (participants produced or ‘light’). Hypernyms were followed by ‘Associative’ as the next-most-common subtype. Contrasting Hypernyms’ uniform trends, the third-most-common Semantic subtype, Explanation, saw considerably more diversity in targets and responses. As an illustration, five participants described ashnav ‘porthole’ as some variation of ‘window on a boat/ship’; meanwhile, although five participants tried to explain the term yaeh ‘dustpan,’ no two explanations were alike2.

Table 2. Cross-linguistic Non-target Responses by Subtype.

Cross-Linguistic Subtype	Target	Response
Borrowing	richrach ‘zipper’ (x6) seren ‘axle’ tavas ‘peacock’ teka ‘plug’ nadneda ‘seesaw’ mad laxatz ‘pressure gauge’	ziper bar pikok charger seesaw meter
Calque	mad laxatz ‘pressure gauge’	sha’on hafsaka (lit) ‘stop watch’
Culturalism	notza ‘feather’ pe’ah ‘wig’	we just did it for bdikat xametz sheitel ‘head covering for religious Jewish women’
L3	matos ‘airplane’	*avion (French)
Blending	magrefa ‘rake’	*merokek
Phonetic	xilazon ‘snail’	snai ‘squirrel’

dives into the cross-linguistic non-responses by subtype. Six different participants produced ziper in place of the target richrach ‘zipper’. This and most other examples of ‘borrowing’ are italicized to indicate that they were produced using an Israeli accent and Hebrew phonology, as opposed to ‘seesaw’ which was said in American English pronunciation. There are several possibilities here: first, participants might be improvising a lexical gap in the hopes that this term in English, like many others in the contemporary lexicon, has been hebraicized and serves as an acceptable Hebrew response. Furthermore, this behavior reflects a general tendency of extensive borrowing of English words into Modern Hebrew (Laks 2018). Alternatively, these might be words that are used specifically within the participant’s family, in place of the standard Hebrew term. If the latter is the case, we could be looking at examples of rapid language change in a bilingual environment (Kupisch and Rothman 2018).

The calque sha’on hafsaka translates literally as ‘clock/watch break’, where hafsaka is derived from the verb lehafsik ‘to stop’. It was the result of a perceptual error in which the participant misidentified the pressure gauge image. The culturalism produced for notza ‘feather’ references bdikat xametz ‘checking for xametz’, which is a Jewish cleaning ritual conducted just before the Passover holiday, in which a feather is used to find any remaining breadcrumbs (or traces of other foods containing leavening agents) around the home. Thus, rather than producing the target word for ‘feather’, the participant associated it with a recent religious activity for which it was used. Another participant produced the term sheitel ‘head covering for religious Jewish women’ in place of pe’ah ‘wig’. This response could have alternately been categorized as L3, as it is a Yiddish word. However, due to the integration of Yiddish terminology into standard American Jewish English, as described in Section 1.3, we opted to categorize this response as a culturalism stemming from the religious circle in which the participant was raised.

One participant, fluent in French as an L3, produced the word avion ‘airplane (French)’ in place of the target matos ‘airplane,’ potentially also triggered by the word’s similarity to the Hebrew aviron, an older term for ‘airplane’ most commonly depicted in children’s songs and picture books. Another participant took the English word ‘rake’ and drew on knowledge of the Hebrew root system and grammatical agent construction to form the neologism merokek, which, were this a real Hebrew root for ‘rake’, would mean ‘raker’. Finally, a particularly interesting phenomenon occurred in one participant’s production of the Hebrew xilazon ‘snail’. The participant produced snai, the Hebrew word for ‘squirrel,’ which sounds like the English ‘snail’, thus likely perceiving the picture prompt first in English, and then applying Hebrew phonology to produce the same sounds, in spite of its divergent meaning.

Not all morpho-phonological non-target responses subscribed to a particular pattern. Most responses involved an erroneous phoneme paired with a target one for the target word. These types of responses indicate a production, not an access problem. It is likely that participants who produced these kinds of non-target responses would accurately recognize and identify the target word in context (for example, in a comprehension task); however, they struggle in the production phase, only accurately remembering part of the word. Other responses, however, fell into a clear category of responses joined by multiple participants, as summarized in

Table 3. Some Examples of Morpho-Phonological Non-target Responses.

Morpho-Phonological Non-Target Response Pattern	Target	Response
General phoneme error	xetz boreg aniva mitznax	ketz bored aneva mitzax
Correct root–incorrect formation	maFTeaX maN’uL Be’ER BoReG PEaH	miFTaX miN’aL Ba’OR BeReG PaEH
Gender marking	nadneda salsala mazleg	nadned salsal mazleget

.

The two most notable categories were root conjugation errors and erroneous gender marking. Recall that Hebrew nouns and verbs use a root system: three- or four-letter roots, arranged into binyanim (lit. buildings), or grammatical verb class paradigms. Several participants produced non-target responses using the correct root but with an inaccurate vocalic pattern3. In addition to root-based word formation, Hebrew nouns mark masculine and feminine grammatical gender. With a few exceptions, feminine nouns end in -a or in -et, and masculine nouns end in a consonant other than t. Multiple participants took a feminine noun and produced it without the target -a at the end, or, on the contrary, took a masculine noun and added the feminine marking -et. Both of these non-target response categories indicate, beyond just the difficulties in production and not access that are common to all morpho-phonological non-target responses, that participants have a familiarity and strong grasp of various Hebrew morphosyntactic features, and struggle primarily with lexical assignment.

This conclusion is supported further by examples from the ‘Both’ category, which included (1) real words both semantically and morpho-phonologically related to the target and (2) morpho-phonological errors on words semantically related to the target. Of the 15 responses in this category, 5 participants produced the word menora ‘lamp’ in place of the target nura ‘lightbulb’. An additional two produced naul ‘locked (adj)’ instead of man’ul ‘lock (noun)’. One produced tzanxanim ‘paratroopers’ in place of the target mitznax ‘parachute,’ potentially more familiar with the former as a prestigious Israeli military unit. These and several other examples provide more evidence for Hebrew HSs’ strong morpho-phonological awareness and understanding both of the target and of some/most of its morpho-phonological components, with a struggle only in final-stage production4. The examples from this category in particular seem to draw on the familiarity of more frequent terms in the production of morpho-phonologically related ones.

Next, we consider additional observations from participant responses. Two participants, both in their 30s, used the term shteker, a synonymous term common in midcentury Israel, instead of the target teka ‘plug’. This usage suggests that the participants heard a term from older generations (parents or grandparents) and solidified it, without updating it to a modernized Hebrew term. Previously, the usage of old-fashioned lexical and grammatical forms has been noted for other HLs (see Lee and Shin 2008; Johannessen and Laake 2015; Wiley 2007). We also took note of ‘child-like’ vocabulary and forms, and found 11 such cases. Of these, 7 cases saw the prompt matos ‘airplane’ produced as aviron. As noted previously, aviron is the original Hebrew term for airplane coined by Itamar Ben-Avi, the first native speaker of Modern Hebrew. However, the poet Haim Nahman Bialik proposed the word matos instead, and this latter term became standardized and popularized, while the original aviron is now primarily relegated to children’s media. Of the remaining 4, two were the pet name dubi ‘teddy bear’ in place of dov ‘bear’. In the final two instances, for the prompt xilazon/shavlul_‘snail’, participants produced the word beraleh, which is the name of a fictional snail from children’s songs and books. Most of these cases indicate childhood exposure to written and spoken media in the HL, and the preservation of these child-like terms in the mental representation, over more ‘adult’ standard terminology.

3.3. Narrative

In this section, we present a quantitative analysis of lexical breadth and a qualitative analysis of code-switching and calquing in participant narratives. While we made note of some of the most common pain points for participants, gender assignment and agreement, definiteness, pro-drop, tense conjugation, prepositions, and some syntactic constructions, these are not discussed here in detail as they are beyond the scope of the present work.

Participants produced narratives of a wide range of lengths (from 199 to 692 lexical units, tokens) and of a wide breadth of vocabulary (ranging from 68 to 246 unique lexical units, types). The number of target lexical units (in Hebrew) did not differ much from the total number of lexical units, such that the average TTR and average target TTR were very close, at 0.36 and 0.35, respectively. Similarly, the average number of code-switching instances (indicating a switch from HL to SL, whether one or several words long) and the average number of calques (direct translations from the SL, could also be one or several words long) were also very close at 3.5 and 3, respectively. However, the range of code-switching instances was almost twice as wide as the range of calques. The above data are summarized in

Table 4. Means and Ranges of Selected Narrative Measures.

	Total Number of Lexical Units, Tokens	Number of Unique Lexical Units, Types	Total Number of Target Lexical Units, Target Tokens	Number of Unique Target Lexical Units, Target Types	TTR	Target TTR	Number of Code-Switching Instances	Number of Calques
Average (SD)	379.5 (114.9)	134.3 (34.4)	368.8 (119.9)	127.1 (37.3)	0.36 (0.06)	0.35 (0.06)	3.5 (4.80)	3 (3.25)
Range	199–692	68–246	171–692	46–246	0.24–0.52	0.24–0.52	0–23	0–13

.

The number of code-switching instances and the accuracy on the Hebrew MINT had a Pearson correlation coefficient of −0.52 (p = 0.01), such that more code-switching instances were correlated with lower picture-naming performance. Likewise, negative correlations were found between this measure and the number of unique and total target and overall lexical units, with the strongest correlation coefficient with unique target lexical units, 0.37 (p = 0.03). The number of calques had no significant correlations with any of the other measures. The number of unique target lexical units predictably correlated strongly with accuracy on the Hebrew MINT (r = 0.66, p < 0.0001), confirming the notion that both measures serve as indicators of broader HL vocabulary. While the averages in both groups do not differ significantly, Figure 4 indicates that there is often, but not always, an inverse relationship between the number of calquing and code-switching instances. This suggests that most participants either strive to use only the HL and in moments of uncertainty translate from the SL into the HL (which they know is the target language for this task), or they mostly switch between the languages fluidly, either inserting words from the SL to fill an HL lexical gap or inserting sections of SL inner speech into their HL narrative.

We present the following examples of code-switching instances for consideration:

1.

ve- yotzet yanshuf I THINK yotze yanshuf

‘and out comes.F an owl I THINK out comes.M an owl’

2.

yesh lahem mishpaxa AW THAT’S CUTE

‘they have a family AW THAT’S CUTE’

3.

ha- kelev leyad ha- I HAVE THE WORD TREE ON THE TIP OF MY TONGUE THAT’S THE THING I SAY THIS WORD ON THE DAILY I DON’T KNOW WHY I’M FORGETTING IT BUT HE’S NEXT TO THE TREE ve- hem od mexapsim

‘The dog next to the [English] and they are still looking’

4.

ha- tzfardea yatz’a me- ha- GLASS JAR

‘the frog came out of the GLASS JAR’

5.

ve- pitom ex omrim OWL

‘and suddenly how do you say OWL’

6.

leyad ha- mashu shel dvorim I DON’T KNOW WHAT A BEEHIVE IS CALLED

‘next to the something of bees I DON’T KNOW WHAT A BEEHIVE IS CALLED’

7.

ve- az ha- BEES hem osim dvash ex korim lahem loh zoxeret OKAY ve- az ha- xarakim

‘and then the BEES they make honey what are they called don’t remember OKAY and then the bugs’

Example 1 shows an SL insertion when the participant considers, then self-corrects, gender assignment. This insertion seems to portray the participant thinking through a problem or approaching a challenge, in which the thinking act itself occurs in the framing SL. In Example 2, a participant adds external commentary on the narrative using the SL. This, too, seems to indicate a distancing from the narrative wherein the narrator (the participant) is still primarily acting in the SL. The inner speech in Example 3 follows a participant’s struggle to remember a particular word. The length of this insertion before the decision to just convey the intended message directly in the SL substantiates the idea that the SL is the main language used for thought and processing, even when the speaker is in ‘HL-mode’, telling a story in the HL. This stands in direct contrast to Example 4, in which the participant simply inserts the missing HL term with a familiar SL term, without verbalizing the frustration of missing the fitting HL term. Contrast Examples 3 and 4 with Example 5, in which we do see inner thought included, although in the HL, with only the missing lexeme inserted in the SL. In Example 6, we see a similar inner speech insertion to Example 3, only this time as an addendum to a first-attempted explanation in the HL. In Example 7, the participant’s first reaction was to fill the lexical gap in the SL, then think out loud in the SL, accept the memory gap, and continue with an associative word in the HL. Although the above examples are only seven out of many utterances across the 40 participants, they provide an understanding of the breadth of code-switching strategies used by Hebrew HSs, which span lexical insertions, inner speech, discourse markers, and beyond. We also observe processing of lexical gaps happening in real time, sometimes in the HL and other times in the SL, shedding light on the ways in which HSs juxtapose multiple languages in their mental lexicon.

Next, we present some examples of calques:

8.

hu *mitkasher la- tzfardea shelo

‘he calls his frog’

9.

eifo ha- tzfardea shelo *bifnim ha- etz

‘where is his from inside the tree’

10.

hem korim ve- *mistaklim la- tzfardea

‘they call and look for the frog’

11.

hem karu *bishvil ha- tzfardea

‘they called for the frog’

12.

im *davar al ha- rosh shelo

‘with a thing on his head’

13.

exad mihem *loh *afilu al ha- etz

‘one of them not even on the tree’

14.

hem kaasu *ito

‘they were angry with him’

In Example 8, the participant uses the term mitkasher ‘to call [on the phone]’ instead of koreh ‘to call [out to]’. In Example 9, the correct term would be betox ‘inside [something]’, but instead, the participant produced bifnim ‘inside,’ which does not take an object, and is also often used to mean ‘indoors’. In Example 10, the participant uses mistaklim ‘they look’ with la ‘for’, instead of mexapsim ‘they look/search for,’ which would be followed by the definite accusative particle et and the object. In Example 11, the participant uses bishvil ‘for /in order to’ with ha ‘the’, instead of la ‘to/for the’. The term bishvil, whether due to its meaning of ‘for’ or its (perhaps more common) meaning of ‘in order to’, which in English can be condensed to just ‘to’, has converged with le ‘to/for’ in the participant’s lexicon. In Example 12, the term davar ‘thing’ is used instead of the grammatically and contextually correct mashehu ‘something,’ to indicate an unknown object. In Example 13, the participant directly translates ‘not even’ as loh afilu when grammatically this should be afilu loh ‘(lit.) even not’ or bixlal loh ‘not at all’. In Example 14, the correct format in Hebrew would be kaasu alav ‘(lit.) angry on him’, rather than the English translation of ‘angry with him’. Thus we can see that there is no uniform approach to calquing in these narratives, with some participants struggling with prepositions, others with pronouns, and yet others with verbs or expressions.

4. Discussion

The present paper set out to characterize the HL lexical abilities of a previously unstudied group of heritage speakers, adult Hebrew HSs dominant in American English. To this end, 40 participants completed MINT vocabulary tests in both their HL and their SL and produced a narrative in their HL. Here, we summarize our findings and contextualize them among previous studies of HSs.

When comparing our participants’ HL-Hebrew and SL-English MINT performances, we see the classic picture of HSs: HL performance is highly heterogeneous, with a range of 67% from lowest to highest score (15–82%), while SL performance has a range of only 21% (79–100%), and a significantly higher average. Among non-target responses, a breakdown of Semantic responses found that the top three subtypes were ‘Hypernym,’ ‘Associative,’ and ‘Explanation,’ replicating the top three subtypes found for adult Russian HSs in the US and Israel (Fridman and Meir, forthcoming), albeit in a different order. This finding supports our hypothesis and adds credence to the idea of universal trends in semantic associations, irrespective of HL and SL.

Responses demonstrating cross-linguistic influence spanned an array of subtypes, with many participants directly borrowing English terms. While it is possible that they were only guessing what the Hebrew term may be for a target word, it is likely that several of these terms are part of their Hebrew–English blended lexicon, lending support toward the idea that HSs speak a new language variety as a result of linguistic contact (Rothman and Treffers-Daller 2014). Through some cross-linguistic and morpho-phonological non-target responses, we can observe a strong grasp of Hebrew morphosyntax among our participants (documented for syntactic knowledge HSs of other languages, as well. See Montrul 2013; Polinsky 2000), with uncertainty mainly focused around specific lexical items, gender assignment, or the construction of conjugation classes. Morphology is known to be a vulnerable grammatical phenomenon among HSs, especially when English is the SL (Polinsky and Scontras 2020), so it appears that retention of Hebrew morphological patterns may be connected to Hebrew language-specific word formation. We also found cases of child-like and antiquated speech patterns, echoing results from Ringblom and Dobrova (2019), Lee and Shin (2008), Wiley (2007), and Johannessen and Laake (2015). Based on these results, we posit that Hebrew HSs follow documented trends for HSs both in and beyond the US, and thus we draw closer toward a generalized model of HL language systems.

We found no significant differences between the average number of code-switching instances and calques in the narratives, although most participants seemed to use more of one strategy than the other, even if both were present. The participants’ code-switching spanned a wide range of forms, from single-lexeme insertion of content relevant to the narrative or brief commentary on the narrative, to inner speech expressing frustration at not remembering a particular word, to a blend of all of the above and beyond. Of the provided examples, single-lexeme content insertions were the least common form of code-switching in the narratives, suggesting that for the majority of HSs, the SL remains the dominant framing language in the mind even when engaged in telling a story fully in the HL. Whether it is used to mentally process a visual or the absence of a desired term, or to complement a mental process in the HL, the SL is always active and accessible for any of the HSs’ needs in processing a task and conveying a message.

We found a significant inverse correlation between performance on the Hebrew MINT and the number of code-switching instances in the narrative, such that individuals who produced more instances of code-switching tended to reach lower Hebrew accuracy and vice versa. By contrast, no correlation was observed between MINT accuracy and calquing frequency. We propose two interpretations for this phenomenon. First, while the average number of code-switching and calquing instances did not differ, the range of code-switching instances was almost twice as wide as that of calquing ones (0–23 vs. 0–13, respectively). Thus, it is possible that this variability played a role in determining a correlation effect. Second, individuals who code-switched invoked English words in an otherwise Hebrew narrative as a strategy to fill Hebrew lexical gaps. Thus, it is not unreasonable that they would have more of these lexical gaps than individuals who consistently used Hebrew terminology, if through an English framework. Calquing patterns indicated that participants were thinking in English and translating directly into Hebrew without recalibrating the linguistic paradigm, but the participants still used Hebrew vocabulary to do so, and thus we do not find a link between increased calquing and lower MINT performance.

When calquing, participants similarly do not fit a single template, with calques expressed in both verbs and prepositions. However, as shown in the example explanations in Section 3.3, the translated expressions are often close or similar in meaning to their grammatical targets, showing a systematic and semantically consistent creative strategy (Rakhilina et al. 2016). Thus, while these expressions would distinguish the HS from a monolingual native speaker of Modern Hebrew, they are not completely incomprehensible. Hebrew-speaking listeners of the narratives would be able to parse them, if at times with greater ease than at others, supporting Polinsky’s (2008) claim that HSs are coherent communicators despite divergence from the baseline. They are able to draw from the language systems with which they are familiar for lexical and expressive support in the face of uncertainty, in order to move forward with their narrative.

5. Limitations and Next Steps

The present work presented an overview of adult Hebrew HSs’ lexical performance in the HL, the number and types of non-target responses they produce in a structured vocabulary task, and their vocabulary breadth and code-switching patterns in a narrative task. By narrowing in on production data, this study presents only a partial picture of Hebrew HSs’ lexical abilities. Future studies might expand on the present work by combining measures of both productive and receptive linguistic abilities, in order to better represent HL lexical development. Additionally, assessing narrative production in the SL as well (as was carried out with the vocabulary task) would facilitate a within-participant comparison of verbosity, as the current design does not contextualize the variability of TTRs and target TTRs.

Overall, much work remains to be carried out on Hebrew HSs to build on the findings presented above. First, we should investigate the effects of background factors such as input quantity and quality over time, motivation to maintain the language, and more, on final performance. Moving away from the lexicon, we could analyze the narratives elicited for this study for other (non-lexical) factors, such as disfluencies and discourse markers. We could delve into morphosyntax: multiple morpho-phonological non-target responses hinted at a strong comprehension of the Hebrew root system and verb class formation, which could be further explored through morphosyntactic tasks. Finally, similar work on lexical development or beyond could be extended to child Hebrew HSs, to assess the extent to which the patterns observed in the current paper are stable over time.

6. Conclusions

As the field of HL studies grows, it becomes increasingly important to expand the range of linguistic contexts and language pairs under consideration, in order to avoid generalized characterizations that may hold in only a few cases. To contribute to this expansion, we have presented an analysis of the lexical abilities of a previously unstudied group of HSs: adult Hebrew HSs living in the United States.

Our participants had significantly higher vocabularies in the SL-English than HL-Hebrew (producing on average almost twice as many terms in the former than in the latter on the MINT vocabulary task), while their non-target response patterns resemble demonstrated patterns for adult HSs of other language pairs. Furthermore, multiple examples of participants’ non-target responses indicate a strong grasp of Hebrew morphosyntactic constructions and grammar rules, which we suggest as a focus for future studies.

When telling a story, participants used a fairly even number of both code-switches and calques. The latter span a variety of forms, from contextual single-word insertions to long strings of inner speech and commentary, all pointing to the idea that the SL is used as the framing language, and the language of thought, which is always active even when producing an HL narrative. Calques were also found across a breadth of lexical categories but were always somewhat close in meaning to contextually accurate forms. Thus, we conclude that Hebrew HSs are competent communicators who can successfully convey complex thoughts while drawing on their dominant language, whether through calques or code-switches, to fill lexical gaps.