Linguistic creativity in heritage speakers

This paper presents and analyzes lexical and syntactic evidence from heritage Russian as spoken by bilinguals dominant in American English. The data come from the Russian Learner Corpus, a new resource of spoken and written materials produced by heritage re-learners and L2 learners of Russian. The paper focuses on lexical deviations from baseline Russian at a single- and multi-word level, which we divide further into transfer-based structures and novel creations , showing that the latter are used by heritage speakers, but generally not freely available to L2 learners. In constructing innovative expressions, heritage speakers follow general principles of compositionality. As a result, such innovative expressions are more semantically transparent than their correlates in the baseline or dominant language. We contend that semantically transparent, compositional patterns are based on structures that are universally available across languages. However, L2 speakers resort to these universal strategies for creating novel phrases much less often than heritage speakers. In their linguistic creativity, heritage speakers’ utterances parallel those of L1 child learners rather than L2 speakers.


Introduction
Heritage speakers are typically defined as "unbalanced bilinguals": those who grew up exposed to a minority language at home, but feel more comfortable with the dominant language of the society in which they live. The category of heritage speakers covers a wide range of abilities, from those who can understand but not speak their heritage language (HL) to those who are quite proficient in their heritage language but limited in some registers associated with literacy (Valdés 2001;Polinsky & Kagan 2007;Rothman 2007). The wide range of proficiency and aptitude levels among heritage speakers raises a number of questions concerning their status as native speakers (see Montrul 2008;2016;Benmamoun et al. 2013;Scontras et al. 2015 for a discussion).
The objective of this paper is to identify and characterize some distinctive features of the lexical-semantic knowledge manifested by heritage speakers. We introduce data from heritage speakers' lexical production and use it to trace patterns; we then establish the uniqueness of those patterns to heritage speakers by comparing them to data coming from L1 and L2 learners. On a broader level, this paper aims to enrich an area of inquiry that has so far received insufficient attention in the growing field of heritage language research.

Data
Our data come from heritage Russian as it is spoken by American English-dominant bilinguals. Lexical issues in Heritage Russian are traditionally discussed in terms of calquing, code-switching, and stylistic violations (see, for example, Zemskaja 2001). corpus contains texts produced by students who started studying Russian as adults. For those students who were enrolled in classes, data were also collected from their written language exams. In both groups, the proficiency level of respondents was no lower than intermediate-mid on ACTFL scales, with the majority performing at the advanced level. Unfortunately, further demographic details were not available for these students.
Oral (elicited) production data were collected from the materials reported by several researchers. Materials on heritage Russian production include "frog stories" (based on the methodology developed by Berman & Slobin 1994;Slobin 2004) and narratives based on short, silent video clips. 4 Data on and discussion of heritage Russian frog stories are presented in papers by Isurin & Ivanova-Sullivan (2008) and Polinsky (2008); several of our illustrative examples below are drawn from these sources.
To focus on the lexical and semantic aspects of heritage Russian, we have chosen in this paper to investigate deviations tagged 'LEX' or 'CONSTR' in RLC. Fragments marked as 'LEX' correspond to improper lexical items; 'CONSTR' indicates mismatches in grammatical or phrasal constructions (collocations), 5 including variations in government patterns, prepositions, lexical restrictions and combinability, etc.
All the expressions tagged as divergent required partial or complete restructuring in order to be acceptable in standard Russian. Such restructuring often involves subtle semantic, syntactic and pragmatic changes. Consider the following example, in which the standard expression is strictly limited to an idiomatic unit and doesn't permit any variations. Despite being an idiom, this unit has a clear passive-like syntactic structure with a specially marked verb form nazyvajutsja (call.prs.3sg) and an inanimate subject (professija 'profession'). The animate agent in this case is left unexpressed: (1) a. The analysis of divergent expressions presented below allows us to compare lexical strategies used by heritage speakers and speakers of baseline Russian. We present data from both oral production (frog story and video clip narrative; only heritage speakers) and written production (heritage speakers and L2 learners).

General approach
Our focus here is on phrasal structures, which usually lie beyond the scope of research. To illustrate a standard analysis of heritage Russian grammar errors, consider the following example from Isurin & Ivanova-Sullivan (2008: 83), which shows a heritage Russian phrase that is ungrammatical in the baseline language: (2) a. Heritage Russian idjot v morju go.prs.3sg in sea.dat (invalid case) 'goes into the sea' b. Baseline Russian zaxodit v vodu go.prs.3sg in water.acc 'steps into the water' Isurin & Ivanova-Sullivan (2008) interpret the deviation in this phrase as a failure to use the proper preposition. Indeed, if the speaker had used the preposition k 'toward', the sentence would have been grammatically correct. Our approach parts ways with this formal analysis in considering the whole phrasal structure to have been misused. Consider: in (2), the speaker describes a simple spatial scene: a person is standing not far from the sea, then moves towards the sea and steps into the water. In this case, the construction idti k 'go to' does not seem to be acceptable in standard Russian because its semantics implies a spatial gap between the subject and the landmark, without contact between them. 6 If the subject steps into the water, another construction, zaxodit' v vodu (lit. 'enter the water'), must be used.
Violations in the use of phrasal structure are particularly noteworthy within a broader context of language interference and calquing. Heritage speakers are heavily influenced by the dominant language, so it is reasonable to expect them to use calques: word-for-word translations from the dominant language. Calquing indeed occurs in heritage language; since Benson (1960), numerous accounts have provided evidence of direct translations from English into Heritage Russian (see Mikhaylova 2006;Polinsky 2008;Dubinina & Polinsky 2013). For example 7 : (3) a. Heritage Russian (Laleko 2010: 28) princessa Below, we show that calquing does not fully account for all the cases of lexical deviations we address; furthermore, this phenomenon cannot explain the mechanisms underlying the emergence of lexical and syntactic deviations. Indeed, overall, the RLC data suggest that direct borrowing from the dominant language is relevant in only a limited number of instances. It seems that, when heritage speakers fail to find a proper Russian phrase to express their semantic intention, rather than turning to their dominant language, they build phrases of their own.
In example (2), we saw that, when attempting to verbalize a simple spatial scheme, the heritage speaker simply combined the semantics of the basic motion verb idti 'to go', the preposition v 'into', and the noun more 'sea' (used in the wrong case form). The resulting construction is awkward, if not wrong, in both standard Russian and Standard English when the subject of the "going" is a person, as it is in (2). This awkwardness arises because both languages have a restriction on the lexical meanings of the relevant nouns. The Russian idti/ vyxodit' v more 'go into the sea' is appropriate only if the "goer" is a vessel: (5) The ship went into the sea.

(6) Baseline Russian
Korabl' vyšel v more. ship [nom] go.pst[sg.m] in sea.acc 'The ship went into the sea.' To gauge the frequency with which calques are produced by non-native speakers of Russian, we examined RLC data (as of 2014) for heritage and L2 learners. A total of 473 sentences produced by L2 learners and 624 sentences produced by heritage speakers were examined; the data, with heritage language/L2 identifiers removed, were independently analyzed by three raters and subsequently compared. As the Figure 1 below shows, out of 310 deviations from standard lexical distribution produced by heritage speakers, 25% were calques. Meanwhile, 64% of the 285 deviations produced by L2 learners of Russian were calques.
In the remainder of this paper, we will investigate in detail the lexical distribution in Russian language produced by heritage speakers, setting aside L2 word-combination strategies that do not rely on dominant-language calques (an issue that warrants further investigation). Among heritage speakers, cases of lexical deviation that are not due to calquing can be divided into two types: a) structures that lack calques altogether and b) structures that we will describe as semi-calques. In Section 2, we discuss the linguistic mechanisms that heritage speakers use to derive new expressions, thereby avoiding calquing; Section 3 presents the hybrid expressions that we refer to as semi-calques; Section 4 discusses the actual calques that our research has uncovered.
To anticipate the discussion below, we propose that heritage speakers prefer compositional structures, 8 avoid non-compositional ones, and tend to rely heavily on conceptual structures when producing lexical content, thereby bypassing language-specific requirements. In that sense, heritage speakers show similarities with young L1 learners, whose lexical production we address in Section 5. Section 6 presents our conclusions and outlines directions for further research.

Absence of calques
In order to understand the creative behavior of heritage speakers, let us first revisit the familiar distinctions among conceptual structure, semantic structure, and grammatical structure. According to Jackendoff, "[c]onceptual structure is not a part of language per se -it is a part of thought. It is the locus for the understanding of linguistic utterances in context, incorporating pragmatic considerations and 'world knowledge'; it is cognitive structure in terms of which reasoning and planning take place" (Jackendoff 2002: 123). Conceptual structure includes presumably universal categories such as Event, State, Object, Path, and Property. These categories underlie the lexical-conceptual structures of the lexical items that compose phrases and clauses. They also play a role in allowing speakers to build semantic categories by combining functions and arguments. It is only in the final stage of the process, when semantic categories are put into grammatical structures, that language-specific properties play a primary role.
One of the possibilities we consider is that heritage speakers rely more heavily on conceptual structure than native speakers, often foregoing the requirements placed on their language by semantic and grammatical structures (cf. Polinsky 2006 for similar observations). In relying on conceptual structure par excellence, heritage speakers may create new formations that do not bear any clear similarity to specific phrases in either of the languages they have access to. A particular subcase of this phenomenon is heritage speakers' creation of compositional expressions where one would otherwise expect calques from the dominant language. The fact that heritage speech contains fewer calques than L2 speech indicates that dominant-language transfer is less strong for heritage learners than for L2 learners; this finding, in turn, suggests that heritage speakers possess linguistic intuition for both languages. The absence of calques is not categorical, nor does it imply that heritage speakers are consciously making the choice to avoid this form of dominant-language transfer. Rather, we contend, this tendency simply indicates heritage speakers' general dependence on conceptual structure. The examples we present below of the heritage non-calquing strategy will be further divided into pattern-based structures (built on conceptual primitives) and decompositional structures (based on compositionally simple "building blocks").

Conceptual primitives as building blocks: Pattern-based structures
The notion conceptual primitive captures the intuition that certain elements are fundamentally irreducible from a conceptual perspective. Conceptual primitives comprise a cross-linguistically universal set of compositional patterns (including patterns of motion, giving, destruction, etc.), which are traditionally described in linguistic theory in formal or cognitive terms (cf. Jackendoff 2002 for the former, and Langacker 2000 for the latter). When mapped onto the lexical and syntactic structures of a particular language, these patterns may be expressed in various ways, as compositional grammar imposes few limitations. In this section, we argue that this mapping has its own specific nature in heritage speakers' production. We pursue this argument by addressing those instances of heritage output that display simple conceptual structures and universal patterns. We also show how these patterns deviate from those found in the standard language. Structures of this type will be referred to as "pattern-based." To begin with a simple instance of pattern-based structures, consider the example below, repeated from (2) Example (7a) illustrates a basic spatial pattern of motion, which, by default, implies TRAJECTOR, SOURCE, GOAL and PATH; consider the more fleshed-out instance of this pattern in (8), where train is TRAJECTOR, Cambridge is SOURCE, London is GOAL and fields is PATH.
(8) The train goes from Cambridge to London across the fields.
In most cases, this pattern is reduced to TR + G, due to the Goal-bias effect, which highlights GOAL as the more salient thematic role (Ikegami 1987;Stefanowitsch & Rohde 2004), for instance as in (9): (9) The train goes to London.
Assuming this basic pattern of motion, we can posit that, in (7a), the speaker takes the direct meanings of words and combines them in a straightforward compositional way to produce a construction that fits the basic TR + G scheme, illustrated in common instances such as idti v školu 'go to school', idti v teatr 'go to the theatre', etc. Note that "open space" goals also conform to this structure; cf. idti v pol'e 'go to the field', idti v l'es 'go to the woods', idti v gory 'go into the mountains', etc.
The frequency of the construction idti v + Location 'go to a location' is around 9,000 in the RNC; the frequencies of VESSEL idti v more and PERSON idti v vodu are 46 and 48, respectively. The heritage speaker clearly follows the more frequent pattern at the same time, ignoring the non-compositional restrictions on the complement of the preposition v in the PP that combines with idti.
To reiterate, in example (7a) above, the word voda 'water' is a much better candidate for a "goal" than the word more 'sea' for the typical native speaker (note that the same generalization also holds true for English, where in the corresponding construction water is definitely preferable to sea when denoting location for non-vessel subjects). The heritage speaker, however, looks beyond these limitations to produce a semantically transparent phrase.
As another example of non-spatial pattern-based structures, consider (10), where the intended meaning is 'giving money to a beneficiary in exchange for something valuable'. The transparent pattern that accounts for (10) entails SUBJECT and BENEFICIARY and optionally expresses MEANS and REASON; consider (11), where President is SUBJECT, general is BENEFICIARY, and excellent service is REASON.
(10) Heritage Russian idei o pooščrenii kul'tury 9 idea.nom.pl about rewarding.obl culture.gen 'ideas concerning the stimulation/encouragement of culture' 10 910 (11) The President awarded his general with a medal for excellent service.
In Russian, however, the word pooščrenije (lit. 'stimulation/encouragement') is used in this pattern only when the BENEFICIARY is expressed by an animate noun: (12) Baseline Russian pooščrenije rabotnikov [BEN] rewarding.nom worker.gen.pl 'stimulation/encouragement of workers' When, instead, the beneficiary is an abstract noun or an organization (something cognitively less primitive), pooščrenije in Standard Russian is more likely to be replaced by the word podderžka (lit. 'maintenance'): podderžka proekta / predprijatija / nauki / kul'tury / sporta etc. 'maintenance of a project / an enterprise / science / culture / sport'). In the RNC, pooščrenije kul'tury is not attested at all, while a direct Google search returns less than half as many matches for pooščrenije kul'tury than for podderžka kul'tury. This number also includes repetitions and contexts that differ in meaning.
In the next example, the divergent pattern has to do with the marking of the by-phrase as ablative:  (13) is to express a direct relation between entities. One entity (strany tret'ego mira 'Third World countries') experiences negative influence (èkspluatatsija 'exploitation') exerted by another entity (lica s vysokim VVP, lit. 'persons with high GDP'). The structure of (13) is determined in large part by the heritage speaker's decision to use a nominalized construction. Had the negative influence been expressed by a verb instead, the syntactically simple transitive structure 'A exploits B' would have sufficed. However, this simple structure would have caused problems for the sentence as a whole: (13) contains two predicates, 'exploit' and 'become a reason for', with the first serving as an argument for the second. REASONs are often conceptualized as entities and verbalized as nouns; the speaker therefore nominalizes and partially passivizes the verb èkspluatirovat' 'exploit'. Now that a passive construction has been introduced, the idea of "directedness" becomes more salient, with the associated semantics [SOURCE + directed relation (negative influence) + GOAL]. Seeking to adhere to these semantics, the heritage speaker selects the preposition ot 'from', which is a standard marker of SOURCE in Russian. 11 Thus, the heritage speaker arrives at a semantically well-specified but grammatically invalid marker, ignoring the restrictions imposed by both English and Russian in order to produce a semantically transparent pattern-based collocation.
In (15), the speaker's intention is to describe the means that the author used to write the play. For this reason, s/he marks the noun monologue as an instrument and puts it in the instrumental case. In standard Russian (as in standard English), however, this context doesn't permit an instrumental pattern (AGENT + INSTRUMENT + OBJECT); rather, it requires a separate lexical expression, as illustrated in (16)  Finally, (18a) is an attempt to express a CONTAINER + OBJECT pattern. In order to convey the idea of placing one entity into another, which seems logical for this sentence, the speaker uses the preposition v 'in', but fails to follow contextual restrictions that prohibit this lexicalization pattern for entities like people and society. To conclude this section, we have presented several instances of structures based on simple, widespread patterns. The knowledge and use of such patterns is certainly affected by their frequency but it is important to keep in mind that frequency alone cannot be the determining factor in the selection of patterns.

Decompositional structures
Decomposition is an explanatory strategy that speakers can use to unpack the meaning of an idiomatic structure; under this strategy, each element of the structure's semantics is interpreted in as detailed a manner as possible. This strategy parallels the process that lexicographers go through when defining the meanings of words in a dictionary. In decompositional structures, speakers tend to avoid translating expressions from their dominant language word-for-word. For particularly complex constructions, this means that these expressions must first be disentangled before being translated. When heritage speakers attempt to deconstruct complex constructions, they often resort to strict compositionality, breaking the concept into simpler semantic items, each one of which is lexicalized by a separate word. This strategy can lead to problems if a given language's way of expressing a complex concept involves non-compositional elements and does not correspond to a clear universal pattern.
Idioms and set expressions present an obvious instance of non-compositionality, and speakers who do not have access to those non-compositional expressions have to invent replacements for those. The corpus data we have at our disposal suggest that heritage speakers reshape non-compositional expressions into compositional equivalents.
This strategy of decomposing a complex meaning into simpler elements and avoiding non-compositionality is illustrated in the following example: In (19), the speaker provides a literal interpretation of a concept that is expressed by an idiomatic structure in the baseline, učit'sja na ošibkax 'learn from one's mistakes (lit.: learn on mistakes)'.
The principal reason for the non-transparency of the baseline expression, which motivates the speaker to search for a clearer way to communicate the meaning, seems to be the interpretation of the preposition na 'on'. The semantics of this preposition within the baseline expression is quite vague. Furthermore, the construction itself is not only highly idiomatic but also very rare: the expression učit'sja na ošibkax has only 22 occurrences in the RNC, which is low compared, for instance, to such idiomatic expressions as brat' primer 'follow one's example' (227 occurrences in the RNC) or brat' v svoi ruki 'take into one's own hands'/'control' (140 occurrences in the RNC). Since this idiom is both structurally opaque and uncommon, the speaker breaks down the complex meaning into a set of simple elements and comes up with a new, strictly compositional, expression to convey the necessary concept.
Heritage speakers also apply the decompositional strategy to frequent phrasal units when their meanings are complex and their structures differ in the heritage and dominant languages. In (20), the speaker may know that there should be a non-compositional way to express the idea she wants -which prevents her from calquing the English constructionbut fails to retrieve the appropriate Russian phrase, however common: proper.acc 'Many countries did the same.' As in the previous example, (20) decomposes the semantics of <to act> in the same way. The correct Russian phrase in this context, sdelat' to že samoje, captures the symmetrical-event concept with a holistic verbal expression, lit.: 'to do + that (+ intensifying particle) + most'. Since this situation affords no opportunity for heritage speakers to guess the right expression, either by appealing to the dominant language or by seeking some standard semantically transparent pattern, decomposition surfaces as a last resort to convey the desired meaning.
Heritage speakers show attempts to make the semantics of structures they use more precise in order to avoid ambiguity. Sometimes this desire to be extra clear leads to a complete rephrasing of an idiom, as we witnessed earlier. In other cases, the speaker will slightly 'tune' an expression to eliminate all hints of idiomaticity, as in the example below:  (22) is perfectly fine when interpreted through the lens of common sense: clearly, no one can literally start a new life, but people often change their lifestyle. Strange as it may seem, this idea is expressed with an identical idiom in both standard Russian and English: načat' novuju žizn' (start a new life). The heritage speaker, however, makes the effort to decompose this construction in order to clarify its meaning.
Our data thus show that heritage speakers readily create new units; however, in doing so, they generally avoid complexity and non-compositionality.

Semi-calques
The next major strategy adopted by heritage speakers involves "semi-calques," which we define as newly created expressions that rely simultaneously on the two linguistic systems available to a bilingual speaker. The following example serves as an illustration: Example (23) contains two deviations from standard Russian, only one of which will be of relevance here. First, the Russian adjective sčastlivyj 'happy' cannot licitly combine with nouns like rasskaz 'story'. A Russian speaker would use vesjolyj 'cheery' or razvlekatel'nyj 'entertaining' in this position. Sčastlivyj rasskaz is a clear calque from the English happy story.
The second deviation, the one that we will focus on here, is the non-standard phrasal unit porovnu A kak i B (lit.: 'equally A as B'). Structurally, this phrase is very close to the corresponding English phrase equally sad and happy. The English construction expresses the intensity of two qualities as applied to one and the same object (story). That's the main import of equally in this phrase.
Russian does not have a direct counterpart of equally that could be used in this context. The adverb odnovremenno 'simultaneously/at the same time' doesn't imply the "intensity" comparison that the speaker obviously wants to express. The speaker could use ravno 'in the same way, equally' in this context, but this adverb is rare and somewhat obsolete in this function. (Only 29 instances of the construction ravno A & В occur in the RNC, and they are limited to the data from the 18 th -and 19 th -century language). 12 Russian does, however, have a special construction used for focusing on the juxtaposition of two qualities: the highly idiomatic two-part construction stol' zhe A skol' i B (lit. 'as much A as B'): In (24), two different properties are set in opposition (nevežestvennoje 'ignorant' implies insufficient knowledge, while vysokomernoje 'arrogant' implies excessive confidence in one's knowledge). The English A and B equally and the Russian stol' že A skol' B constructions are semantically close, but the Russian construction is not semantically transparent and its components are quite rare. 13 The semantic transparency and dominant frequency of equally renders this unit salient enough to form the first part of the heritage speaker's novel construction in (23). The second element is the Russian phrase kak i (lit. 'as and'), which decomposes to the classical comparative marker kak 'as' and the connective i. The resulting new construction, porovnu A kak i B 'equally A as and B', is thus glued together compositionally from frequent and salient fragments of the relevant English and Russian constructions.
Another illustration of a new constructions composed from parts of both languages is given in (25). This example is taken from a frog-story production experiment (Isurin & Ivanova-Sullivan 2008: 89). In this particular fragment, the dog is lying down, and the frog is sitting nearby with its leg on the dog's back. Then the frog moves its leg: Although the situation described is visually quite simple, the way it is articulated is noncompositional in both English and Russian. The English construction used in this context is built around the verb to take, which can be followed by different adverbial modifiers or particles depending on the context. Ordinary possessive contexts (taking an object into one's hands) require from/off (cf: he took a book from the shelf / he took a picture off the wall, etc.), while the motion of a body part is normally described with the adverbial modifier away.
Similar to English take, the Russian verb vzjat' occurs in canonical possessive constructions (taking something from a person), cf. (26), as well as in locative (ablative) constructions (for example, taking something from the surface, as shown in (27) The donkey in (28) plays the same grammatical role as the table in the previous example, but it cannot be regarded as a standard surface. Hence, the situation described in (28) retains its strong locative semantics; conceptually, taking a sack off a donkey is understood to be more like taking off clothes (worn by an animate participant) than like taking an artefact from a table.
Yet another restriction concerns the OBJECT role in (25). Both vzjat' 'take' and snjat' 'take off' are incompatible with a body part in the OBJECT position. Both verbs specifically denote mechanical displacement of an item by means of hands. The natural movement of a body part, if controlled by the body's owner, is not mechanical; it is caused by the psycho-cognitive impulse of an animate agent and needs no instrument in order to be performed. This type of motion is lexically distinguished in Russian and is encoded by the verb ubrat': To express the meaning encoded in (25), the heritage speaker minimizes the linguistic options by making the construction as transparent as possible. The immediate counterpart of the English verb take 'vzjat', the most frequent and salient verb, is combined with the preposition s, which translates the whole variety of English items -from, off, with, and away. The resulting construction is a compositional semi-calque.
Example (30)  This piece is a hybrid of the respective English and Russian non-compositional constructions: in contrast to and po kontrastu s ètim (lit. 'along contrast with'). The heritage speaker uses the first preposition po 'along' from the proper Russian construction and goes on to borrow the second preposition k 'to' from English. This choice, like those discussed above, is based on several conceptual considerations. First, the need to compare two ideas in terms of their similarity or difference is resolved in baseline Russian with the help of the preposition po 'along', whereas English employs a variety of elements: What about the choice of the second preposition, k 'to', which the heritage speaker borrows from English in (30)? Although the choice of this preposition ignores the Russian system, it is not arbitrary either. While, as mentioned, the speaker is comparing two entities with respect to their dissimilarity, this is not the full meaning of the construction. The semantics of (30) also implies that one of the entities being compared is more salient than the other, and the relation is thus one of unilateral directedness. Seen in this context, the preference for k over any other lexical element may be attributed to the strong association of k with directional semantics. Thus, again, we see that the heritage speaker does not hybridize Russian and English elements together arbitrarily, but deliberately draws on the simplest items from each construction, in terms of both semantics and lexical combinability, to form a semi-calque.

Calques
We have so far concentrated on strategies other than direct dominant-language calques that heritage speakers use to create novel expressions. However, heritage speakers are by no means immune to this more direct form of language interference. Furthermore, calquing has received significant attention in L2 acquisition studies (Odlin 1989;Ellis 1997), and thus offers an important opportunity for direct comparison between L2 and heritage speakers. We explore that comparison in this section.
Unlike L2 learners, heritage learners rely heavily on their intuitions when producing Russian. 14 Thus, when heritage speakers do create calques, they typically import dominantlanguage constructions that happen to be associated with similar licit and readily available structures in Russian. This tendency was shown in example (3a), repeated below: The deviation from baseline Russian in (35a) is quite subtle. It can be explained by the fact that the Russian language distinguishes both between dynamic and static situations and between temporary and permanent (again, static) situations. Blizko k (lit. 'close to') is attested in the baseline language, but it typically surfaces either with verbs of movement or with static verbs that denote temporary locations. It is not normally followed by city names; cf. shore.gen 'The most beautiful and unusual church is on the small island not far from the shore.' In the RNC, this construction occurs more than 1200 times with town/city names.
How do heritage speakers compare to L2 speakers with respect to calquing? As was mentioned above, corpus data show that L2 learners of Russian use significantly more English calques than heritage speakers do. Furthermore, the calques they produce are markedly distinct from those that we find in heritage speech. The driving force behind L2 calques is the copying of form, without much consideration for concomitant semantics. Consider the following example, which shows an L2 calque for the construction for two hours that clearly comes from the dominant English: (39) a. L2 Russian dlja dva časa for two hour.pl 'for two hours' (lit. for to two hours) b. Baseline Russian Ø dva časa two hour.pl 'for two hours' The preposition dlja 'for', unlike its English equivalent, is restricted in Russian to mark the addressee, recipient, or beneficiary, and has no temporal interpretation. Because dlja has no meaning in the temporal context, example (39a) is ungrammatical in the baseline.
Similarly strong semantic divergence between English and Russian can be seen in the calque in (40), also produced by an L2 learner: In place of this unfamiliar construction, the L2 speaker produces an (ungrammatical) word-by-word translation of the standard English construction called/named X. The situation is further confused by an error in the Russian verb form: the speaker uses the active past form of the causative verb (nazyval, lit. 'he called') instead of a passive participle (nazvannyj, lit. '(one that) was called'). However, neither verb form produces a phrase that is interpretable in baseline Russian: From these examples, we can see that the L2 calquing method is strongly motivated by form rather than semantics, and therefore tends to be difficult for native speakers to make sense of. By contrast, the heritage calquing strategy, illustrated in (34) and (35a), produces constructions that are interpretable, if not entirely natural, in baseline Russian. L2 learners, on the other hand, do not refer to the semantics of Russian when they calque constructions from English, but rather borrow ready lexical units that may not be understandable to standard Russian speakers. In doing so, they demonstrate less linguistic creativity.
To further illustrate our point, we present below two short texts produced by an L2 speaker (Text 1) and a heritage speaker of the same proficiency level (Text 2); both speakers took the same language class. The author of Text 1 relies more heavily on her dominant language, producing 10 calques from English. The author of the second text cannot avoid calques either, but produces only 4 calques. Conversely, the L2 speaker produces notably fewer non-calque deviations: one in Text 1 versus seven in Text 2. Our preliminary analysis of these texts focuses on phrasal calques and non-calques; divergent structures are marked in bold and each is assigned the status of calque or non-calque. The commentaries are presented in Tables 1 and 2 after the texts. 16
'As I understand these words, success is achieving a goal. Success cannot be defined without the notion of what it is not, that is without the notion of «failure». Classical composers of Soviet times also resorted to that principle. For example, Vano Muradeli, a Soviet composer and a contemporary of Shostakovich, decided to write disharmonious and dissonant music for his opera "The Great Friendship" for the 30th Anniversary of the Great October Revolution. On October 11th, 1948, the authors of the Central Committee of Bolshevik Party Decree About the opera "The Great Friendship" presented this opera not only as a disharmonious, dissonant musical piece, but also as a "chaotic" piece. They thought the opera was chaotic because clear and easy-to-remember tunes which a common listener could easily sing after the concert were absent in the music. According to the Decree authors: the opera music is expressionless and impoverished. It doesn't contain any easy-to-remember melody or air. Straightforward elements of this kind would be considered unacceptable in Soviet music halls and theatres but Shostakovich probably wanted to present these unpleasant elements to the Soviet audience to show how bad and intolerable life was in pre-revolutionary Russia and how bad people needed revolution which would relieve people from the oppressive bourgeois gloom of the Tsar government. In that case the complicated, hard to understand and to remember opera music was still realistic because the musical dissonance perfectly expresses the pathetics and bitterness (Pathos? How should I translate 'pathos'?). If all this was really Shostakovich's intention, the audience, critics and the Soviet government understood "Lady Macbeth of Mtsensk" awfully wrong.'
'I think that people in Russia hit into entertaining programs. Many people got tired of action movies and movies with human violence. They want movies without scenes, indecencies and blows after the characters get up as if they hadn't lost their consciousness. If we depart from this, this exactly will contribute to the development of the civil society in the country. Сensorship should exist -you just can't broadcast anything you want. Most of my friends who are political scientists believe that his movie does not deal with the ambassador's murder or any other type of violence, this is a particular case and the accusations towards the director are only for the sake of appearance. I agree with this point of view. Nevertheless, there are no laws that limit the freedom of speech about what is politically incorrect. For example, in America you can cry about how you hate the President burning the American flag. At the same time there are limitations on the freedom of speech at the workplaces -for example the Equal Opportunity Act. This does not mean that the State infringes on our freedom of speech. The purpose of this law is only of the wish to protect the other people at the workplace from slander. Also we are cursing the government and waiting for fantastic pension. Wake us up in the morning and we immediately remember "On seashore far a green oak towers, and to it with a gold chain bound. . ." We are proud of Tolstoy, Dostoevsky and the people who remember more than what Raskolnikov did and who Natasha Rostova is. And despite all this, we don't take the cross off and we never forget to sit before the long journey.'

Heritage language speakers and L1 learners
We have suggested that heritage speakers are highly creative in filling their lexical gaps, and that they deploy resources made available to them by their heritage language when doing so. Another group of highly creative non-standard speakers are small children acquiring their first language. It has already been shown that children are much more creative than adults, readily ignoring certain linguistic restrictions and overgeneralizing patterns (Ceitlin 2009 The idea of "ripening" in this example is generalized by Ženya to cover not only plants but also animals. On his reinterpretation, animals become ready to be utilized (for fur coats) in the same way that fruits and vegetables become ready to be utilized (for food and drinks).
Young children frequently overuse templates in this manner, generalizing their meanings. In the context of Russian, this effect is illustrated most often with reference to derivational schemas: (45) a. Russian child language Naša kurica о-cypljat-i-l-a-s'! our hen.nom prefix-chick-pst-3sg.f-refl Lit. 'Our hen has chicked!' b. Baseline Russian Naša kurica snesla jaico. our hen.nom lay.pst.sg.f egg.acc 'Our hen has laid an egg.' While conceptual patterns involving SOURCE are not restricted, languages impose specific structural constraints on the expression of this concept. In particular, Russian limits the range of verbs that can be used with a SOURCE that is a body part (47b). Only the verb snjat' 'take off' may be used in this case, accompanied by the preposition s, which duplicates the verb prefix. In (48a), where the animate SOURCE is also the POSSESSOR, the preposition ot is used instead of the default preposition u.
These examples suggest that the child has acquired the general ablative pattern with its most frequent and cognitively salient marker ot 'from' (the frequency of ot 'from' in ablative contexts, according to RNC, is seven times higher than the frequency of u in the same environment). However, she has not yet mastered the relevant lexical restrictions, based on subtle semantic differences that dictate the use of this pattern. This limitation compels her to follow the principle of compositionality when constructing a novel phrase. Despite the fact that heritage speakers have two languages to resort to, they frequently display the same mechanism (see section 2.1 above).
Additionally, heritage speakers and young L1 learners alike produce comparable decompositional structures (see Section 2.2 for the discussion of heritage speakers' decompositional expressions). Consider the following example from Gvozdev (1961:  The child is clearly trying to articulate the idiomatic construction odin i tot že 'one and the same' (lit. 'one' + 'and' + 'that' + intensifying particle), but fails to locate the structure correctly in his lexicon and elects to give a more transparent explanation instead: one + intensifying particle že. Recall that heritage speakers adopt the same compositional strategy when they encounter problems with idiomaticity.
Despite the parallels shown above, however, the two groups are not entirely similar. For instance, some of the deviations that characterize heritage speech may never be found in the speech of children, since heritage speakers can rely on the additional resources of their dominant language to produce formal structures (including calques and semi-calques). Less evident, perhaps, are lexical violations that can be committed only by children. Consider the following example: The answer given in (50) is unlikely to be produced by an adult speaker of any natural language, including a heritage speaker. The body is ascribed greater salience than any piece of clothing; hence, slapping the/my pants is a less natural utterance than slapping the/ my knees. Heritage speakers acquire this principle along with their dominant language and import it into the heritage language thereafter, whereas children continue to produce this sort of error even after they learn simple constructions.

Conclusions
Linguistic creativity is often associated with literature, not language or linguistics. In the present work, we import the concept of linguistic creativity into the study of heritage languages, where we characterize the phenomenon as involving two main facets: the violation of co-occurrence constraints in non-compositional phrasal units, and the creation of innovative lexical material, including multi-word expressions. Based on this conception of linguistic creativity, we have examined lexical distribution in the production of heritage speakers, comparing our findings to similar data from L1 and L2 learners.
In this paper, we showed that heritage speakers demonstrate greater linguistic creativity than proficiency-matched L2 learners. Particularly, heritage speakers create new phrases using structures that are absent in both their languages, relying on pattern-based behavior and decomposition of meaningful elements. These strategies result in novel phrasal formations, which we associate with calque avoidance. We also observed partial calquing (semi-calquing) and selectively motivated calquing, along with direct borrowings from the dominant language.
In all their novel productions, heritage speakers abide by the basic principle of compositionality: they decompose meanings that would otherwise be idiomatic and deploy resources from both languages when expressing those meanings. Even when borrowing from their dominant language, heritage speakers follow the principles of semantic consistency and transparency. On the contrary, L2 learners most often rely on straightforward calquing.
We also compared the linguistic strategies of heritage speakers to those adopted by monolingual first language learners. Strategies that heritage speakers share with child L1 learners include the use of pattern-based constructions, the use of meaning-based decomposition, and the conflation of fragments taken from different standard constructions into a single novel expression.
Although the non-calquing and semi-calquing strategies used by L2 learners require further investigation, it is clear that the use of set expressions and the lower reliance on calquing fundamentally distinguishes heritage speakers from L2 learners, underscoring the differences between the two groups. However, it is also too simplistic to posit that heritage speech is frozen at an early acquisition stage. Although heritage speakers and young L1 learners (between ages 2 and 4) deploy similar strategies of strong compositionality, clear differences also exist between the two groups. Certain types of non-compositional constructions are more readily acquired by children because they are not exposed to the interference from another, dominant, language.
Looking ahead, it is our hope that this study will set a precedent for future investigations into the patterns of linguistic creativity in L1, heritage language, and L2. A more detailed discussion of non-calquing and semi-calquing strategies in L2 production is due in the near future. New structures produced by heritage and L2 speakers show some inevitable dominant-language interference, but they also reveal some general syntactic and semantic patterns that should be investigated beyond Russian-English bilinguals.
Abbreviations hl = heritage language, l1 = first language, l2 = second language, rnc = Russian National Corpus, rlc = Russian Learner Corpus, actfl = American Council on the Teaching of Foreign Languages, ben = beneficiary