If that’s what she said, then that’s what she said: a usage-based, constructional analysis of pleonastic conditionals in English

This paper investigates so-called pleonastic conditionals (PCs) in Present Day English, e.g. If you have to, you have to or When duty calls, duty calls. PCs are rhetorical statements in which the antecedent and the consequent are formally and semantically identical. The construction is characterized by stylistic reduplication and semantic redundancy. It is exactly this tautological nature which has a considerable effect on the discourse-pragmatic functions of the construction. The paper analyzes the construction’s frequency, form and functions by looking at data from three BYU corpora: the Movie Corpus (Davies, 2019), the TV Corpus (Davies, 2019), the Corpus of American Soap Operas (Davies, 2011). 2151 examples of if- and when-PCs are analyzed qualitatively and quantitatively. It is argued that PCs are functionally too different to be classified as default conditionals in the strict sense simply because they do not express a condition or inference. PCs are constructions used to express acceptance, indifference or certainty of the proposition. On top of that, they also code prototypicality which is a function completely overlooked in the literature so far. A Logistic Regression Model is fitted to predict the choice of the connector if or when. The statistical analysis confirms that the two types have different constructional profiles. With regards to theoretical modeling, the paper analyzes PCs from a usage-based, cognitive, construction grammar point of view and sketches the constructions’ form-meaning parings as well as their horizontal and vertical connections in the constructicon.


Introduction
"When you're hot you're hot and when you're not you 're not" (Jerry Reed) This paper investigates so-called pleonastic conditionals (henceforth also PCs) qualitatively and quantitatively in Present Day English.In general, a pleonasm -from Ancient Greek pleonasmós ('to be in excess') -is defined as a rhetorical figure in which the speaker uses "more words than those necessary to denote mere sense". 1  In other words, a pleonasm describes the stacking of words without adding further semantic content.In that sense it is a type of tautology used for emphasis.A famous example would be free gift or actual fact.
Pleonastic conditionals are statements in which the antecedent and the consequent are formally and semantically identical.As exemplified in (1)-( 4), the consequent is an exact repetition of the antecedent (Declerck & Reed, 2001: 359).
(1) Sweetheart, I understand.If you have to go, you have to go (SOAP, 2003, Pass) (2) If I donʼt remember, then I donʼt remember, and thatʼs fine (SOAP, 2001, AMC) (3) You see that door.When itʼs shut, itʼs shut.I can pull on it as hard as I can (TV, 2016, GMW) (4) Okay, Iʼll see you when Iʼll see you (SOAP, 2007, GL) With regards to formal shape, they are similar to conditional clauses by having a bi-clausal structure with a so-called 'conditional connector' (if or when) and an antecedent which is followed by a consequent.In the "Constructional Sketch" section, various constructional templates will be postulated.For instance: Note that X has to be a clausal unit (with a few exceptions) and represents a proposition that can either have a positive (1) or negative polarity value (2).In these templates then can be inserted optionally (2).Using the superscripts highlights the identical nature of the elements.As will be discussed more later, there is also a reverse pattern [X cl i if/when X cl i ] Cx in which the conditional connector is postponed to the second clause (4).
PCs are characterized by semantic redundancy and stylistic reduplication.The truth or interpretation of the consequent is clearly not dependent on the evaluation of the antecedent because they are the same.This is very different from default conditional statements (5 & 6), where we most often find a causal link, condition or inference between two propositions.if X i cl (then) X i cl Cx whenX i cl (then) X i cl Cx 1 https:// www.merri am-webst er.com/ dicti onary/ pleon asm.
1 3 If that's what she said, then that's what she said: a usage-based,… (5) If it rains tomorrow, the lawn will get wet.
(6) If the light is blinking, the ferry is operating.
If one compares default conditionals with PCs there is similarity in form but not so much in function.This is why PCs have been categorized as pseudo-conditionals, or non-conditional conditionals (Declerck & Reed, 2001: 359).In this paper, ultimately it will be argued that PCs should not be classified as conditionals in the strict sense (see "Pleonastic Conditionals" section for details).
In the examples above, the consequent does not contribute any new propositional meaning.This begs the question why, for instance, it is not enough to state I am allergic.What is the added value for a speaker opting for the PC in (7b) instead of the simple clause in (7a)?(7a) I am allergic.(7b) If I am allergic, I am allergic.
The reduplication seems to be a discourse-pragmatic matter.A pleonastic conditional flouts the maxim of quantity (Grice, 1975: 46) and it is exactly this tautological nature which has considerable effects on the discourse-pragmatic meaning and communicative function of the construction.
Unfortunately, the discourse-pragmatic functions of PCs have not been discussed sufficiently in the current literature.Declerck and Reed (2001) are one of the few authors which mention PCs briefly.They consider the basic function of a PC to be "conveying the message that the speaker accepts what she considers as an inescapable fact" (Declerck & Reed, 2001: 360).Moreover, some instances of PCs signal that "the hearer should not doubt the inevitability of Q following from P" and/or "that there are no alternative conclusions to be drawn from P" (Declerck & Reed, 2001: 360).In a nutshell, according to Declerck and Reed, a PC expresses some kind of acceptance of the inevitable.
While this offers a convenient starting point for the analysis of the examples, it is by no means a fully comprehensive analysis of the constructions' functions.As a matter of fact, the constructions and their distribution have not been investigated in a quantitative and corpus-based manner at all.This lack of inquiry leads to the following research questions regarding the constructions' frequency, form and function: RQ 1 Which syntactic variant is most often used to form PCs? RQ 1.1 Which conditional connecting device(s) are recruited to formulate a PC? RQ 1.2 Can preferences with regard to polarity value be observed?RQ 2 Which discourse-pragmatic functions can be identified?RQ 2.1 What are the functional differences between the variants [if X cl i (then) X cl i ] and [when X cl i (then) X cl i ]?
1 3 RQ 3 Are PCs conditional clauses?RQ 4 How are the different PCs connected vertically and horizontally in the Constructicon and how are they related to default conditional constructions and/or other related constructions?
The 3rd and 4th RQ are of a more theoretical nature.The 4th RQ will be addressed by analyzing the constructions within the framework of usage-based, cognitive construction grammar (CxG; e.g.Goldberg, 2006;Hilpert, 2014;Diessel, 2019;Hoffmann, 2022).
To approach the constructions in a corpus-based manner, this paper analyzes data from the offline versions of three BYU corpora (https:// www.engli sh-corpo ra.org/) that approximate informal everyday speech: the Movie Corpus (Davies, 2019), the TV Corpus (Davies, 2019) and the Corpus of American Soap Operas (Davies, 2011).Taking their communicative function of {requesting and expressing acceptance} into account, it is assumed that PCs are rhetorical devices and consequently most often occur in conversational, dialogical, spoken (informal) registers.This increases the likelihood to find examples of this low-frequency construction in corpora which are representative of this type of language.The data was extracted with Python 3.0 (https:// www.python.org/).In the corpus study, 2151 examples of if-and when-PCs are being analyzed qualitatively and quantitatively.A Logistic Regression Model was also fitted using R-studio (https:// www.rstud io.com/) in order to test if the two types have different constructional profiles and deserve independent constructional status. 2he paper will be structured as follows: The "Pleonastic Conditionals" section will briefly present some theoretical background on default conditionals with some well-known classification schemes (2.1).Afterwards, it will zoom in on some characteristic features of pleonastic conditionals, which are very different from default conditionals (2.2).The "Methodology" section sheds light on the used corpora and methodology (3.1), but also discusses the discourse-pragmatic functions of PCs which have been identified and coded for (3.2).This is followed by a presentation of the empirical findings with descriptive and inferential statistics ("Empirical analysis" section).Among other things, 4.1 primarily discusses the distribution of functions and syntactic types, while 4.2 reports the results of the regression model.The "Constructional Sketch" section turns to theoretical issues and the constructional sketch.The paper ends with a summary and desiderata for future research ("Conclusion" section).

Pleonastic Conditionals
This section provides information on PCs.Before we can discuss some of the constructions' formal and functional features, let us briefly revisit some characteristics of default conditionals.
Many philosophical and logical theories (e.g.Grice's "material theory" (1989)) focus on the truth value of conditionals and not so much on linguistic aspects.In a conditional statement we find a relationship between two propositions P and Q.This type of formal conditional is referred to as a 'material implication'.
An instance of a material implication can either be true or false.P and Q denote the so-called antecedent/protasis and consequent/apodosis respectively (Stanford, 2003: 3). 3n linguistics, many definitions focus on the form that conditionals take in natural speech: "A conditional is a two-clause structure in which one of the clauses is introduced by if (possibly preceded by only, even or expect) or by a word or phrase that has a meaning similar to if" (Declerck & Reed, 2001: 9).The antecedent and consequent of a conditional statement are considered "a proposition which is ultimately expressed by a tensed modalized clause, with a positive or negative polarity value" (Declerck & Reed, 2001: 10).Additionally, Dancygier remarks that "the conjunction then […] often introduces main clauses of conditional constructions, but is usually not necessary" (1999: 12).She continues that if "seems to signal […] that at least some of the felicity conditions for asserting do not hold: the speaker does not have enough grounds for asserting p [i.e. the antecedent] as a factual statement and may in fact not believe p to be true" (1999: 18-19).
Several classification schemes for default conditionals have been proposed in the last years (Table 1): While Huddleston and Pullum follow the traditional classification into realis, irrealis and counterfactual (Huddleston & Pullum, 2002: 738-759), Sweetser (1990) distinguishes between epistemic conditionals, content conditionals and speech act conditionals.In epistemic conditionals (8) an inference can be drawn: (8) If the light is blinking, the ferry's operating.(COCA) In the example, one may observe that the light on the ferry is blinking and infer from that that the ferry is operating.Epistemic conditionals imply "that knowledge of the truth of the hypothetical premise expressed in the protasis would be P => Q {if P then∕therefore Q} a sufficient condition for concluding the truth of the proposition expressed in the apodosis" (Sweetser, 1990: 117).
(9) If it rains tomorrow, the lawn will get wet.
In contrast, in so-called speech act conditionals (10) the antecedent does not describe a scenario under which the consequent is true (Fintel, 2009: 4)."Speechact conditionals are cases where the if-clause appears to conditionally modify not the contents of the main clause, but the speech act which the main clause carries out" (Sweetser & Dancygier, 2005: 113).
(10) If you need a hand with anything, I'm not far away.
The proposition I'm not far away is not a direct consequence of the proposition in the antecedent.What we find in the consequent is an indirect speech act (I can help you) which is linked to the antecedent; i.e. needing a hand motivates the offer of helping.The speaker "asserts the "apodosis" (or rather, engages in the speech act it represents) and contextualizes it with the protasis, which is an evoked and shared context" (Sweetser & Dancygier, 2005: 113).
(11) If you are hungry there are biscuits on the sideboard (Austin, 1961) 'Biscuit' conditionals are a type of speech act or relevance conditional (see Sweetser (1990) above).The consequent (in this case and indirect offer) is said to be the main speech act.In biscuits, the antecedent P "puts some aspect of the consequent q into metalinguistic focus, with the purpose of commenting on some non-propositional aspect of q" (Elder, 2019: 73).The speaker accesses a context in which it is possible that the hearer may be hungry, and sets it up as the background to the main indirect message {take some biscuits if you like}.Here, the truth-value of the antecedent clearly does not represent a condition for the truth or falsity of the consequent.Hence, the primary meaning of a biscuit conditional is not conditional (Elder & Jaszczolt, 2016: 42).
It would go beyond the scope of this paper to spend more time on the presented classification schemes.They are mentioned to make the following crucial point: PCs share many formal features with default conditionals (e.g.bi-clausal structure with if/when X Cl (then) X Cl ).At first glance, they could be classified as a type of biscuit or speech act conditional because they seem similar with regard to their non-conditional but rather communicatively suggestive character.Moreover -like biscuit/speech act conditionals -they are never logically false.However, I argue against this categorization.The problem is that PCs are simply too different to belong to any of the subtypes just mentioned.PCs are a logical tautology and one finds a relationship of sameness between P and Q.In contrast, in biscuit or speech act conditionals we find a relationship of (indirect/inferred) dependence.Therefore, it should be concluded that PCs only share some formal structure with default conditionals.
Of course, the lexical semantics of {if} express a non-factual state in both constructions.In that sense, some PCs (but not all) -like default conditionals -deal with the hypothetical.Be that as it may, this paper argues for their independent status as a constructional family.This, however, does not mean that from a construction grammar point of view, PCs are not linked in some way or another to default conditionals in the English Constructicon (i.e. the network of all constructions of English; see "Constructional Sketch" section).This brings me to a more detailed discussion of PCs and their features.

Formal Features
As already mentioned in the introduction, we find three syntactic variants.One with a connector (if or when)4 including then, one without then and finally a reverse pattern.It remains to be seen how frequent the respective types are in Present Day English.Another interesting aspect is the following: The Q-clause is a precise echo of the P-clause.When this exact duplication happens, Declerck and Reed (2001: 360) call this the "standard form" (18a) and contrast it with a "non-standard form" (18b).Speakers prefer to repeat the whole clause (18a), which "overrule[s] the normal grammatical tendency to replace noun phrases by proforms on their second occurrence" (Declerck & Reed, 2001: 360).Yet the NP gets repeated in its full form, instead of it.This already hints at the stylistically extravagant nature of the construction.In this paper, only the standard form will be investigated.An investigation of the non-standard form and how frequent it is needs to be postponed to future research.
Finally, in many of the examples we find some kind of stacking of pairs which sometimes expresses contrast and opposition: Such combinations are interesting because one needs to ask what underlying constructional templates might license these combinations.Unfortunately, a detailed analysis of these patterns is not possible in this paper and has to be postponed to the future.

Related constructions
We find pleonastic conditionals in many languages, for example, in Portuguese ( 22), French ( 23), or Turkish ( 24 These examples are not supposed to imply that PCs are a high-frequency phenomenon, quite the opposite: they are quite rare.On top of that, in some of the languages the reduplication is not 'perfect', because the antecedent and the consequent differ with regards to inflectional marking or have a different word order. At the same time, PCs seem similar to other constructions which include reduplication. 5For example, (27) Que sera sera?? Whatever will be, will be?? (TV, 2017, Bates Motel) (28) Listen, it is what it is.(TV, 2016, Cops) (29) A: We will go out on a date.-B:What?Like a real date date…?(COCA, 1997, TV) The construction in ( 29) is called "Contrastive focus reduplication" (Gomeshi et al., 2004).The speaker highlights that he/she is talking about a 'real', prototypical date.It will be argued below that one function of PCs is very similar to this (3.2).

Corpora and Queries Used
In order to investigate the construction empirically data was extracted from three corpora: The SOAP corpus, the TV corpus and the MOVIE corpus (https:// www.engli sh-corpo ra.org.).The SOAP corpus is composed of 100 million words in scripts from ten soap operas like All my Children from 2001 to 2012.The TV Corpus is composed of 325 million words in 75,804 texts from the 1950s to the 2010s (the last texts are from 2017).The Movies Corpus is composed of 200 million words in 25,094 texts from the 1930s to the 2010s (the last texts are from 2018).These three roughly correspond to 625 million words; a database which is large enough to investigate a relatively infrequent construction.Admittedly, those three corpora are a bit of a mixed bag especially regarding variety (US, UK, AU, NZ) and timespan.Moreover, one could criticize the data to be 'scripted reality', but I argue that the examples found in these corpora are representative of spoken, (informal) conversational English.
The used off-line data comes in text format and examples were extracted with the help of regular expressions using Phyton 3.0.The following three queries were run to approximate the three main syntactic variants: These regular expressions are quite complex but they guarantee that the antecedent and the consequent are identical.This is ensured by using groups.The antecedent represents a group indicated by "(.{2,})".This group of string characters, as indicated by ".", must take a length of at least two characters but is not restricted with regards to an upper limit in length, indicated by "{2,}".The antecedent group has to be matched exactly in the consequent, something that is indicated by "(\1)".As a consequence, examples like (30) and ( 31) are not included because there we find substitution by a pronoun.Furthermore, [Q3] does not allow when to precede the consequent to avoid examples such as (34), in which the interlocutor intends to draw attention to the possibility that the event described in the when-clause might not occur after all by addition of an if-clause with the same proposition.
(34) When we win this thing, if we win this thing, you'll be my chief assistant (Movie, 2013, Dark Power) Implementing the above restrictions as well as an additional word-boundary restriction (indicated by "/b") to avoid matches such as (35), the regular expression [Q3] returns pleonastic conditionals like ( 36 The sentence basically means {if you were not with the right crowd, then you were not safe}.Also (38) was excluded as the first part refers to biologically fathering a child whereas the second part rather seems to refer to raising a child.
(38) I gotta be involved-If Iʼm gonna be a father, then Iʼm gonna be a father (TV, 2008, Boston legal) Note that early on, additional queries based on [Q2] were run to check if other conditional connectors are used in the construction: [as|so] long as (.{2,}), (\1 No hits were found with these other connectors, which is why this paper only analyzes if and when.In the end, 2151 examples of 'real' if/when-PCs remained.These were coded manually for connector type (if, when), syntactic variant (con_incl_then, con_excl_then, reverse), verb type, polarity value (positive, negative) and pragmatic function (see "Discourse-Pragmatic Functions of PCs" section below).

Discourse-Pragmatic Functions of PCs
One major aim of this paper is to analyze the different discourse-pragmatic functions of PCs.Inspired by Declerck and Reed (2001) who already mention the function of acceptance, I propose the following four main functions of pleonastic conditionals: acceptance-indifference-factuality-prototypicality.As will be shown, it is not easy to demarcate these functions.After all, meaning is always co-text and context dependent and a construction can be multi-functional.In that sense, this paper is only a first attempt to establish a functional classification of PCs, being well aware that the proposed categories might have to be adjusted or fine-tuned in the future.

a. Acceptance
The first most basic function of a PC is to express acceptance.The speaker expresses their own acceptance or nudges the addressee to accept something in case it happens.For example, one can accept the necessity of a proposition (39), fate/kismet (40) or a proposition with all it entails (41): To communicatively request or express acceptance also entails that the construction is often used in a context that is conversational and dialogical.In ( 42 At the same time, speakers can also comment on their own actions and express acceptance of their own situation: (43) A: I was hoping to make it to my prom.But if I can't, I can't (TV, 2001, ER) Note that in both cases, the construction is what has been called 'backwardpointing', 'responsive', or 'reactive' (Fried & Östmann, 2005;Linell, 2009;Põldvere & Paradis, 2019).Almost always, the PC comments on a proposition or topic which has already been discussed in the conversation before (see "Constructional Sketch" section for details).

b. Indifference
In some contexts, acceptance leads to indifference, expressing the notion of disregard.In this paper this was coded as a separate function but admittedly it is debatable whether this is not just a subtype of acceptance: (44) If it breaks up, it breaks up!Who cares?(SOAP, 2004, YR) (45) I donʼt care-Iʼm not going to stop you-so if youʼre in, youʼre in (SOAP, 2007, AMC) (46) It is an investment, and if we lose it, we lose it.There are more important things (SOAP, 2008, GL) That the speaker is indifferent is primarily expressed in the existing co-text (Who cares?;I don't care).Thus, it could be argued that indifference is not really coded by the constructional template but that this meaning only emerges in some contexts and in combination with other constructions.Even if this is an argument not to assign this function to the PC template, I still propose that indifference is coded 'construction-internally' by the reduplicating structure of the constructional template.After all, reduplication in language is used elsewhere and in other constructions to express indifference: (47) A: Right, unless they're conservative approved links.
B: Yeah, yeah, yeah, heard it all before (COCA, 2012, BLOG) (48) A: I only know one project worldwide that makes money in Scandinavia.
B: Laura!Bla bla bla I believe it when I see it (NOW corpus, Tech Central 10-03-08) This paper follows Linell (2005) and others who speak of "meaning potential".No construction has a completely fixed meaning, but rather a potential to mean.What a construct means is always "negotiated locally" and is influenced "by constraints that emerge in and through situated interaction" (Ono & Thomson, 1995: 220).Still, as soon as a particular meaning (e.g.indifference) gets repeatedly inferred from the co-and context it can be internalized by the speaker and stored directly with the formal shape of the template.I argue that this has happened with the indifference function.

iii. Factuality
The next discourse-pragmatic function also includes the notion of acceptance but the main function of the PC seems slightly different: In this type, the speaker primarily uses the PC to highlight the factuality and certainty/ irreversibility of the proposition.In the sense of: {X is a fact.period.nothing can be done about it}.(TV, 1999, Farscape) This factual type is especially interesting because here the {if/when} no longer expresses a hypothetical or future scenario but refers to an (irreversible) fact.In (49) the person is definitely allergic, in (50) the store is closed and in (51) the addressee has already lost.In that sense, the reduplication of the proposition stresses the inevitability of the fact.Obviously, acceptance of this factuality plays a role here as well.Still, I consider these examples different from the acceptance ones because the speaker does not necessarily demand or express acceptance of the proposition but primarily requests acknowledgment.For instance, (50) does not foreground {please accept that the store is closed} but rather expresses annoyance {Can you not see that the store is closed?}.Admittedly, it is often not easy to clearly demarcate the two functions (see mixed types below).

d. Prototypicality
The final function, which I argue has been overlooked completely in the literature is very different.It is one of intensification expressing prototypicality or 1 3 If that's what she said, then that's what she said: a usage-based,… stereotypicality.It is a set-shrinking expression to a prototypical event.This is very similar to contrastive focus reduplication mentioned in 2.2.2.
(53) They are the best beepers that bucks can buy.I mean when they beep, they beep (TV, 1981, Soap) (54) Iʼm sorry, Kevin.If youʼre paying for dinner, youʼre paying for dinner.
None of this coupon shit (TV, 2017, Inside No.9) (55) What a sickly little bunch of violets?When I buy flowers, I buy flowers (Movie, 1947, Copacabana) For example, the statement When I buy flowers, I buy flowers expresses that the speaker prefers a rich, beautiful, representative, prototypical bunch of flowers and not a small, cheap, sickly bunch.In this function, the speaker does not express acceptance at all.

v. Mixed Types
When coding the data, one also finds mixed types where it is simply impossible to decide on a single function.For instance, ( 56 As can be seen, assigning functions is not a simple task.Also note that the final interpretation of the construction is strongly influenced by the (lexical) semantics of the connector, something which the regression model in the next chapter will confirm.If is much more hypothetical than when.Aware of all these caveats, a classification was attempted along the lines outlined above.
I proceeded in the following way when coding for pragmatic function: Always, the extended context (4 sentences to the left; 4 sentences to the right) was taken into consideration.I distinguished between the four main types: (i.e.acceptance, indifference, factuality, prototypicality) and two mixed-types: Acc_Fact and Acc_Ind.As the prototypicality function could almost always be identified in a straight-forward manner, no mixed type was created here.A rest category was created which subsumes a) examples with more than two meanings, b) examples where there was not enough context to allow for assigning a function and c) examples where the overall meaning of the sentence remained completely unclear.

Empirical Analysis
What follows is the presentation of results.The main idea was to empirically investigate how frequent PCs are in general and to check whether the if-and the when-type differ significantly in their preferred functions and their preferred syntactic variant.Finding evidence for such a difference strengthens the argument that the two types deserve independent constructional status.What implications the findings should have for a sketch of the English constructicon will be discussed in the "Constructional Sketch" section.

Descriptive Statistics
Figure 1 shows that in general PCs are a low frequency phenomenon.In a corpus of 625 million words, we only find 2151 examples, which suggests that they are a rather rare stylistic device.Among the syntactic variants, the type without then (con_ excl_then; e.g.If/when you have to, you have to) is the preferred majority pattern with 1957 examples.In contrast, the reverse type You have to if/when you have to is least common.
With regard to connector use, if-PCs (1273) are more frequent than when-PCs in my sample (878) (Fig. 2).Among both types the con_excl_then type is by far the most frequent one.When is used relatively often in the reverse type (85), e.g.It ends when it ends.In contrast, [X cl i if X cl i ] Cx is rare (8) and [when X cl i then X cl i ] Cx only occurs twice.The examples were also coded for their polarity.An example was classified as having a positive polarity if it expressed a positive proposition.PCs including a negation with not but also clauses including negative lexical items like never, nothing, nobody etc. were considered to have a negative polarity value.As shown in Fig. 3, the if-type includes a much higher number of examples with a negative polarity value (964).When is used in a negative statement only 19 times.This is not surprising because the semantics of when are closer to reality than the semantics of if which has always been considered to be more hypothetical.In that sense, if goes together better with negative polarity.The result indicates that the two types clearly differ with regards to polarity.
Figure 4 shows that the two types also have different functional preferences: The if-type mostly expresses acceptance (721), followed by factuality (300), whereas the when-type mostly expresses factuality (504).This again is not surprising, because the semantics of when construes a world of non-hypotheticality.In other words, the hypothetical semantics of if expresses the notion of {in case X happens, accept it} whereas when tends to be used when expressing facts.Moreover, indifference is coded by the if-type (117) but almost never by the when-type (5).Finally, the when-type is clearly the preferred type to express prototypicality (93).

Logistic Regression
All these findings and preferences are confirmed by a binomial logistic regression model (Levshina, 2015;Winter, 2020) using R/Rstudio (R Core Team, 2022; R Studio Team, 2021). 6The model was fitted to predict the choice of connector IF or WHEN.Logistic regression models are a useful tool to capture the statistical relation between a set of predictors and a categorical outcome (the dependent variable = in this case 'choice of connector').This method is commonly used in variationist studies, where it has been typically applied to predicting what determines the choice between two grammatical variants that are deemed to be somewhat equivalent (Table 2).

Fixed Effects
In Table 2, if is the baseline hidden in the intercept, which means that the given slopes in the estimate column are for when.Categorical variables were treatmentcoded and the following baseline levels were chosen: For syntactic variant the baseline is 'con_excl_then' because this is the most frequent pattern.For polarity value the baseline is 'positive' and for pragmatic function the baseline is 'acceptance' as I consider this to be the most basic function.There are no issues with collinearity: all variance inflation factors are very small; the variance accounted for by this model is pretty decent with 34.4%.
As Table 2 reveals, there were several significant effects and correlations: As indicated in the estimate column, the reverse type increases the likelihood for when to be used.The syntactic type con_incl_then prefers if.Negative Polarity value does decrease the likelihood for when.These results are also visualized for convenience in Fig. 5 which shows the predicted probabilities for the when-type to be used.
When it comes to pragmatic functions (Fig. 6), factuality and prototypicality correlate with when.In contrast, indifference is almost always expressed by if.

Interactions and Random Effects
I also tried to fit interactions between polarity and pragmatic function, and between syntactic type and pragmatic function.This would have told us whether the different combinations of the variables involved in the interaction have different influences (compared to their separate effects) on the predicted probabilities of the connector.Unfortunately, there was an error message when trying to fit both interactions which warned that predicted probabilities of 0 or 1 had occurred, as well as large standard errors.After removing one of the interactions, the warning disappeared, but the standard errors were still large, so a model without interactions was fitted. 7ust because fitting interactions did not work in the regression model, does not mean that these variables do not interact in the real world.Figure 7 suggests that syntactic type and pragmatic function affect each other.
For instance, it can be observed that the con_incl_then type expresses a higher proportion of acceptance than con_excl_then.sample.Why this is the case is unclear.There is no obvious reason why the reverse type should block an indifference reading.Most likely, this finding is an artifact due to the small sample and I expect the indifference function to occur with the reverse types in a larger sample (see "Constructional Sketch" section for further interpretation of the results).
For corpus data, it is often important to treat the individual texts in which the examples occur as random effects.However, for this data, one runs into singular fit issues both when choosing 'text' and 'corpus' as random effects.These issues could not be resolved even with all slopes removed.This indicates that there is not enough variance between the different transcripts and the three corpora to justify fitting a mixed effects model, and/or that there is not enough data for the random effects structure, especially when individual texts are used as random effects.In the future this shortcoming might be overcome by collecting more data points, i.e. examples from other corpora.At the same time, I do not consider this issue a major problem because the three corpora all represent informal conversational language used 'on screen' and I never expected major differences between the corpora in the first place.
Although any regression model has its limitations, I argue that the one presented above is sufficient to confirm that the two connector types behave differently in many respects and therefore deserve separate constructional status as individual constructions; an idea which is outlined below.
Constructions are relatively stable, cognitively entrenched patterns which have emerged from language users' cumulative experiences of utterances.They are abstractions/schematizations over sets of linguistic expressions parallel in form and function.What is special about CxG as a model of linguistic knowledge is that "different traditional 'levels' of grammatical description (phonology, syntax, semantics, prosody, pragmatics, discourse, etc.) are integrated in a single complex sign" (Fried & Östmann, 2005(Fried & Östmann, : 1754)).A construction directly maps a formal linguistic code not only to semantic but also to discourse-pragmatic function (Fig. 8).It is assumed that speakers are aware of the "functions constructions have and what sort of pragmatic effects one can achieve with them" (Linell, 2009: 99).At the same time, speakers internalize probabilistic information about variation, for instance that specific constructions are used more or less in certain regions, genres, registers or by specific socio-economic classes etc.Finally, several interactional linguists who work in CxG stress that a speaker also understands that some constructions require sequential links to connected discourse and other (previously uttered) constructions and communicative turns in the co-text (Auer & Pfänder, 2011).Constructions may have responsive or projective properties (e.g.Fried & Östmann, 2005;Põldvere & Paradis, 2019).
In general, constructions have been classified on various dimensions, regarding their complexity (ranging from atomic to complex) and specificity (ranging from specified/substantive to schematic) (see Croft & Cruse, 2004: 255;Hilpert, 2021: 6-7).Moreover, related constructions are linked to each other via horizontal and vertical links, a design feature which ultimately creates networks of constructional families (Smirnova & Sommerer, 2020;Sommerer & van de Velde, accepted).Vertical links express so-called inheritance relations where linguistic information percolates down from more abstract mother constructions at higher positions down to more specific lower-level daughter constructions.The lower-level construction is thus a special, more specific instance of the higher-level construction (see Goldberg, 1995: 79-80;Hilpert, 2014: 59).Constructions also entertain lateral or horizontal links.Constructions are seen as horizontal sisters to each other if they are on the same level of complexity and are similar to each other in form and/or function (Diessel, 2019: 199-200).
It goes beyond the scope of this paper to discuss the architecture of the model and the different types of vertical and horizontal links in detail.For this paper, it is sufficient to state that it is a construction grammarian's goal to a) determine if a linguistic string deserves constructional status as an independent node in the network, b) describe the construction's form-function mapping and c) sketch in what way the target constructions are linked among themselves and to other related constructions.

PCs and their Form-Function Mapping
Most researchers in CxG subscribe to the following definition of a construction: Any linguistic pattern is recognized as a construction as long as some aspect of its form or function is not strictly predicable from its component parts or from other constructions recognized to exist.In addition, patterns are stored as constructions even if they are fully predictable as long as they occur with sufficient frequency.(Goldberg, 2006: 5) Following this definition, it is clear that PCs deserve independent constructional status.First, their functions are different from default conditional which express a condition or inference.Moreover, the proposed conventionalized discourse-pragmatic meanings of the constructional template are not predictable from the template's component parts but are linked to the observable reduplication.These meanings (acceptance, prototypicality, etc.) emerge on top of any propositional semantics of the X clause.In that sense, it is the overarching Gestalt of the pleonastic template and not its individual parts which carries a PC's pragmatic meaning.
PCs are considered a constructional family with three variants and with if an when as the main conditional connectors.An utterance like (59) is a construct which is licenced by the following constructional template: In this constructional template X has to be a clausal unit (with a few elliptical exceptions, e.g. if not then not) and represents a proposition that can either have a positive or negative polarity value.Using the superscripts highlights the identical nature of the elements.Next to the propositional semantics of the clause, this formal template is directly connected to the pragmatic functions that have been established in "Discourse-Pragmatic Functions of PCs" section.I also suggest that the speaker has stored the 'experience' that this is a construction to be used mainly in informal conversational interactions.Moreover, following Linell (2009) and Põldvere and Paradis (2019), I argue that the construction is 'backward-pointing' and 'reactive' (see discussion in "Discourse-Pragmatic Functions of PCs" section).
In comparison, an utterance like (60) instantiates the template: The subscript numbers 1 to 4 express a functional ranking based on the observed frequency of use.For the if-type, the acceptance function is most frequent, whereas for the when-type it is factuality (as shown in "Descriptive statistics" section).Due to their higher usage frequency, acceptance and factuality are seen as the prototypical meanings of the construction.Obviously, this ranking is based on a rather small dataset and might change if more texts from other genres are taken into consideration.To integrate the findings from the regression model, the bold font indicates that the acceptance and indifference function increases the likelihood for if, whereas the speaker knows that when he/she wants to express factuality or prototypicality when should be used.
Finally, the two constructional templates above indicate that then is optional in both templates.The if template licenses examples with and without then.The same holds for the when template.The analysis with optional then-insertion is preferred to an analysis where four different types of constructions are postulated, as it is more economic and assumes fewer nodes.Moreover, then-insertion is not very frequent.I consider this a fact to be expected for the following reasons: The [if/when…then…] combination most often expresses some kind of causal or conditional relationship-at least in default conditional clauses.However, expressing conditionality is something which is missing from the PC's functional repertoire.This is why it is expected that then-insertion is not the preferred speaker option.If it happens, I interpret it as an example of what has been termed 'constructional contamination' (Pijpops & van de Velde 2016) which is a process where a construction is structurally influenced by another functionally non-related construction due to superficial formal similarities or similar linguistic environments.In other words, constructional contamination happens when semantically unrelated constructions with similarity in formal shape nonetheless exert an influence on another construction and thus on the speaker's behavior.In our case it is assumed that some speakers feel the need to add a then in some of their PCs due to the fact that this is a frequent formal template with default conditionals.
To conclude, the when-and the if-type are considered to be individual constructions.The empirical analysis has revealed that both types occur relatively often.Moreover, the empirical analysis shows that the various functions are distributed differently.The constructions also differ with regards to their expression of polarity.The when-type favors positive polarity statements.This warrants making a distinction and postulating separate templates.

PC Network
With regard to network structure, the PC constructions which have been investigated could be arranged in the following network with horizontal and vertical links (Fig. 9): The idea is that speakers-based on their input-at one point abstract the following constructional templates on the lower levels.From these, the speakers abstract a more abstract template with an open conditional connector slot in first position and optional then insertion.
In the reverse type, the conditional connector appears in the second part.Again, constructs with both when and if exist.The problematic aspect here is that the reverse type is structurally different from the other two syntactic variants, with the connector being postponed, which makes it difficult to integrate it into the network.However, by opting for a so-called constructeme analysis (in the sense of Cappelle, 2006), it is possible to connect the third type to the other two types.The motivation for this is that all PC types are functionally and formally partially related (Fig. 10).
In a constructeme analysis we have a mother node called constructeme that remains somehow underspecified, for instance with regards to any word order.The word order of the constructeme is not fully fixed, so that it can license all syntactic patterns which in the model are seen as 'allostructions'.Perek describes the relation between constructemes and allostructions as follows: "the constructemes capture the level at which constructions are semantically equivalent and the allostructions specify exactly how these constructions differ" (Perek, 2015: 153).
The template on top of the network in Fig. 10 merely expresses the necessary ingredients of a PC construction, namely a reduplicating clause and a conditional connector in one of two positions and an optional then.
In the template, the curly brackets enclosing the connector slots and the superscripted v express that the connector can occur in more than one position but the slots are mutually exclusive.The sketch in Fig. 10 only captures the PCs that have been investigated empirically so far.It is thinkable that if the empirical investigation was extended, we might find examples in which other conditional connectors are used (e.g. even if, unless).If future corpus analyses reveal that other connectors are also recruited into the connector slot, the network would have to be extended.The current network also does not include the templates for the non-standard form which was discussed in "Formal Features" section.Ultimately, also the non-standard template must be added to the PC network.Moreover, the current network sketch focuses on the PC family and the assumed internal links between the templates.That being said, it also visualizes assumed links to other constructional families which PCs will most likely be related to, e.g.default conditionals, other reduplicating structures and other back-ward pointing constructions.However, future research is required to investigate these inheritance relations in much more detail.

Conclusion
This paper has discussed PCs as stylistic, rhetorical, constructional templates of low frequency.The empirical analysis has revealed that one finds three main syntactic variants which are distributed unevenly regarding their raw frequency.The con_exl_ then type is by far the most frequent one.The corpus analysis also showed that the construction only recruits if and when into the connector slot.The if-and when-type have different preferences when it comes to syntactic variant, polarity and discoursepragmatic function.From a CxG modelling perspective it was argued that the two types deserve independent constructional status due to their different profiles.Moreover, they are considered more specific instantiations of a more abstract template [Con cond X cl i (then) X cl i ].The reverse type [X cl i Con cond X cl i ] Cx was integrated into the network via a constructeme analysis.The paper has also shown that the extravagant, pleonastic nature of PCs is used to express various discourse-pragmatic functions (some of which have been overlooked in the literature).These communicative functions are of a conversational, interactional nature.The main functions are a) to express or request acceptance of a proposition or b) state the factuality of a proposition.Here, the if-type mostly expresses acceptance, whereas the when-type most of the time expresses factuality.The when-type almost never expresses indifference but it is clearly the preferred type to express the prototypicality function, which is a function very different from the others which has nothing to do with acceptance.
It was also argued that PCs are very different from default conditional clauses.They are only similar in form but semantically do not express a condition or dependency.The tautological, reduplicating nature of the template encodes non-conditional, pragmatic meaning like acceptance or indifference.Still, it has been argued that this family might be linked to the larger network of default conditionals after all, as it seems to be the case that especially then-insertion comes about due to constructional contamination, which in the network design corresponds to some linking between the different constructional families.
The following aspects should be investigated in future research: It seems fruitful to investigate the constructions' productivity and to conduct distinctive collexeme analysis to analyze potential lexical preferences of the respective types.For instance, certain verbs might be attracted to the if-or when-type and some verbs might prefer a particular function.It will also be necessary to investigate what Declerck and Reed (2001: 360) have called the non-standard (pro) form (see "Formal Features" section).Here, the question would be how frequent this variant is and if it is really true that speakers prefer the standard form with full reduplication of the antecedent.
Another goal should be to investigate default conditional clauses and their distribution in more detail.This would enable a more fine-grained analysis of how PCs are different from normal conditionals with regard to their syntactic variation and formal preferences.The corpus analysis must also be extended to different corpora and genres as it has to be investigated if the construction really is more frequent in interactional conversational registers as I have claimed.In other words, the question is whether and how often it is used in more formal (academic), and or written genres; something which needs to be confirmed empirically.
If we want to successfully position PCs in the English constructicon, we have to dig deeper and look at the potential connections to other (similar) constructions.It has to be investigated in what way PCs are linked to default conditionals, other reduplicating constructions and/or other dialogical constructions which are interactional and responsive (backward-pointing).I have mentioned constructional contamination and potential inheritance relations in the "Constructional Sketch" section but this idea needs to be researched further and elaborated on in the future.Last but not least, it would be very interesting to investigate pleonastic conditionals in other languages and trace their diachronic development.
The attempted functional classification of PCs once again shows how tricky it is to clearly demarcate a construction's discourse-pragmatic communicative function; what a string ultimately means always emerges in context and is heavily influenced by co-text.The ultimate question is which meanings and functions one identifies as stable and decontextualized enough to deserve integration into the constructional template (e.g.acceptance) and which meanings are more inferred from co-and context and thus should be seen as 'secondary' and not stably codified (e.g.indifference).In any case, this paper agrees with Fried and Östmann (2005) and Auer and Pfänder (2011) who argue that CxG is capable of modeling the discourse-pragmatic side of constructions and can successfully show how communicative, interactional meaning is directly mapped onto syntactic form.
Funding Open Access funding enabled and organized by Projekt DEAL.

Declarations
Conflict of interest On behalf of all authors, the corresponding author states that there is no conflict of interest.'If that's what she said, then that's what she said': a usage-based, constructional analysis of pleonastic conditionals in English Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made.The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material.If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.To view a copy of this licence, visit http:// creat iveco mmons.org/ licen ses/ by/4.0/.

(
18a) If Ted wonʼt come, then Ted wonʼt come (Standard Form) (18b) If Ted wonʼt come, then he wonʼt come (Non-Standard Form) This is especially interesting in (19) where the inside of his house is a very long and complex NP.(19a) I mean, if I have seen the inside of his house, Iʼve seen the inside of his house (Standard Form) (19b) I mean, if I have seen the inside of his house, Iʼve seen it (Non-Standard Form) If that's what she said, then that's what she said: a usage-based,…

(
20) I need to do this on my own.If I fail, I fail.If I pass, I pass.(TV, 2012, The Big Bang Theory) (21) I'm not going to stop you-So if yourʼre in, youʼre in.If youʼre out, youʼre out (SOAP, 2007, AMC) ): (22) Se não dá, não dá (If it does not work, it does not work) (23) S'il ne veut pas, il ne veut pas! (If he does not want to, he does not want to) (24) Bittiyse bitti (If its done, its done) PCs also exist in German: (25) Wenn es sein muss, muss es sein! (If we must, we must) (26) Wenn die Mama das gesagt hat, hat die Mama das gesagt!(If that's what Mommy said, that's what Mommy said)

( 30 )
If John has moved on, he has moved on.(31) You do what you have to do to survive, and if what you have to do sucks, well then it sucks.(Movie, 2016, Black Tar Road) This condition holds true for all regular expressions used in the empirical investigation.The regular expression in [Q1] and [Q2] also does not match 1 3 If that's what she said, then that's what she said: a usage-based,…examples in which the consequent is followed by any element that is a letter, something expressed by "(?![a-zA-Z])".This measure aims to exclude examples like (32):(32) Right, but if you were there, you were there with him, right?(Movie, 2014, A Murder in the Park)This restriction reduces the set of hits to only those cases where the PC occurs next to a punctuation mark.[Q2] is the same as [Q1] without then.The third regular expression [Q3] aims to identify examples of the reverse type.In order to avoid examples like (33), which are considered a case of repetition, stutter or false start on behalf of the speaker, the expression does not allow the connecting device if to precede the consequent (?<![if|when]).(33)If I killed him, if I killed him, I'd make a break(TV, 2003, Law and  Order)

( 39 ) 3
If you gotta go, you gotta go (SOAP, 2010, ATWT) (40) As far as you and my sister go, you know, if itʼs meant to be, itʼs meant to be (SOAP, 2005, OLTL) (41) If youʼre for the death penalty, youʼre for the death penalty.Even if the condemned murderer is cute and cuddly (TV, 1998, Law & Order) 1 If that's what she said, then that's what she said: a usage-based,…

( 49 )
If you are allergic, you are allergic (TV, 2015, A to Z) (50) I don't know what to tell you, if they are closed, they are closed (TV, 2004, Curb your enthusiasm) (51) I donʼt know why you need an excuse.If you lost, you lost (TV, 1976, The Waltons) (52) When you die, you die If that's what she said, then that's what she said: a usage-based,…

Fig. 3 Fig. 4 3
Fig. 3 Distribution of if-and when-PCs by polarity Fig. 5Predicted probabilities of connector when depending on syntactic type (left) and on polarity (right)

Fig. 6 Fig. 7 3
Fig. 6 Predicted probabilities of connector use when based on function

F
If that's what she said, then that's what she said: a usage-based,…

Fig. 10 3
Fig. 10 PC network in Present Day English If that's what she said, then that's what she said: a usage-based,… 1 3

Table 1
Classification schemes for conditional clauses

If he doesnʼt want to be with her, then he doesnʼt want to be with her
What do you want from him, Jenna.(SOAP, 2007, ATWT) (13)

You can't tell me if you can't tell me (
Movie, 2008, Synecdoche)Despite all the set restrictions, [Q1]-[Q3] return approximately 4500 examples, which is a recall full of noise.Going through the examples one by one reveals that roughly 50% are not the target construction.One finds many false positives but also doublets, which had to be excluded manually.For instance, I consider (37) a false positive.The two you weren't refer to two different propositions.
), speaker A states that he cannot help himself and speaker B reacts with a PC signaling acceptance:

if Iʼm going crazy, Iʼm going crazy. Lock
) is compatible with acceptance and me up(TV, 2009, Defying  Gravity)Here, the example is basically compatible with acceptance, indifference, and prototypicality.