To appear in: Herbst, Thomas & Peter Uhrig (eds.). 2015. Yearbook of the German Cognitive Linguistics Association, Volume III. Berlin: De Gruyter.

Let’s go look at usage: A constructional approach to formal constraints on go-VERB Susanne Flach Freie Universität Berlin

Abstract The English go-VERB construction as in I go get the paper every morning is said to be subject to the BARE STEM CONDITION, which states that neither verb can bear inflection (thus resulting in the ungrammaticality of *she went see/saw a doctor). Studies that address the constraint attribute it to underlying formal parameters, without paying attention to functional properties and/or usage events. The fact that we find occasional violations of the constraint in large amounts of data raises the question of the systematicity of this data and how to account for it. Arguing from a usage-based perspective, this paper assumes that the entrenched schema of go-VERB is hortatory (e.g., commands, advice, invitations), which make inflectional variants increasingly unlikely for semantic-functional reasons. But where exceptions do occur, they are assumed to occur in contexts predicted by the construction’s semantics. These assumptions are borne out by data from a large corpus of web data. Potential counterexamples are tested for systematicity vs. random noise in (web) data using collostructional analysis and simple correlation measurements. The central arguments are thus (i) that the BARE STEM CONDITION is better conceived of as the result of go-VERB’s constructional semantics, and (ii) that rare exceptions can be framed in terms of likelihood of occurrence and distance from the licensing schema.

Keywords: serial verbs, formal counterexamples, corpus linguistics

1

constraints,

Construction

Grammar,

Introduction

The ‘quasi-serial’ go-VERB construction (Pullum 1990) as in I go get the paper every morning or Go see a doctor! has seen continued interest over the years. It has primarily been studied with respect to constructional non-synonymy (Wulff 2006, 2008; Nicolle 2009), its historical development (Bachmann 2013; Nicolle 2007), or, in formal analyses, the morphological or syntactic status of the verbs involved (Bjorkman 2015; Jaeggli & Hyams 1993). Despite a number of contentious issues, two properties of go-VERB are largely undisputed: the fact that (i) both verbs occur in the uninflected form (BARE STEM CONDITION), and that (ii) both verbs share the same form (IDENTITY CONDITION). Therefore, examples such as *she went see a doctor or *he goes get(s) the paper are said to be ungrammatical. From the perspective of usage-based, constructivist corpus linguistics, such judgements inevitably raise questions: are ‘ungrammatical’ examples also unattested, or is it plausible to assume that speakers violate the condition? If yes, where would we expect exceptions, and, ultimately, are exceptions systematically licensed by the construction? Following the outline of descriptive problems in previous theoretical accounts on the BARE STEM CONDITION (Section 2), the major claims are, first,

A constructional approach to formal constraints on go-VERB

that the constructional meaning of go-VERB is hortatory (Section 3). Second, its prototypical—and highly entrenched—use types are therefore largely incompatible with contexts that are subject to inflection in English. Third, analysing data from a large webcorpus (ENCOW14, Section 4), rare but systematic violations of the BARE STEM CONDITION occur in contexts predicted by constructional semantics (Section 5). Thus, the central argument is that while the BARE STEM CONDITION can be confirmed as empirically accurate, it is better conceived of as the result of constructional meaning, rather than deriving from underlying parameters or operations (Section 6).

2

Theoretical perspectives & descriptive problems

Almost all treatments of the go-VERB construction mention the BARE STEM CONDITION as a key property, which states that neither go nor the verb in V2 can be inflected. While in most formal analyses this property is a theoryinternal object of study (Bjorkman 2015; Carden and Pesetsky 1977; Jaeggli and Hyams 1993), the constraint is usually mentioned in functional analyses, but generally not elaborated on further. The examples in (1) exemplify ‘grammatical’ uses, those in (2) the ‘ungrammatical’ ones (following Pullum 1990:219; Nicolle 2007:49; Jaeggli and Hyams 1993:313–318, inter alia): Every day I go get the paper. Let’s go find the paragraph marker. Go climb a rock! The police insist that Bill go talk to the counselor. Didn’t he go eat there yesterday?

(1)

a. b. c. d. e.

(2)

a. *Every day my son goes gets the paper. b. *She goes visits her friends throughout the year. c. *He went ate at that restaurant yesterday.

The first observation is that the grammatical examples occur with either first or second person subjects and/or in various non-finite or imperative syntactic environments: a let’s-construction in (1b), an imperative in (1c), a subjunctive in (1d), and do-support in (1e). In other words, the go-VERB construction is not subject to inflection here. With the exception of (1a), all acceptable examples are thus in considerable contrast to the syntactic (and semantic) contexts of the ‘ungrammatical’ examples, which are syntactically indicative and semantically representative. This observation is not trivial. While all authors note the preference of goVERB for non-inflecting contexts, this fact has not been taken into consideration to account for the BARE STEM CONDITION. Rather, the theoretical centrality assigned to the BARE STEM CONDITION assumes that the non-finite or imperative contexts of (1b–e) follow from the BARE STEM CONDITION. This implicit assumption leaves the declarative context describing a habitual event in (1a) as the major reference point against which the ungrammaticality of inflectional variants is assessed. This could constitute a methodological problem: in the absence of systematic empirical analyses, there seems to be a considerably 2

A constructional approach to formal constraints on go-VERB

deceptive skew in the theoretical literature towards suggesting that indicativerepresentative contexts as in (1a) are typical and central contexts of go-VERB. However, there might be something salient in the constructional properties of go-VERB that affects its occurrence probability in inflection-prone contexts for semantic-functional reasons. In other words, the BARE STEM CONDITION could plausibly be the result of go-VERB’s distributional preference for nonfinite contexts (rather than the other way around). Thus a cognitively oriented constructional approach (Goldberg 1995; Langacker 1987) seeks to add a perspective from the other end to account for a seemingly puzzling morphosyntactic constraint.

3

Constructional meaning of go-VERB

What is noticeable from observing even a few concordance lines of go-VERB is that many uses of go-VERB seem to share, semantically, an affinity for contexts of orders, suggestions, and recommendations. The question is whether this is indicative of a more schematic property of the go-VERB construction. This section will explore the meaning of go-VERB, arguing that the construction displays a number of closely related meaning clusters with a strong ‘hortatory’ core meaning. The term ‘hortatory’ will be used here as a relatively broad cover term for directives or commands, but is also used to include utterances of recommendations, advice, encouragements, promises, or invitations (i.e. where the illocutionary force is closer to commissives than to directives), especially between speaker and hearer. It should be borne in mind that ‘hortatory’ as used here is distinct from ‘hortative’ as a grammatical category. 3.1

The use of go-VERB

The examples in (3) repeat the contexts of go-VERB that are generally cited in support of formal properties, here with comments on their functional spectrum: (3a) is a request for the speaker herself, (3b) illustrates intention with a slightly negative evaluation, (3c) acts as an invitation to speaker and hearer(s), and (3d) conveys subjective necessity in a wider sense:1 (3)

a. b. c. d.

I need to go get me a copy of that Express paper. I said well we’ll go get him, we’ll try again. Oh, let’s go get pizza and go watch a movie spontaneously. Maria tells Nagi that she should go take a bath.

The examples in (4) have contexts with invitations or encouragement to do (or enjoy) something—a ‘promotional’ directive—while the contexts in (5) add negative or belittling evaluation towards actions or beliefs of the addressee: (4)

1

a. Go read Atlas Shrugged, go see Gran Torino, go inspire yourself in a way that works for you. b. Now go have a nice relaxing bubble bath or something.

All examples cited here are from ENCOW14AX01 (cf. Section 4).

3

A constructional approach to formal constraints on go-VERB (5)

a. Nothing that actually defends Dr. Wakefield against the real charges. Seriously, go read for yourself. b. Go look it up: A theory is not the same as a theorem.

The evaluative function also often materializes in sarcastic uses (6), which transport the speaker’s strong negative comment on a situation: (6)

a. And yes, Mark Antoinette, I’ll go eat cake now. b. “If you like the BBC so much why don’t you go live there?”

Go-VERB in interrogatives often occurs in the pattern why don’t X go-VERB? with a primarily derogatory overtone (especially with live), but is also used in neutral or positive recommendations with varying degrees of urging or inviting force (why don’t you go have fun?). One highly prominent ‘derogatory’ sub-sense is the go-fuck subconstruction (possibly more salient due to the source containing data from interaction of people on the web). It involves a serious personal insult, uttered in disgusted reactions towards some behaviour or attitude. It typically occurs with fuck, as in (7), but can take other verbs (go screw yourself, go shove X up Y’s ass). Extended uses in (8) specify the manner of ‘recommend action’ (spam, terrorize, troll, rant). Note that although go fuck is a directive, it often occurs outside of imperatives (7b): (7)

a. Go fuck yourself and quit trolling. b. And then I hope someone tells them to go fuck themselves.

(8)

a. ‘Leave us alone. Go terrorise someone else’. b. Go suck on someone else’s a-hole and try and do something productive, like not losing.

While the hortatory force in go-VERB is prototypically associated with imperatives, the directives also appear in other syntactic environments. A sizeable portion of go-VERB occurs in to-complements of semi-modals (have to, need to), speech act verbs (tell, ask, require, also cf. (7b)), or suggestion verbs (suggest, encourage, recommend), as in (9). It is also used in make/have X do Y patterns, as in (10): (9)

a. I have to go give a mission report to my Father, we’ll make sure its cleared with him, so the council doesn’t bite our heads off later … b. I encourage you to go read the entire Opinion.

(10) a. Successful product marketing isn’t arguing semantics with your customers on your website or making them go read the fine print ... b. ‘Oh, the radical men kept women under their thumb, had them go make the coffee in SDS,’ but I didn’t see any of that.

Such contexts act as directives through a wider linguistic co-text. The same applies if go-VERB is used as infinitival complements of modals, where they occur in ‘suggested or recommended action’ settings, see (11). Uses about ability or future prospect often have less hortatory force, and they tend to foreground a strong ‘intention’ or ‘prospective motion’ reading, as in (12): 4

A constructional approach to formal constraints on go-VERB (11) a. You should definitely go check out the book. b. Ginny, perhaps you could go help your mother. (12) a. If he wants police reports, he can go get them. b. (Maybe I’ll go check the garage again first)

One isolated type of the go-VERB schema is the idiom-like go figure. It shares with many other uses a directive function and/or a negative-sarcastic evaluation, but differs in the inability to drop go to retain the meaning: (13) a. He achieved MUCH more, with MUCH less at his disposal. yet his recognition is practically zero. Go figure!! b. Cause in my head (yeah, I hear voices. go figure) it sounds awfully like Patrick Stewart.

The final use that needs addressing is the context that makes up the majority of indicatives. These uses are often representative and (occasionally) focus on habitual events (cf. (1a)). Consider (14) and (15): (14) a. Three times a week after school I go visit my dad. b. Say you’re a plumber - You go fix someone’s sink and then they say your quote is too high and don’t pay you. c. This does not excite the cyber criminals, so they just go play somewhere else. (15) a. You go tell Professor Evil I got a new class for him to teach – it’s called Who Framed Dan Rather 101, and I’m the star pupil. b. You go watch a film of a surgical abortion being done. I double dare ya!

Where hortatory force is entirely absent, as in the habitual contexts in (14), uses display a fairly strong focus on ‘intended’ or ‘(incipient) motion’ (over their uses without go). Some carry a negative overtone, such as annoyance with the attitude of customers (14b) or criminals (14c). The two examples in (15) illustrate a potential constructional coercion effect: they appear to be formally indicative (though (15b) is likely an imperative), but have a strong directive reading, in (15b) with particularly negative emotion. If we assume a hortatory constructional meaning of go-VERB, it seems that as the hortatory force is backgrounded, the element of (incipient and/or intended) movement is foregrounded, and hortatory readings can be coerced on indicatives. 3.2

Semantic clusters of go-VERB

While this pilot study is not meant to give an exhaustive or definitive semantic analysis of go-VERB (and no schemas are yet available), the following suggested clusters summarize our discussion above; some of them elaborate on previously suggested concepts and properties of ‘hortatoriness’ (Zwicky 2003) and ‘initialevent focus’ (Wulff 2006). The first three clusters below pertain primarily to the hortatory force of the motion or action involved, while the latter two focus more on connotation and evaluation. Most uses of go-VERB combine several of these characteristics, while some exhibit fewer (or are much harder to classify): 5

A constructional approach to formal constraints on go-VERB

−

−

−

− −

or REQUESTED MOTION/ACTION. Central cluster of requests, recommendation, appeals, or otherwise strong endorsement of ‘movement for action’. If strong(est) directive force is absent, weaker directive readings prevail. Prototypical imperatives illustrate cluster. PROMOTIONAL or FACILITATIVE MOTION/ACTION. Cluster with less directive force, and with stronger focus on encouragement, invitation, or concession to move for anticipated action. Prototypical let’s is part of the cluster, but imperatives and infinitival uses are more frequent. ANTICIPATED, INTENDED, or HABITUAL MOTION/ACTION. Heavily backgrounded hortatory force; intended action/motion is foregrounded. Incipient motion/event tends to be in focus over perceived necessity. Most indicatives fall into this category. DEROGATORY EVALUATION. Contributes a strong emotional evaluation including mocking, condescending, or sarcastic overtones. PROVOCATIVE INTENTION. Use types with insults, often involving sexual or violent action verbs in V2. It has an inherent, though indirect and non-literal directive element. SUGGESTED

With the exception of habitual and ‘pure’ anticipated or prospected action/motion descriptions, all uses include some hortatory or hortatory-like element. The strength of the hortatory force moves along a continuum from strongest directive, to various types of urging appeals or advisory remarks, to encouragements, suggestions, or invitations, and may often be accompanied by an emotional evaluation. The two elements that seem to unite all types— directives and representatives alike—are INCIPIENT MOTION, i.e., an inherent focus on the beginning of the action, and a slightly less salient focus on INTENTION TO PROCEED TO ACTION. Given the presence of a motion verb this observation may not seem too surprising as such. However, the salience of ‘motion’, ‘incipiency’ and ‘intention’ is inversely proportional to the intensity of the hortatory force. It should also be noted at this point that habitual uses, as in (14a)—the ‘model’ for the ‘ungrammatical’ examples (2a,b)—are extremely rare. 3.3

Preliminary discussion & hypotheses

If we assume the above classification as an approximation to go-VERBS’s constructional semantics, many puzzling aspects of the BARE STEM CONDITION seem to fall into place. In the absence of a sizeable amount of representative uses, the entrenchment of a strong hortatory constructional meaning appears to prevent go-VERB from being used in contexts that are subject to inflection. First, the hortatory core makes it far less likely for go-VERB to be used in situations other than that direct speaker-hearer interaction (excluding third person contexts). Second, it also makes it less likely to use go-VERB for making statements of intended motion on behalf of third parties. Third, use types that background the hortatory meaning by foregrounding the INCIPIENT MOTION reading are unlikely to attract progressive and perfect constructions for semantic incompatibility with their prototypical uses. Assuming hortatory semantics does, however, explain the unproblematic, or ‘grammatical’ use of go-VERB with 3.SG subjunctives (e.g. The police insist 6

A constructional approach to formal constraints on go-VERB

that Bill go talk to the counselor once a week), which Jaeggli and Hyams (1993:318) call ‘striking’. These (can) occur here not (only) because go-VERB satisfies an underlying morphological restriction, but because subjunctives are compatible directive functions. Thus, while non-finite contexts seem to ‘save’ otherwise ungrammatical uses with 3.SG contexts (e.g. she can go troll somewhere else vs. *she goes troll(s) somewhere else), they can also be accounted for by the fact that they correlate highly with hortatory structures (e.g., he told her to go see a doctor). Surprisingly few studies have focused on the meaning or function of goVERB (cf. Section 6). One of the notable exceptions is the collostructional approach taken in Wulff (2006:120–121), arguing that go-VERB’s constructional semantics is characterized by atelicity: go-VERB tends to focus on the ‘initiation’ or ‘initial phase’ of some action or event, as opposed to go-and-VERB (which tends to have a ‘dynamic event-like reading’, cf. Stefanowitsch 1999). This is compatible with the constructional semantics discussed above, in which the ‘incipient motion’ focus is a property that connects hortatory (urging the initiation of some action) and non-hortatory usage events. The hypothesis is thus that, if speakers violate the condition, we expect them to do so in cases where hortatory force is absent (or heavily backgrounded) and incipient or intentional motion/action is foregrounded. Thus, counterexamples are more likely to occur with simple past, but less likely in perfect or progressive constructions, as the latter are prototypically used for conceptualizing situations that have increasingly fewer of the core semantic properties that characterize go-VERB uses.

4

Data & method

As go-VERB is a rare phenomenon highly characteristic of spoken ‘face-to-face’ language (Zwicky 2003), sufficiently large amounts of data are hard to come by in written-medium corpora (Wulff 2006 identified 454 tokens of go-VERB in the BNC). Thus, the search for probably very rare exceptions in reliable amounts would likely be futile (and there are none in the BNC). Turning to corpora of web data with language from the informal end of the continuum is a viable approximation. However, a number of caveats must be dealt with, due to the rarity and syntactic quirkiness of the construction as well as the nature of web data. This section also briefly addresses the question of the potential skew in syntactic distribution hinted to in Section 2. 4.1

Data source, extraction & cleaning

This analysis is based on ENCOW14, a corpus of 16.8 billion tokens of web data (Schäfer & Bildhauer 2012). The association measures used in this section and the next are based on a 616-million token subset (ENCOW14AX01). I extracted the case-insensitive strings followed by what the tagger identified as noun or verb, but excluding strings ending in if longer than 6 characters (excluding go fishing, but retrieving go bring/sing).2 Instances of uninflected go-VERB can generally be identified semi-manually, but considerable manual work was required with verb-adjective ambiguity (updates go live tomorrow). By comparison, identifying instances of inflectional variants was very difficult and time-consuming due to the lack of punctuation (and she went like seen this?), ambiguous adjectival uses (the company went pay, ‘became a paid service’) or ‘typos’ (it goes show that), i.e. where it is not clear whether these are instance of {go}-VERB or {go}-to-VERB. Therefore, tokens were included as representing potential counterexamples if there was no clear contextual indication to the contrary. While this is a rather generous definition of ‘counterexample’, it was chosen as a deliberately inclusive approach to avoid a confirmation bias corroborating ‘received wisdom’ of ‘no exceptions’. Sections 5.1 and 5.2 will address qualitative and quantitative aspects of the systematicity of these putative counterexamples, respectively. After cleaning, a total of 11,447 instances of go-VERB, 436 of going-VERB, 32 of gone-VERB, 29 of went-VERB, and 12 of goes-VERB remained from ENCOW14AX01 (616m tokens). 4.2

Association measures

Assuming uninflected go-VERB as the central type of the construction, a simple collexeme analysis was performed on this set of data, producing log likelihood scores of attraction strengths of V2 to the construction (Stefanowitsch & Gries 2003; Gries 2007). As data extraction was string-based, the association measures are based on string frequencies (i.e. the association score for get as V2 is based on get’s string frequency, excluding got, gotten etc.). String-based collostructional analysis will underestimate the go-VERB types whose V2-verbs have considerably higher corpus frequencies as nouns (e.g., water or party), and vice versa. Note, however, that lemma-based calculations have an identical problem from the other end, ignoring skews in word-form distributions. String-based analysis is a more pattern-based approach that makes fewer prior assumptions, especially with respect to automatic annotation. Choosing log likelihood scores over the more widely used Fisher-Yates exact (FYE) p-value has the advantage that association measures for the most strongly associated types are different from 0. The idea, in other words, is to have an attraction measure for each V2 in go-VERB; these measures will be used to assess the systematicity of inflectional variants (cf. Section 5.2). Table 1 lists the 30 most strongly attracted verbs for go-VERB, along with their expected and observed raw frequencies. The goverall collexemic patterns discussed in Wulff (2006) are confirmed, with expected slight variation in ranks. 3 Throughout the entire list, there is a noticeable absence of telic, Excluding adjectives and adverbs here is probably a fairly safe guess: a preliminary inspection of followed by adjectives and adverbs yielded a precision rate of less than a 0.1%. Including ‘nouns’, on the other hand, is surprisingly sizeable: roughly a fifth of go-VERB occurrences have V2 tagged as a noun. 3 Items identified by Wulff (2006) as repelled collexemes—do, have, and say—are all attracted in the ENCOW sample (ranks 13, 36, and 141, respectively, of 483 significant types). While Wulff attributes the repulsion to the stative-vs.-action character of verbs and construction, the results here are not necessarily 2

8

A constructional approach to formal constraints on go-VERB

perception, or non-action verbs such as feel, believe, know, or think. A few (unexpectedly) repelled types are ‘victims’ of the string-based analysis, as these have much higher corpus frequencies as nouns (face, hand, or place). Table 1: 30 most strongly attracted verbs in go-VERB in ENCOW14AX01 go-VERB

(O:E)

LogL

see

(1226:10.7)

9338.8

fuck

(140:0.1)

1669.1

tell

(99:2.5)

536.59

get

(1197:12.8)

8622.8

watch (192:1.2)

1559.0

make (147:10.8)

497.8

check

(440:1.6)

4101.9

do

(406:28.8)

1408.1

hang

(54:0.3)

473.3

buy

(435:1.6)

4013.7

talk

(189:1.9)

1374.8

try

(91:2.8)

459.8

find

(485:5.9)

3341.6

pick

(134:0.8)

1092.2

work (142:11.3)

458.3

read

(417:3.7)

3125.2

take

(200:7.7)

919.7

vote

(63:1.0)

404.7

look

(424:5.1)

2932.8

sit

(109:0.7)

894.9

fetch

(34:0.0)

382.1

figure

(280:1.3)

2473.0

grab

(76:0.2)

751.8

meet

(68:1.6)

380.2

visit

(234:1.4)

1919.7

ask

(110:2.0)

668.0

kill

(56:0.7)

378.6

play

(253:2.9)

1762.2

eat

(89:0.9)

648.8

hide

(45:0.3)

368.8

4.3

go-VERB (O:E)

LogL

go-VERB (O:E)

LogL

Syntactic distribution

To quantify the hypothesized deceptive syntactic skew hinted at in Section 2 with regard to the assessment of ungrammatical uses against an infrequent type of go-VERB, a random subset of 1,000 tokens of the go-VERB set was coded for syntactic environment (Table 2). Go-VERB overwhelmingly occurs in nonindicative contexts, with a noticeable preference for imperatives (~38%), while indicatives constitute only slightly more than 5%. Although the preference of go-VERB for non-finite contexts is not at all a novel finding, its actual magnitude casts significant doubt on the centrality of the indicativerepresentative use type of (1a) above. The distribution hints at far more than an accidental or arbitrary preference for a certain syntactic environment—it is considerably skewed towards non-indicative contexts. Table 2: Distribution of go-VERB among sentence types

NON-INFLECTING

Syntactic environment Imperative Infinitival to-complement Modal complement Let’s-construction Subordination/embedded Interrogative DO-support (emphatic/negation) Indicative FIN Unresolvable by context/data Total

384 328 164 30 16 11 8 52 7 1,000

Example Go see a doctor! I told her to go see a doctor. You should go see a doctor. Let’s go see a doctor. First thing I did was go see a doctor. Did you go see a doctor? I did go see a doctor. I go see a doctor.

contradictory. First, Wulff’s analysis is based on a much smaller sample and corpus. Second, the verbs’ attraction is plausible assuming hortatory constructional semantics (e.g., Let’s go have a drink).

9

A constructional approach to formal constraints on go-VERB

4.4

Preliminary summary

The number of potential ‘violations’ of 509 tokens is very low, considering the huge amount of data (though this assumes that all violations are indeed instances of inflectional go-VERB; cf. Section 5.1). However, the syntactic environment skew towards non-indicative uses somewhat relativizes the rarity of exceptions. Great caution should thus be taken in writing these off as flukes a priori. Despite their relative infrequency, the next section discusses that the counterexamples fall into two distinct groups—those that are random noise in the data and those that are valid inflectional variants—but that the behaviour of both types are predicted by constructional semantics go-VERB.

5

Constructional semantics & constraint violation

The aim of (large-scale, statistically-minded) corpus linguistics is of course not to find the one counterexample for illustration. Isolated exceptions can always be found given sufficiently large corpora, and they don’t necessarily invalidate constraints. The more interesting issue is how far the exceptions are systematic and/or predictable. While the question of the validity of violations and exceptions can’t be answered by referring to individual example(s), patterns of systematicity may emerge once we look at the data collectively. 5.1

Inflectional go-VERB

Remember that the hypothesis was that constructional semantics make it more likely that exceptions occur with simple past, but less so with perfect or progressive constructions. Consider (16) and (17): (16) a. So u just gone let another man grab yo a * s when u married b. What are we gone do about it, now? (17) a. I’ll have knowledge that you are going attack me. b. Anyone going seeing Alexisonfire this week?

The problems with ‘perfects’ (16) and ‘progressives’ (17) are obvious. The examples with gone are highly likely not instances of perfects; they seem to be contexts where Standard English speakers use went (16a) or gonna/going to in (16b). While cases of going like (17a) are possibly typos of missing to (in going to); though (17b) could be an example of a very rare going-VERB. In the case of goes in (18) a go-VERB reading is possible; there is an incipient action element, similar to similar non-hortatory uses of go-VERB. But goes-VERB is overall extremely rare (12 tokens) with only a handful examples, of which (18) is the clearest one: (18)

Jefferson goes look for Coberly and he too disappears into the fog.

Went-VERB cases are, interestingly, more more plausibly classified as instances of inflectional go-VERB, particularly if they violate the IDENTITY CONDITION, too (which remains undeveloped in this paper). Compare the examples in (19) to 10

A constructional approach to formal constraints on go-VERB

those in (20); the latter could more easily be questioned to be cases of ellipsis, typos of missing and, and/or missing punctuation: (19) a. On this September visit I went see one of the teachers who has taught since the school began. b. After the war Baruchwoic who went live in Mexico and be a dentist. c. Pirates of the Caribbean 2 .. it was rubbish :( :o I went see it last night ... at 11 o clock :eek: d. But that setback did not deter Brown he went earn his Masters of Pepperdine University in California. (20) a. Sir Patrick’s on the case then it must be good, I reckon, so I went searched about for more details regarding this. b. Just went did this a few days ago.

Both types of went-VERB are interpretable as fairly clear violations of the BARE STEM CONDITION. And they occur where expected: the semantics are parallel to those of habitual-representative counterparts in the uninflected construction. They denote incipient motion to go see or visit something, anticipated change (of residence), or prospective motion or action (academic education). They focus on the beginning of motion or action. The major difference is of course that the point of reference is in the past rather than the present (or immediate future). 5.2

Violations, infrequency, and systematicity

The question is whether these counterexamples have systematicity beyond being (mostly typographical) flukes in the wealth of unedited, informal data. In fact, in many cases it is impossible to determine with reasonable certainty whether the token contains an accidental typo or whether the inflectional variant was intended. However, if there was no underlying systematicity, the inflectional variants should be distributed randomly. Thus, to assess the systematicity in some inflectional variants, the raw frequencies of the verbs that occur in V2 in the inflectional tokens where correlated with the respective verbs’ association scores in uninflected go-VERB (cf. Section 4.2) and with the V2 string corpus frequencies (goes was not tested as it only has 12 tokens/12 types): Table 3: Correlating constructions, frequencies, and association scores

Inflectional go-VERB type going-VERB gone-VERB went-VERB

N 436 32 29

F{go}-VERB~ LogLgo-VERB .17 –.12 .61

F{go}-VERB~ Fcorp .90 .49 .51

paired.r z = 19.14 z = 2.5 z = .52

p < .001 *** p < .05 * p = .61 n.s.

As predicted, going and gone correlate significantly more strongly with their respective corpus frequency than with their association score (LogL) in goVERB, suggesting randomly distributed occurrences. These cases, collectively, are probably due to the data rather than constituting systematic patterns.

11

A constructional approach to formal constraints on go-VERB

However, the reverse holds for went, where the raw frequency of the verbs in V2 correlate more strongly with their association score in go-VERB (r = .61) than with their corpus frequency (r = .51). While the difference between the correlation coefficients is not statistically significant (N = 29, z = .52, p = .61), the correlation points in the opposite direction than for going and gone. WentVERB thus seems to have a greater affinity to the collexemic pattern of go-VERB than to raw corpus frequencies, indicating an overlap distinct from random noise. This is consistent with the hypothesis provided by constructional semantics. To validate this pattern, the sample for went-VERB was extended. 5.3

Extending the sample

To test whether the trend with went-VERB can be corroborated, a further 490 went-VERB tokens were extracted from the full 16.8bn corpus using the same methods described in Section 4. In the extended set, the frequencies of V2 in went-VERB correlate best (and strongly) with their association scores in go-VERB (r = .82), but much less with corpus frequencies (r = .34); the difference in correlation coefficients is highly statistically significant (N = 490, z = 12.53, p < .001). This is taken to confirm the trend identified above in that went-VERB is not a random occurrence. However, this still does not fully address the possibility that examples like those in (19) and (20) are flukes in the data (especially typo-missing to or and). This cannot be ruled out in every case, of course. But it is unlikely to account for the pattern as a whole: the frequencies of V2 in went-VERB correlate less strongly with their association scores in went-and-VERB (r = .68) or went-toVERB (r = .62) than to the association scores in go-VERB (r = .82). 4 The strong(er) correlation with the and-/to-alternatives over corpus frequencies are of course expected given the close semantic relationship between them, especially between went/go-(and)-VERB. So while it can’t be ruled out that some went-VERB occurrences are due to confounding factors in the data, this is unlikely to be the case for the entirety of the data. Moreover, the correlation continuum paints a clear picture, in that went-VERB, though rare, is more than arbitrary and it occurs where expected by constructional semantics of go-VERB. 5.4

Preliminary summary

The data presented here can be taken as evidence to confirm the descriptive accuracy of the BARE STEM CONDITION at large. But two things are important to note: on the one hand, the likelihood of exceptions—both those that occur and those that do not—are predicted by the assumption of constructional schematic meaning discussed in Section 3.3, and follow a stable continuum. By the same token, consulting the ‘informal writing of millions of people’ (Liberman 2007) is not simply and eclectically fishing for isolated counterexamples—rigorous corpus linguistics can provide tools for the description of systematic patterning vs. random noise in large amounts of messy and unedited data. The association scores for went-and/to-VERB are string-based loglikelihood scores, based on ENCOWA14AX01, i.e. calculated in the same way as those for uninflected go-VERB (cf. Section 4.2). The difference between the correlation coefficients is highly statistically sigificant in both comparisons, i.e. went-VERB vs. went-and-VERB (z = 5.11, p < .001), went-VERB vs. went-to-VERB (z = 6.74, p < .001).

4

12

A constructional approach to formal constraints on go-VERB

6

Discussion

This paper has two major findings with respect to the go-VERB construction. First, while the construction shows a variety of sub-senses and meaning clusters, the construction on the whole has a highly entrenched core hortatory schema that is not predictable from the construction’s components (Goldberg 1995, 2006). Second, the evidence presented here suggests that constructional semantics and entrenchment, conceived of as pragmatic association with ‘external situations’ (Schmid 2014:251), i.e. in the context of recommendations, suggestions or highly emotional situations, can account for the BARE STEM CONDITION. In other words, go-VERB is simply less likely to be ‘useful’ in contexts that are subject to inflection in English. None of the observations, taken individually, are particularly novel. The preference for non-finite contexts has been mentioned by all authors (and from the results of the quantification of the syntactic distribution skew, it’s not very difficult to see why). In terms of the functional-semantic spectrum of go-VERB, Zwicky (2003) uses the term ‘hortatory’ referring to ‘sentence-initial’ uses (being mainly imperatives). The ‘volitional quality’ (≈ INTENDED ACTION) and ‘motion away from the viewpoint location’ (≈ INCIPIENT MOTION) was discussed in Pullum (1990:226) and atelicity and ‘initial phase focus’ (≈ INCIPIENT ACTION) observed by Wulff (2006). The strong distributional skew towards imperatives was described in Bachmann (2013) and Nicolle (2009); imperatives have also been suggested to be go-VERB’s historical origin (Nicolle 2007; Zwicky 2003, 2012). Finally, of course, the possibility to find inflectional variants in ‘the writings of millions’ is also far from surprising (Liberman 2007, on went-VERB). However, all these properties tend to be seen, implicitly, as something epiphenomenal of go-VERB, not as something inherent to, or explicable by, the construction’s schema itself (with the exception of the gist in Wulff 2006, 2008). So what this account adds is an approach that takes the entire bundle of formal, syntactic, semantic, functional, pragmatic, and distributional properties into consideration to explain the preponderance of the BARE STEM CONDITION— and ultimately to account for the rarity of its violation. This alternative perspective of go-VERB frames the construction and its constraint in terms of Langacker’s concepts of entrenchment, sanction, and ‘elaborative distance’ (Langacker 1987:66–69). Let’s assume that hortatory goVERB is the entrenched, fully schematic sanctioning unit—this holds both qualitatively and distributionally (cf. ‘syntactic skew’). The more some instantiations background the hortatory force of go-VERB, the more they move away from the central properties of the sanctioning schema. But as long as some specifications of the sanctioning unit are sufficiently present (such as incipient motion and/or intended action), the instantiation, or ‘elaboration’, is taken to be partially sanctioned (Langacker 1987:69). Therefore, the probability of finding ‘deviating’ structures, i.e. violations, is progressively a question of ‘elaborative distance’ from the sanctioning unit: finite and inflectional variants are increasingly rare(r), because they need to cover more and more semanticfunctional distance from the sanctioning unit. (I remain uncommitted at this point towards the question of how much the absence of inflectional forms can be entrenched, too.) 13

A constructional approach to formal constraints on go-VERB

The ‘elaborative distance’ continuum accounts for the observation of ample evidence for the ‘grammatical’ structures of the types in (1), i.e. less distance, but also for the grammaticality judgements for those in (2), i.e. too much distance. On the other hand, the continuum predicts that we find violations in contexts that have less ‘elaborative distance’ from the sanctioning schema, as with went-VERB (compatible with INCIPIENCY and INTENTION) than with goingVERB or gone-VERB (incompatible with almost all properties). The advantage of this approach is that inflectional variants and their infrequency can be formulated as a matter of probability of occurrence. In other words, violations don’t have to be judged ill-formed, dialectal, or oddities in masses of data. (Though the gist of the usage-based account leaves open the possibility that some speakers or communities have ‘microvariation’ due to different degrees of entrenchment with sanctioning and/or extended schemas.) The perspective of a cognitively oriented Construction Grammar on goVERB allows interesting hypotheses for the BARE STEM CONDITION in general. First, the key difference between go-VERB and go-and-VERB in formal properties is that the latter is not subject to the constraint. If the account presented here bears some generalization, then this difference could be due to the fact that the hortatory aspect is not as salient in go-and-VERB (thus inflectional contexts have less ‘elaborative distance’ to cover). Evidence in terms of syntactic distribution for go-and-VERB (fewer imperatives, cf. Bachmann 2013) and a different semantic focus (‘unexpected event and process readings’, cf. Wulff 2006; Stefanowitsch 1999) suggest that this might plausibly be the case. Second, the related come-VERB construction (Come visit me in Berlin!), which is also subject to the BARE STEM CONDITION, seems similarly skewed towards imperatives (Nicolle 2009). This could be indicative of a larger hortatory macro schema for VERB{go,come}-VERB. But why do many inflectional variants feel ungrammatical and why are most inflectional variants almost unanimously rejected? The question is probably how far the judgments are themselves influenced by go-VERB’s constructional schema and its entrenchment. In other words, it remains unclear what type of acceptability the judgments actually judge. If she goes gets the paper is rejected by most speakers, it might be because they judge purely formal properties. But under the account presented here, it seems more plausible to assume that what is actually judged here is the ‘elaborative distance’ from the sanctioning schema (which is assumed to be speakers’ entrenched representation). In other words, the morphological form, which happens to be bare in English, is likely just one of several aspects that influence this judgement. Put in terms of the methodological problems raised in Section 2, the indicative-representational contexts (which are themselves very infrequent) that serve as a point of reference for ‘ungrammaticality’ are already non-central uses once removed from the schema (i.e. partially sanctioned). Thus, the distributional skew in the conventionalized use of—and frequent experience with—uninflected go-VERB might not only account for the rarity of violations even in the largest corpora currently available, it could also explain the acceptability of some forms over others.

14

A constructional approach to formal constraints on go-VERB

7

Conclusion

Arguing from a constructional perspective, the present study has three major contributions to the explanation of the formal constraint in English serial verb constructions. First, go-VERB’s constructional meaning is decisively hortatory, centering around directives (with variably salient aspects of intentionality and incipient motion). Second, these schematic properties make go-VERB unlikely to be used in contexts that are subject to inflection in English (3rd-person; past, perfect, or progressive constructions). Third, infrequent violations of the constraint are predicted to occur in contexts that are at least partially sanctioned by the construction’s schema, and much less likely in contexts that are too far removed from the licensing schema. These predictions are borne out by empirical data. Based on an association measure approach, the analysis of go-VERB in web data confirms a systematic ‘inflection violation continuum’ identifying the theoretically most plausible went-VERB as also the (only) empirically systematic inflectional variant. The analysis separated the plausible from the less plausible violations, concluding that at least the former are neither accidental flukes in the wealth of messy data, nor have to be written off as necessarily (or inherently) idiosyncratic variation. While different types of data might produce different results in terms of varying frequencies of individual (sub-)types, it is probably safe to assume that the central claim of an entrenched hortatory core of go-VERB can be upheld to account for the rarity of constraint violations in general. While formal analyses attribute the lack of inflection to underlying morphological and/or syntactic parameters (ignoring functional aspects entirely), the usage-based perspective offers interesting insights from the opposite angle. It concludes, most importantly, that the BARE STEM CONDITION should better be seen as the result of constructional meaning, not as the underlying factor for the unacceptability of what would be very non-central, functionally inappropriate use contexts anyway. The results have direct implications for the analysis of go-VERB within a larger family of related constructions. The conclusions also have wider relevance, both theoretically and empirically, for questions pertaining to the relationship between formal constraints, grammaticality, and constructional meaning, as well as the interplay between language use (frequency) and schema representation (entrenchment).

Acknowledgements This paper is an extended version of a talk presented at the DGfS 2014 workshop on Web data as a challenge for theoretical linguistics and corpus design. I thank the workshop organizers, Roland Schäfer and Felix Bildhauer, for providing access to the full ENCOW data. I’d also like to thanks especially Anatol Stefanowitsch, Peter Uhrig, and Thomas Herbst for feedback and valuable comments on earlier drafts of this paper. The usual disclaimers apply.

15

A constructional approach to formal constraints on go-VERB

References Bachmann, Ingo. 2013. Has go-V ousted go-and-V? A study of the diachronic development of both constructions in American English. In Hilde Hasselgaard, Jarle Ebeling & Signe Oksefjell Ebeling (eds.), Corpus perspectives on patterns of lexis, 91–112. Amsterdam: John Benjamins. Bjorkman, Bronwyn M. 2015. Go get, come see: Motion verbs, morphological restrictions, and syncretism. Natural Language & Linguistic Theory. aop. doi:10.1007/s11049-015-9301-0. Carden, Guy & David Pesetsky. 1977. Double–verb constructions, markedness, and a fake co– ordination. CLS(13). 82–92. Goldberg, Adele E. 1995. Constructions. A construction grammar approach to argument structure. Chicago: University of Chicago Press. Goldberg, Adele E. 2006. Constructions at work. The nature of generalization in language. Oxford: Oxford University Press. Gries, Stefan Th. 2007. Coll.analysis 3.2. A program for R. Jaeggli, Osvaldo A. & Nina M. Hyams. 1993. On the independence and interdependence of syntactic and morphological properties: English aspectual come and go. Natural Language & Linguistic Theory 11(2). 313–346. Langacker, Ronald W. 1987. Foundations of cognitive grammar. Volume I: Theoretical prerequisites. Stanford, CA: Stanford University Press. Liberman, Mark. 2007. Ask Language Log: “Would you mind go checking on the Laundry?.” LanguageLog. http://itre.cis.upenn.edu/~myl/languagelog/archives/005028.html (16 October, 2007). Nicolle, Steve. 2007. The grammaticalization of tense markers: A pragmatic reanalysis. In Louis de Saussure, Jacques Moeschler & Genoveva Puskas (eds.), Tense, Mood and Aspect: Theoretical and Descriptive Issues, 47–65. Amsterdam: Rodopi. Nicolle, Steve. 2009. Go-and-V, come-and-V, go-V and come-V: A corpus-based account of deictic movement verb constructions. English Text Construction 2(2). 185–208. doi:10.1075/etc.2.2.03nic. Pullum, Geoffrey K. 1990. Constraints on intransitive quasi-serial verb constructions in modern colloquial English. In Brian D. Joseph & Arnold M. Zwicky (eds.), When Verbs Collide: Papers from the 1990 Ohio State Mini-Conference on Serial Verbs, 218–239. (Working Papers in Linguistics). Columbus, OH: Ohio State University, Department of Linguistics. Schäfer, Roland & Felix Bildhauer. 2012. Building large corpora from the web using a new efficient tool chain. In Nicoletta Calzolari, Khalid Choukri, Thierry Declerck, Mehmet Uğur Doğan, Bente Maegaard, Joseph Mariani, Jan Odijk & Stelios Piperidis (eds.), Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC’12), 486– 493. Istanbul: ELRA. Schmid, Hans-Jörg. 2014. Lexico-grammatical patterns, pragmatic associations, discourse frequency. In Thomas Herbst, Hans-Jörg Schmid & Susen Faulhaber (eds.), Constructions collocations patterns, 239–293. Berlin: De Gruyter. Stefanowitsch, Anatol. 1999. The go-and-VERB construction in a cross-linguistic perspective: Image-schema blending and the construal of events. In Dawn Nordquist & Catie Berkenfield (eds.), Proceedings of the Second Annual High Desert Linguistics Society Conference, 123–134. Albuquerque, NM: High Desert Linguistics Society. Stefanowitsch, Anatol & Stefan Th. Gries. 2003. Collostructions: Investigating the interaction of words and constructions. International Journal of Corpus Linguistics 8(2). 209–243. Wulff, Stefanie. 2006. Go-V vs. go-and-V in English: A case of constructional synonymy? In Stefan Th. Gries & Anatol Stefanowitsch (eds.), Corpora in Cognitive Linguistics, 101–126. Berlin: Mouton de Gruyter. Wulff, Stefanie. 2008. Das Prinzip der Nicht-Synonymität: V1-and-V2 und V1-V2 im Englischen. In Anatol Stefanowitsch & Kerstin Fischer (eds.), Konstruktionsgrammatik II: Von der Konstruktion zur Grammatik, 189–201. Tübingen: Stauffenburg. Zwicky, Arnold. 2012. QSV. Arnold Zwicky’s Blog. http://arnoldzwicky.org/2012/08/28/qsv/ (18 March, 2015). Zwicky, Arnold M. 2003. Go look at the modern language to test hypotheses about the past. Abstract. http://web.stanford.edu/~zwicky/lsaabst.qsv.pdf. 16

A constructional approach to formal constraints on go-VERB

17