Well, maybe you shouldn’t go around shaving poodles: collostructional semantic and discursive prosody in the go (a)round Ving and go (a)round and V constructions

1 Introduction

Consider the following examples:

While probably appearing to express negative evaluations of their own contents, neither sentence contains anything structural that can be said to overtly express this negative evaluation. However, in a construction grammar perspective, this could be due to the presence of the two constructions go (a)round Ving in (1) and go (a)round and V in (2) part of whose function seems to be the pragmatic expression of disapproval (Stefanowitsch 2000: 262). Although that is not the case in (1) and (2), we can assume that, if they indeed do serve this purpose, they might attract verbs in the Ving- and V-positions, respectively, which express situations that are in some way undesirable and more likely to be assessed negatively. This collocation-based phenomenon is known as semantic prosody within corpus linguistics. In connection with lexical collocation, semantic prosodies are identified by observing semantic patterns in the collocational field of the node word. Seeing that constructions are regarded as pairings of form and conventionalized meaning in construction grammar, constructions may also have semantic prosodies that might be identifiable by exploring what we can call the collexemic field of a construction (i.e., the collexemes that appear in a construction in a corpus).

With this in mind, collostructional analysis should be a helpful tool in the identification of constructional semantic prosodies (indeed, much collostruction-based research already involves semantic categorization of collexemes). Drawing on data from sections of the Corpus of Contemporary American English (henceforth COCA), this article addresses verb-construction interaction in the two constructions so as to explore the potential of using collostructional analysis in connection with semantic prosody. The article is, to some extent, a trial balloon as it involves traditional applications of simple collexeme and distinctive collexeme analyses as well as untraditional applications of standard and multiple distinctive analysis on nonlexical phenomena.

With this in mind, the reader should not take the findings presented in this paper as the final say on these two constructions, but rather as a few steps toward 1) a slightly deeper understanding of their usage and 2) developing a theoretically and descriptively relevant notion of constructional semantic prosody.

2 Usage-based construction grammar

In usage-based construction grammar, grammar is a nonmodular complex adaptive system (Beckner et al. 2009) of constructions (conventionalized pairings of form and meaning) organized at various levels of generality, specificity, and schematicity (Traugott and Trousdale 2014: 258).^[1] The system emerges inductively at a global level from actual usage at a local level. Constructions are thus chunks of conventionalized linguistic knowledge (Patten 2014: 94) and are perhaps best viewed as entrenched routines in a speech community (Croft 2005: 274). Heterogeneously distributed within the speech community, the system is incomplete in the individual speaker (Geeraerts 2017). Frequency of use is central to the conventionalization of a construction (but salience may override frequency), and the system contains not just structural information but also knowledge of contexts in which a construction is typically used. Hence, constructional meaning is both semantic and discourse-pragmatic.

Relevant here are the much-neglected notions of internal and external constructional properties (Fillmore 1988: 36–37). The former covers formal, semantic, and symbolic structures while the latter subsumes recurring contextual patterns. These are essentially association patterns – “the systematic ways in which linguistic features are used in association with other linguistic and non-linguistic features” (Biber et al. 1998: 5). Association patterns count co-textual elements like collocations, colligations, and collostructions as well as more discourse-pragmatic ones. They are arguably generalized reflections of local-level contextual regularities and part of speakers’ constructional knowledge.

External properties would also include semantic prosody, which has been described as a “consistent aura of meaning with which a form is imbued by its collocates” (Louw 1993: 157) or a “halo or profile … based on a semantically consistent set of collocates” (Bublitz 1996: 9). Less mystifying, Tognini-Bonelli (2001: 111) argues that speakers’ linguistic choices “will involve the more local and grammatical constraints around the word, but will also include perhaps more remote semantic preference and its correspondent on the pragmatic side, semantic prosody,” and Hunston and Francis (1999: 137) write that “a word may be said to have semantic prosody if it can be shown to co-occur typically with other words that belong to a particular semantic set.” Constructional semantic prosody, then, rests on emergent semantic patterns in the collexemic field of a construction.

Semantic prosody is contrasted with semantic preference (Partington 2004: 149; Tognini-Bonelli 2001: 111):

1. Semantic preference: specific semantic classes in the collexemic field. For instance, undergo prefers as direct objects nouns from the semantic domains medicine, education and assessment, and change (Stubbs 2001: 89–95).

2. Semantic prosody: abstract attitudinal patterns in the collexemic field. Since many of the semantic preferences of undergo are often associated with unpleasant experiences the verb has negative semantic prosody (Stubbs 2001: 89–95).

Semantic prosodies are divided into negative semantic prosody, positive semantic prosody, and neutral semantic prosody and are functionally akin to stance-markers: “[s]ince they are evaluative, prosodies often express the speaker’s reason for making the utterance, and therefore identify functional discourse units” (Stubbs 2001: 65). Hence, a construction with negative or positive semantic prosody can itself serve pragmatically as a marker of negative or positive stance.

The terms “semantic preference” and “semantic prosody” are unfortunate, as both phenomena are prosodic in that they extend beyond the individual unit (Stubbs 2001: 65). Moreover, only the former deals with semantics, while the latter deals with discourse-pragmatics. In this article, “semantic prosody” is used with reference to the former, while “discursive prosody” is used with reference to the latter.

3 The two constructions

Lexicographers appear to be aware that go (a)round Ving carries meaning beyond the words in the construction. For example, MacMillan offers the following definition: “to behave or be dressed in a particular way,” and the definition in Cambridge Learner’s Dictionary reads “to spend your time behaving badly or doing something that is unpleasant for other people.” The vocabulary resource PhraseMix includes the following description: “someone doing something bad again and again, with different people.” From such definitions, two general observations can be gleaned regarding what the construction expresses:

1. Some type of iterativity (which may be habitual)

2. Actions/behaviors associated with badness or unpleasantness

The unpleasantness does not have to be experienced by the patient only. It can also be experienced by the agent, as seen in the following examples:

Arguably, both actions are primarily socially unpleasant to the agent, being self-face-threatening acts. It is particularly interesting that, in (5), go around apologizing seems to be appended to the main message as a type of self-initiated self-repair serving to adjust the more neutral representation offered in the utterance. The following examples suggest that go (a)round and V, a subcategory of what Stefanowitsch (2000) calls the coordinated verb construction, might be functionally similar:

In all three examples, the actions are construed iteratively. In (5), arresting petty criminals is even construed as a habit of the agent, while (6) deals with a hypothetical mass shooting, which in itself, while not habitual, is iterative in that a mass shooting involves multiple shots fires at multiple victims. In (7), the act of bragging similarly comes across as a recurring action performed by the agent.

Formally, both constructions contain an inflectionally schematic go, which, together with (a)round, forms a symbolic unit with iterativity. In both constructions, this symbolic unit is presumably a metaphorical extension of the literal meaning of go around. Stefanowitsch (2000: 262) suggests that, in go (a)round and V, go (a)round draws on an image schema of motion along a nondirectional path which conveys aimlessness, intentionality, and habituality. Arguably, these features are compatible with iterativity, habituality even overlapping with it. In go (a)round Ving, the Ving-position is filled by a verb specifying the iterative action (henceforth, Ving). While Ving is inflectionally fixed, the V-position in go (a)round and V (henceforth, V) is schematic but inflectionally concordant with go. Like Ving, V specifies the iterative action. It was mentioned above that both constructions are used in the context of unpleasant or bad behavior. This is in line with Stefanowitsch’s (2000: 262) observation that go (a)round and V allows disapproval construals. We assume that this also applies to go (a)round Ving.

4 Data and method

This study draws on a 248,145,425-word portion of COCA containing the registers ACADEMIC, FICTION, MAGAZINES, MOVIES, NEWS, SPOKEN, and TV and limited to the period 2010–2019. Data were retrieved using these search strings: GO around _V?G, GO round _V?G, GO around and VERB, and GO round and VERB (these strings, it has to be said, exclude discontinuous instances of the two constructions). The search yielded 443 instances of go (a)round Ving and 101 instances of go (a)round and V.

Simple and distinctive collexeme analyses were performed using Gries (2022) with log transformation applied to all collostruction strengths. Simple collexeme analyses (Stefanowitsch and Gries 2003) were applied to both constructions. These yielded lists of collexemes ranked according to collostruction strength. The lists were subject to manual sense-based semantic categorization to account for semantic prosodies, which were then divided into the three discursive-prosodic categories. Specifically, in the semantic-prosodic analysis, verbs were divided into inductively established semantic classes based on sentence-level verb meanings; that is, the categorization was based on co-textually disambiguated senses, a consequence of which is that the same verb may have multiple semantic-prosodic values if multiple senses occur across instances of use.^[2]

According to the no-synonymy principle (Bolinger 1968: 127; Goldberg 1995: 67–68), constructions cannot be fully synonymous: small and seemingly intangible semantic, discourse-pragmatic, and sociolinguistic differences can occur systematically. Semantic differences may be reflected in distinctive collexemes which a standard distinctive collexeme analysis will identify (Gries and Stefanowitsch 2004a, 2004b). In this study, then, the two constructions were subject to a standard distinctive collexeme analysis.

Distinctive collexeme analysis has also been applied in the study of register-collexeme interaction by Schönefeld (2013) and diachronic year-collexeme interaction by Hilpert (2012). Both studies use multiple distinctive collexeme analysis – a variant that compares three or more constructions (Schönefeld 2013). Since nonlexical association patterns might serve as external properties of a construction on par with lexical ones, standard and multiple distinctive collexeme analysis were applied at the levels of colligation and speech acts, both of which may intersect with semantic prosody.

Colligation may be distinctive in (at least) two ways: 1) a collexeme might colligate with some categories in one construction and others in another construction and 2) one collexeme might cause a construction to colligate with some categories while another collexeme might cause it to colligate with others. To address 1), a standard distinctive collexeme analysis was applied at the level of the following manually identified VP categories:

1. simple present

2. simple past

3. present participle (i.e., verbal in ing-clause)

4. progressive aspect

5. excessive aspect

6. inceptive aspect

7. modality

8. mandative subjunctive

9. imperative

10. infinitive (i.e., verbal in infinitive clause)

11. controlled future tense

12. perfective aspect

To address 2), a Schönefeld-style (2013: 19, 24–25) multiple distinctive collexeme analysis was applied to either construction individually at the level of interaction between collexemes and verbal categories.

Speech acts can also be distinctive in (at least) two ways: 1) a construction might be preferred by one speech act type while others might prefer another construction and 2) a particular collexeme in a construction might attract some speech act types while another collexeme might prefer others. To address 1), a standard distinctive collexeme analysis was applied at the level of manually identified speech acts of utterances in which the constructions appear. The following speech act types were observed in the analysis:

1. statement

2. question

3. directive

4. commissive

To address 2), a Schönefeld-style (2013: 19, 24–25) multiple distinctive collexeme analysis was applied to either construction at the level of collexeme-speech act interaction.

5 The two constructions in use

It is worth noting that the two constructions are not used equally across the different registers of COCA. Go (a)around and V appears very frequently in SPOKEN (making up 46.53 % of all occurrences of that construction) and MOVIES (17.83 %), while go (a)round Ving appears frequently in FICTION (24.6 %), SPOKEN (24.15 %), MOVIES (20.09 %), and TV (15.8 %). Both constructions are infrequent in ACADEMIC, NEWS, and MAGAZINES. In other words, constructions are associated with spoken language with go (a)round Ving perhaps also serving in fiction as a stylistic representation of spoken language.

5.1 Collexemes, semantic prosody, and discursive prosody

5.1.1 Simple collexeme analysis

Table 1 lists top 20 attracted collexemes in either construction.

Table 1:

Top 20 attracted collexemes in both constructions.

Ranks	Go (a)round Ving		Go (a)round and V
	Collexeme	Collostruction strength	Collexeme	Collostruction strength
1	Say	216.5637	Say	31.0818
2	Kill	138.1451	Introduce	30.3402
3	Tell	136.2567	Tell	27.482
4	Pretend	108.3743	Round up	25.7867
5	Talk	102.4302	Ask	24.4004
6	Try	91.7626	Unscrew	16.8174
7	Blab	70.782	Personify	16.1235
8	Look	66.063	Try	16.0406
9	Ask	64.1325	Check	15.5419
10	Brag	64.0946	Shoot	15.202
11	Accuse	43.1593	Pick	14.4086
12	Shoot	41.0571	Buy	14.3165
13	Beat	28.9036	Batter	14.0515
14	Act	27.3377	Pay	12.7997
15	Murder	26.9148	Speak	12.7419
16	Prey	26.7351	Brag	12.5534
17	Point	25.8336	Scoop	12.4221
18	Save	24.8275	Spy	12.1216
19	Kiss	24.304	Tag	12.009
20	Imaginate	23.7016	Empty	11.9992

Many of these collexemes clearly have negative leanings. However, there are also a few with positive leanings, such as save and kiss, both of which are attracted to go (a)round Ving. Similarly, there is a range of arguably more neutrally leaning verbs – in particular in the collexemic field of go (a)round and V.

Imaginate also appears to be attracted to go (a)round Ving. However, this usage-event is a dis legomenon in COCA in its entirety, both instances occurring in one movie within the same dialog; therefore, it cannot be considered representative of the corpus. As (8) shows, it is clearly designed to be a neologism, used for humorous effect:

(8)

Now he’s going around imaginating anything he pleases.

Imagi … What now?

Imaginating. Verb, meaning “to imaginate.” I got an A+ in made-up words class.

So how do we stop the, uh, imagi-junk that hurts Gammy Gram?

(COCA 2014 MOV … t Wheels: The Origin of Awesome)

This anomaly aside, Table 1 suggests that go (a)round Ving attracts verbs with negative associations while go (a)round and V might have more neutral inclinations.

These simple collexeme analyses are useful in providing overviews of the constructions individually and serve as the basis of identification of semantic and discursive prosodies, which is why they included here. However, for the purpose of cross-constructional comparison, distinctive collexeme analysis is preferable.

5.1.2 Standard distinctive collexeme analyses

To provide a truly comparative perspective, a distinctive collexeme analysis was performed. Table 2 lists the top 20 distinctive collexemes of both constructions.

Table 2:

Top 20 distinctive collexemes of both constructions.

Rank	Go (a)round Ving		Go (a)round and V
	Collexeme	Collostruction strength	Collexeme	Collostruction strength
1	Kill	7.5329	Introduce	10.1764
2	Do	4.1501	Check	6.7678
3	Pretend	4.1501	Round up	6.7678
4	Think	2.4798	Batter	3.3758
5	Look	2.0961	Beg	3.3758
6	Accuse	2.0643	Bless	3.3758
7	Act	1.6498	Build	3.3758
8	Beat	1.6498	Bump	3.3758
9	Blab	1.6498	Clear	3.3758
10	Have	1.6498	Cut	3.3758
11	Hit	1.6498	Discover	3.3758
12	Open	1.6498	Empty	3.3758
13	Point	1.6498	Enforce	3.3758
14	Save	1.6498	Evaluate	3.3758
15	Use	1.6498	Examine	3.3758
16	Wear	1.6498	Gather	3.3758
17	Change	1.2361	Grab	3.3758
18	Fight	1.2361	Keep	3.3758
19	Kiss	1.2361	Knock	3.3758
20	Murder	1.2361	List	3.3758

Due to many tris, dis, and hapax legomena, many collostruction strengths outside the top 3 of either construction are identical (at least with four decimals), which on the face of it suggests that the degree of distinctiveness of most collexemes outside either top 3 drops. Figure 1 provides an overview of the association-frequency ratio, which allows us to identify collostruction strengths that should perhaps be taken with a grain of salt. The figure, in which go (a)round and V is represented by the positive side of the “Association”-axis while go (a)round Ving is represented by the negative side, seems to confirm that kill, pretend, do, think and accuse are genuinely distinctive and attracted to the latter construction.

Figure 1:

Association-frequency ratio in distinctive collexeme analysis.

At the level of collexeme types, again it certainly seems that there are a few verbs that express unpleasant or undesirable situations that prefer go (a)round Ving. As seen in Table 3, our semantic-/discursive-prosodic analysis suggests similarity between the two constructions as they both prefer negative and neutral prosodies.

Table 3:

Semantic and discursive prosodies.^a

Construction	Discursive prosody	Semantic prosody
Go around Ving	Negative	accusation, admission, ambush, apology, apology, appellation, apprehension, argument, assault, change, chatter, coercion, command, concealment, concern, conflict, crime, deception, denial, denouncement, destruction, division, entrapment, espionage, fear, gambling, gossip, harassment, ignorance, immodesty, immorality, injury, intrusion, irreality, jettison, misbehavior, mockery, predation, pretense, question, slander, snitching, threat, vengeance, vexation, violence, warfare
	Positive	affection, assistance, confidence, courtship, diplomacy, disinfection, donation, existence, greeting, honesty, ingestion, intercourse, manufacture, obtainment, ornamentation, pacification, permission, possession, rescue, resurrection, security, socialization, sports, study
	Neutral	action, announcement, anointment, apparel, appearance, appellation, argument, assumption, causation, claim, cognition, collection, communication, decision, diplomacy, discovery, effort, elucidation, empathy, equipment, expression, governance, grasp, inspection, litigation, locomotion, magic, manipulation, perception, photography, repetition, return, search, similarity, situation, song, touch, transaction, transportation, utilization, verbiage
Go around and V	Negative	admission, apology, apprehension, awakening, bribe, coercion, denial, destruction, espionage, harassment, immodesty, imploration, question, violence
	Positive	assistance, benediction, disinfection, game, greeting, manufacture, proffer, socialization
	Neutral	appellation, assessment, causation, cognition, collection, desire, discovery, effort, grasp, inhalation, inspection, manipulation, obtainment, perception, registration, schedule, search, selection, verbiage, transaction

1. ^aManipulation is not to be understood in the sense of social manipulation but in the sense of physical manipulation of objects.

Naturally, go (a)round Ving occurs with a wider range of semantic prosodies, but neither appears aligned with positive prosodies. While useful in mapping the overall semantic and discursive prosodies of constructions, this type of analysis does not account for preferential differences. An application of a standard distinctive collexeme analysis at the level of semantic prosody provides a slightly more fine-grained picture as shown in Table 4.

Table 4:

Distinctive semantic prosodies.

Rank	Go (a)round Ving		Go (a)round and V
1	Semantic prosody	Collostruction strength	Semantic prosody	Collostruction strength
2	pretense	5.8344	apprehension	12.0299
3	violence	5.4066	socialization	7.6394
4	action	4.1501	proffer	6.7678
5	accusation	2.896	disinfection	6.7571
6	change	2.4798	discovery	6.0778
7	threat	2.4798	manipulation	5.8507
8	appearance	2.0643	inspection	4.2473
9	donation	2.0643	assessment	3.3758
10	ingestion	2.0643	awakening	3.3758
11	misbehavior	2.0643	benediction	3.3758
12	rescue	2.0643	bribe	3.3758
13	affection	1.6498	desire	3.3758
14	assumption	1.6498	game	3.3758
15	conflict	1.6498	imploration	3.3758
16	crime	1.6498	inhalation	3.3758
17	utilization	1.6498	registration	3.3758
18	announcement	1.2361	schedule	3.3758
19	argument	1.2361	selection	3.3758
20	assault	1.2361	coercion	3.3525

In the top 20 distinctive semantic prosodies, more negatively inclined semantic ones such as violence, conflict, misbehavior, crime, and assault^[3] prefer go (a)round Ving than go (a)round and V. The latter appears more diverse overall, attracting more prosodies associated with social undesirability ones with physical harm. Table 5 summarizes the discursive- and semantic-prosodic analysis of the entire collexemic fields generated by the distinctive semantic-prosodic analysis.

Table 5:

Semantic and discursive prosodies based on distinctive semantic-prosodic analysis.

Constructions	Discursive prosodies	Semantic prosodies
Go around Ving	Negative (43 %)	accusation, ambush, argument, assault, change, command, concealment, concern, conflict, crime, deception, denouncement, destruction, division, entrapment, fear, gambling, gossip, ignorance, immodesty, immorality, injury, intrusion, irreality, jettison, misbehavior, mockery, predation, pretense, slander, snitching, threat, vengeance, vexation, violence, warfare
	Positive (22 %)	affection, confidence, courtship, diplomacy, donation, empathy, existence, honesty, ingestion, intercourse, ornamentation, pacification, permission, possession, rescue, resurrection, security, sports, study
	Neutral (35 %)	action, announcement, anointment, apparel, appearance, claim, cognition, communication, decision, effort, elucidation, equipment, expression, governance, litigation, locomotion, magic, perception, photography, repetition, return, search, similarity, situation, song, touch, transportation, utilization, verbiage
Go (a)round and V	Negative (32 %)	admission, apology, apprehension, awakening, bribe, coercion, denial, espionage, harassment, imploration, question
	Positive (27 %)	assistance, benediction, disinfection, game, greeting, manufacture, obtainment, proffer, socialization
	Neutral (41 %)	appellation, assessment, causation, collection, desire, discovery, grasp, inhalation, inspection, manipulation, registration, schedule, selection, transaction

Go (a)round Ving tends toward negative discursive prosody, as negatively leaning semantic prosodies account for 43 % of all of the semantic prosody types. In go (a)round and V, negatively leaning ones 32 % (vs. 41 % neutral ones and 27 % positive ones). Tables 4 and 5 suggest that go (a)round Ving furthermore attracts semantic prosodies relating to war, violence, destruction, emotional distress, and immoral and deceptive behavior while go (a)round and V seems to be more oriented toward socially unpleasant situations.

It is interesting that go (a)round Ving also attracts verbs with positive leanings. However, in such cases, the positivity of the collexemes is co-textually overridden:

(9)

We’re not sentimentalists. We don’t exchange heart-shaped boxes of chocolates or glossy cards with manufactured endearments inside, and we don’t go around kissing in public or saying “I love you” twenty times a day. (To my mind, couples like that are always suspect -- really, who are they trying to fool?) (COCA 2018 FIC NewYorker)

(10)

Don’t tempt him. I forgot. Baz, the white knight, goes around saving everybody. (COCA TV 2016 Animal Kingdom)

(11)

It would be a funny world if demons went around trusting each other. (COCA 2019 TV Good Omens)

While kissing, saving people, and trusting one another are arguably positive and socially desirable, they are presented in a more negative light here. Co-textual features – sentimentalists and suspect in (9), white knight in (10), and funny in (11) – suggest negative, maybe even sarcastic, assessment of these otherwise positive actions as undesirable types of behavior. This indicates that go (a)round Ving itself might prefer negative co-texts. Indeed, a simple manual utterance-level stance analysis shows that, overall, there is a preference for negative utterances of various types (such as expressions of disapproval, prohibitive directives, and undesired hypothetical situations) with negative co-texts constituting 89.84 % while positive and negative ones make up 4.74 and 5.42 %, respectively.

6 Final remarks

In this article, we have addressed – using collostructional analyses in both traditional and untraditional ways – the semantic and discursive prosodies of go (a)round Ving and go (a)round and V. It seems that the former has a preference for negatively leaning semantic prosodies such that its overall discursive prosody is negative. Consequently, it probably serves as a negative stance marker, thus aligning with a number of dictionary definitions of the construction. The latter construction seems a bit more diverse and less systematic. This might owe to the fact that only 101 instances were observed, but another explanation could be that perhaps go (a)round and V has not yet been constructionalized to the point that it can be said to be a strongly entrenched routine with a clear communicative function; to determine this, future research must measure the extent to which usage-events are idiomatic or not (but see Wulff 2007); a similar analysis could be applied to go (a)round Ving.

Our applications of standard and multiple distinctive collostructional analysis suggest that they are useful when isolating particular usage-factors and addressing various types of attraction patterns within those. This is undeniably valuable to analysts as it enables the analyst to provide detailed analysis of usage patterns. For the full picture, a multivariate analysis would be preferable, though, as it allows the analyst to take all factors into account. The ultimate strength, one could argue, would lie in the combination of multivariate analysis and more isolating collostructional analyses.

For now, it is hoped that this article has 1) shown that go (a)round Ving does indeed seem to have constructional semantic and discursive prosody serving as a marker of negative stance and 2) provided us with some theoretical notions and some potentially useful applications of collostructional analysis that can be utilized in the study and description of constructional semantic and discursive prosody.