In this chapter, I will give an overview of the semantic scope and evolution of the simple verbs. This will be done by accounting for their frequencies of occurrence (for the operationalization of semantic scope and specialization, see Section 4.2). While Section 5.1 provides the figures for present-day English (with reference to the frequency bands stated for each item in the OED), Section 5.2 opens up a diachronic perspective. By comparing the verbs’ frequencies of occurrence in the 19th and 20th centuries, we will be able to infer whether a simple verb has become more general in the course of time, whether it has specialized or whether it is hardly subject to any semantic changes. In one sense, Chapter 5 is therefore the first empirical chapter of this book. In another sense, the frequencies of the simple verb attested for in the 19th and 20th centuries form the basis for predicting how the CPs will evolve during the same time span (for the predictions, see Section 5.3). In that sense, they form the input for an investigation into the CPs’ processes of ‘specialization’ or ‘no specialization’.

5.1 Semantic coverage and frequencies of occurrence

In this section, we will be looking at the semantic coverage of the simple verbs that morphologically (and in many cases semantically) correspond to the CPs introduced in Section 4.1 (e.g. use in the case of make use of). This is a necessary first step in order to find out whether there is any correlation between the semantic scope and evolution of the simple verbs and the degree to which the CPs have become semantically specialized/restricted.

In Section 4.2, we have argued that the token frequencies of an item may be taken as a proxy of the number of senses that it covers. We will here test this claim against the simple verbs, providing both the number of senses that they cover and their frequencies of occurrence per million words. Both types of information are taken from the OED. In line with the procedure applied therein, each verb is assigned to a frequency band, ranging from 1 (with a frequency of nil) to 8 (with more than 1,000 tokens per million words; for more information, the reader is referred to Section 4.2).Footnote ¹ Since the OED does not give separate counts for prepositional verbs like provide for or account for, Tables 5.1 and 5.2 are limited to the non-prepositional verbs of our set.Footnote ² The tables are designed in the following way: the first column specifies the lexical entry, the second one the number of senses when obsolete ones are excluded, the third one the frequencies per million words (as calculated by the OED) in modern written English and the last one the frequency band assigned by the OED. We will start with the set that contains more lexical entries – that is, with those verbs that correspond to CPs involving take.

Table 5.1 shows that the results attest to a positive correlation between the number of senses that a verb contains and the frequency band that it belongs with. This correlation applies to all the verbs listed in Table 5.1: when we look at the extreme ends of the scale, we see that the verbs hold and leave have the highest number of senses (70 and 27, excluding obsolete senses) and range in FB 7. This contrasts with such verbs as advantage, heed, offend and pity, which have as few as 2–6 senses (excluding obsolete ones) and range in FB 5. The middle ground is taken by the verbs charge, notice and possess, which cover between 9 and 26 different (non-obsolete) senses and range in FB 6.

We will now turn to the set of verbs that correspond to CPs involving make to see whether the positive correlation between the number of senses covered by the OED’s entry for that verb and its frequencies in modern written English also applies to this set.

Notably, this set is less homogeneous than the one on display in Table 5.1. If we abstract from the outliers (answer, love, mention and use), however, we see a positive correlation between the number of senses assigned to an item and the frequencies with which this item occurs in modern written English. Take the verb fun, which contains no more than three different senses and, with no more than 0.07 occurrences per million words, ranges in FB 3. One of the rare examples of the verb to fun is provided in (1).

This contrasts with the verbs doubt, impress and sense, which cover between 8 and 10 different senses and belong with FB 6, occurring 10–20 times pmw.

For the verbs answer, love, mention and use, the relation between the number of (non-obsolete) senses and frequencies is not in line with our predictions. The most surprising candidate is certainly the verb mention, which, according to the OED, has no more than three different (non-obsolete) senses and yet ranges in FB 7. This contrasts with the verb answer, which has as many as 73 different senses and ranges in FB 6 (as opposed to FB 7 or 8). One explanation for what seems an ‘odd’ behaviour may be that the OED provides different meanings for different syntactic constructions, resulting (generally) in more meanings for verbs that have a plurality of different syntactic constructions than for those which do not allow for syntactic variance.

Let us illustrate this contrast for the verbs mention and answer. For one, the verb answer can be used both transitively and intransitively, while the verb mention is generally transitive (with the exception of the very rare construction mention about; see OED online). In addition, the verb answer can be used ditransitively (e.g. answer him the question), while the verb mention cannot. Focussing on the intransitive uses of the verb answer, we further see that it allows for different syntactic constructions internally (containing e.g. a prepositional phrase as in answer to the lady/to the letter).Footnote ³ It is therefore very likely that those verbs which allow for a series of different syntactic constructions also cover more meanings than those verbs which are more restricted in terms of their syntactic options. More research is needed to substantiate this claim.

In order to test whether there is a statistically significant correlation between the number of senses that the simple verbs cover and their frequencies of occurrence, I calculated the Pearson correlation coefficient for the 13 verbs listed in Tables 5.1 and 5.2, excluding the outliers answer, love, mention and use. The analysis provides me with the following correlation coefficient: r_P = 0.7561. This value attests to a strong positive correlation between the number of senses the verbs cover and their token frequencies. In order to see whether this correlation is statistically significant, I calculated the p value, resulting in p = 0.002786. This shows us that the correlation between the number of senses the verbs covers and their frequencies of occurrence is statistically highly significant (with p < 0.01).

Returning to the figures provided in Tables 5.1 and 5.2, the correlation between the number of senses and the frequencies of occurrence (here classified along the lines of the corresponding frequency bands) can be described as is illustrated in Table 5.3.

These findings allow us to operationalize an item’s semantic scope via its frequencies of occurrence: verbs that are semantically more general have higher frequencies of occurrence than verbs which have a more narrow semantic scope.

The verbs’ frequencies of occurrence presented in this section hold for modern written English (for the database used by the OED, see Section 4.4). In a next step, we will need to explore how each of the simple verbs behaves diachronically. Three scenarios suggest themselves: a) a simple verb was more frequent in previous stages of English than it is in the 20th century, b) the verb was less frequent in former times than in 20th-century English or c) its frequencies of occurrence have hardly changed. Section 5.2 will address the question of which scenario holds for each of the verbs investigated as we move from the 19th to the 20th century.

5.2 Frequencies over time

In order to decide which of the three scenarios holds for each of the verbs investigated, we will look at their frequencies over time, spanning the 19th and 20th centuries. While previous research on the history of CPs has primarily studied earlier time periods of English (for Old English, Middle English and Early Modern English, see the studies in Brinton & Akimoto Reference Brinton, Akimoto, Brinton and Akimoto1999), a look at the evolution of the CPs and their morphologically related simple verbs (e.g. take notice of – notice, make use of – use) suggests that the time span covering the 19th and 20th centuries is particularly promising when it comes to semantic and syntactic changes of the CPs. In addition, the retrieval of the data for the simple verbs is so extremely laborious that I had to limit myself to the investigation of only two (adjacent) centuries (for reasons discussed in Section 4.1).

For reasons of better comparability, the frequencies on display in this section will be based on the same corpora that will be used in order to study the evolution of the CPs in Chapters 6–8. These corpora are restricted to British fiction (see Section 4.4). In the case of the simple (and prepositional) verbs, the focus is on the 19th- and 20th-century data. In order for us to record possible changes in the frequencies throughout the 19th century, the 19th-century corpus is split up into its subcorpora NCF1, containing texts by authors born between 1728 and 1799, and NCF2, containing texts by authors born between 1800 and 1869 (for more information, see again Section 4.4). This level of granularity has been chosen here in order to be able to reflect any changes in the frequencies of the simple verb occurring in the course of the 19th century (which might go unnoticed if the 19th century was not split up into different time periods). The 20th century is then represented by the fiction component of the BNC (BNC wrid.1).

Since the simple verbs are much more frequent than the CPs (ranging in the thousands per corpus), the following method has been applied: 100 random tokens of each of the 24 verbs in use have been analysed manually in each subcorpus (i.e. 100 tokens for NCF1, 100 tokens for NCF2 and 100 tokens for BNC wrid. 1), excluding non-verbal or non-prepositional verb uses (but including prepositional verbs like pride in, which have been excluded from the analysis presented in Section 5.1).Footnote ⁴ Subsequently, the proportions of ‘correct’ tokens (subtracting the excluded items from the 100 tokens) have been projected onto the full set of examples (if e.g. 35 examples (out of the set of 100 tokens) are correct, then this results in 35% of the tokens of the entire set (e.g. 3,123 tokens) being included in the counts (which, in our example, results in 1,093 tokens)). As in the overview provided by the OED, the table contains all the senses that a verb (e.g. love) occurs in, irrespective of whether these allow for variation with the CP or not.Footnote ⁵ In those cases where we need to separate the verbal (i.e. participial) from the adjectival use (as e.g. in the cases of impressed, offended and possessed), I follow the distinction between category and function found in Huddleston et al. (Reference Huddleston, Pullum and Bauer2002).Footnote ⁶ The results are summarized in Tables 5.4 and 5.5, which provide both the normalized frequencies and the corresponding frequency bands for each verb in each corpus. The absolute number of tokens found for each verb in the respective corpora (NCF1, NCF2 and BNC wrid.1), the proportion of “valid” examples after non-verbal uses/non-prepositional verb uses have been excluded from the dataset, and thus the number of tokens in the cleared set are available from www.cambridge.org/compositepredicates. As in Section 5.1, we will start our discussion with those simple verbs that correspond to the CPs involving take.

As a first step, let us compare the data retrieved for BNC fiction (BNC wrid.1) to the data representing modern written English in the OED (as illustrated in Tables 5.1 and 5.2 in Section 5.1). Since the analyses presented in Tables 5.4 and 5.5 rely on British fiction exclusively, they will not be fully comparable to the data used by the OED, which, for the time period from 1970 to 2016, come from Google Books Ngrams and for PDE from a 20-billion-word corpus mainly covering online news sources from all major varieties of World English (see Section 4.4). It therefore does not come as a surprise that the outputs produced for 20th-century English in Tables 5.4 and 5.5 are similar in some respects compared to the ones laid out in Tables 5.1 and 5.2 but are different in others. Let us here look at some of the similarities and differences.Footnote ⁷

If we compare the individual simple verbs in terms of their frequency bands (the prepositional verbs cannot be compared for the reason that they do not occur in the OED), we find an identity in frequency bands (although not in the exact frequencies) for ten out of seventeen (non-prepositional) verbs, which is a proportion as high as 60%. The verbs are charge, heed, hold, pity, possess, doubt, impress, love, mention and sense. There are two verbs which do not occur at all in BNC fiction, namely advantage and fun. While it is not quite clear why the verb advantage is not found, the verb fun is also extremely infrequent in the OED, accounting for not more than 0.07 occurrences pmw. Given that, with a total of 20 million words, BNC fiction is much smaller than the database used for the OED (see Section 4.4), this finding is not surprising. The four verbs that have higher frequency bands in BNC fiction than in the OED are leave, notice, offend and answer, whose frequencies each rank one frequency band higher in BNC fiction compared to the OED. We can only speculate that these verbs are overrepresented in fiction. In the case of the verb leave, for instance, expressions like to leave (someone) alone/behind or to leave home seem to tie in nicely with the speech and behaviour of characters depicted in novels.Footnote ⁸ The verb use stands out in that it has about 600 occurrences pmw in the fiction component of the BNC, which is clearly lower than the 1,000 tokens per million words specified for the OED. Here, British fiction does not seem to explore the full semantic range covered by the verb in the OED.

In the following, we will now ask whether we can observe any changes in the normalized frequencies of the simple verbs as we move from 19th- to 20th-century British fiction. In order to answer this question, the results obtained for the two subcorpora NCF1 and NCF2 will be lumped together so that we can compare the entirety of the 19th-century fiction corpus to the 20th-century fiction corpus.Footnote ⁹ Once we have calculated the frequencies per million words, we will, for a better comparability, assign each item to the frequency bands used in the OED. Changes in the frequency bands to which a verb belongs are then considered to be more revealing than changes within a frequency band (for the simple reason that the scale is logarithmic, or, to rephrase the OED online, an item that belongs to FB 8 is ‘around ten times’ more frequent than an item that belongs to FB 7, which, in turn, is ‘around ten times’ more frequent than one that belongs to FB 6). The results will, subsequently, be assigned to the following four scenarios:

• Scenario 1: Scenario 1 contains verbs that show a decrease in frequency bands from the 19th to the 20th century (see Table 5.6)

• Scenario 2: Scenario 2 contains verbs that show no change in frequency bands and have moderate to low frequencies in the 19th and 20th centuries (ranging in FBs 1 to 6)Footnote ¹⁰ (see Table 5.7)

• Scenario 3: Scenario 3 contains verbs that show an increase in frequency bands over the same time period (see Table 5.8)

• Scenario 4: Scenario 4 contains verbs that show no change in frequency bands and have very high frequencies throughout (ranging in FBs 7 or 8) (see Table 5.9)

These four scenarios apply both to verbs whose corresponding CPs contain the light verb make and to those that contain the light verb take. The results for these four groups of verbs are summarized in Tables 5.6–5.9.

Comparing the lengths of these four tables, there is no doubt that most of the verbs in our dataset show no changes in frequency bands from the 19th to the 20th century (see Tables 5.7 and 5.9) but may vary internally as to whether they show a decrease in normalized frequencies inside their respective frequency bands (e.g. account for, delight in, offend or impress) or an increase (e.g. hold, leave or use). Assuming that there is a positive correlation between an item’s frequencies of occurrence and its semantic generality (see Sections 4.2 and 5.1), we can expect to find no changes in terms of the semantic scope of the verbs outlined in Tables 5.7 and 5.9. Yet, there are differences between the two types of verbs presented in Tables 5.7 and 5.9: in total, there are ten verbs that have frequencies in the range of FBs 1–6. Since their frequencies are in the lower to medium range, we can infer that they will be rather restricted in terms of their semantic coverage. On the other hand (see Table 5.9), there are a total of six items which show consistently very high frequencies of occurrence, ranging in FB 7 and FB 8 (the latter FB only applies to the verb leave). In line with the general correlation between high frequencies of occurrence and a high degree of semantic generality unfolded in Section 4.2, these verbs should be semantically very broad.

Although the tables that represent ‘no frequency change’ contain as many as 16 out of the set of 24 simple verbs, there are also verbs which show either a decrease or an increase in terms of their frequencies. Table 5.6 reveals that there are six items which show a decrease in frequency bands. To these belong the verbs advantage, heed, pity, possess, doubt and provide for. Assuming that decreasing frequencies are indicative of the semantic process of narrowing, these simple verbs should become more strongly specialized than they were before.

Finally, the shortest table, with no more than two items, includes those simple verbs which show an increase in frequency bands (see Table 5.8). The two items that meet this condition are notice and sense. For these verbs, we can assume to find an extension in terms of their semantic generality – that is, they should become much broader in semantic scope.

Against this overview of frequency-based changes in the simple verbs, we will now predict how the morphologically (and often semantically) related CPs should develop. More specifically, we want to use our knowledge of either the increasing or consistently extensive semantic scope of the simple verbs to predict which CPs undergo specialization/become restricted over time.

5.3 Predictions for CPs

On the basis of the frequencies assigned to the simple verbs in the 19th and 20th centuries (see Tables 5.6–5.9), I will now predict two different types of development for the morphologically related CPs which have at least a minimal semantic overlap: for those CPs which have simple verbs that either decrease in terms of their semantic scope (Scenario 1) or that are semantically (relatively) restricted throughout (Scenario 2), I expect to find no processes of semantic specialization in the CPs. For those CPs that have simple verbs which become more general in meaning (Scenario 3) or that are semantically very broad throughout (Scenario 4), I expect to find clear signs of semantic specialization or restrictedness. This means that I will treat the semantic evolution of the simple verb as the independent variable and the semantic specialization of the CP as the dependent variable. Only once I have attested to a process of specialization will I go one step further and identify the different types of specialization (see the different manifestations outlined in Section 4.3) and the degrees to which the CPs have become specialized.Footnote ¹¹

Table 5.10 illustrates the four scenarios outlined in Section 5.2, together with the predictions made for the corresponding CPs. In order to distinguish those CPs that have become semantically specialized/restricted from the rest, I will conceptualize Scenarios 1 and 2 as a setting in which there is no need for the CP to specialize. If the simple verb is not extending its semantic scope or if it does not have a (very) broad scope throughout (i.e. occurring in FBs 7 or 8, throughout), it is not necessary for the CP to become semantically specialized (because the meaning of the simple verb does not encroach on the meaning of the CP). This is different in Scenarios 3 and 4. Here, I predict that the greater semantic scope of the simple verb leads to a situation in which the CP adopts a niche function (see Section 3.4). In this case, the CPs are either semantically restricted throughout or they undergo specialization.

Table 5.10 Predictions made on the basis of four different scenarios

scenarios	development of simple verb	expected development of CP
Scenario 1	decrease in frequencies	no signs of semantic specialization
Scenario 2	moderate to low frequencies	no decrease in frequencies
Scenario 3	increase in frequencies	semantic specialization/restrictedness
Scenario 4	consistently very high frequencies	decrease in frequencies/consistently low frequencies

In addition to predicting settings of ‘specialization/restrictedness’ or ‘no specialization/restrictedness’ on the part of the CPs, Table 5.10 also specifies the operationalization of semantic change. Semantic change is obvious from changing frequencies of occurrence, with semantic specialization showing in decreasing frequencies and semantic restrictedness in consistently low frequencies of occurrence (see the information provided at the end of the arrow).

The results on display in Section 5.2 have revealed that only a small minority of the simple verbs analysed show an increase in terms of their frequency bands. If we follow Scenario 3 and take increases in the normalized frequencies of the (simple) verbs to be indicative of their increasing semantic width, then this process of semantic widening can be expected to hold for the two verbs notice and sense (see Table 5.8). We can, accordingly, expect the CPs that correspond to these verbs, namely take notice of and make sense of, to show clear signs of semantic specialization or restrictedness.Footnote ¹² This scenario contrasts with those CPs that have simple verbs which decrease in terms of their semantic scope (see Table 5.6) or which are semantically restricted throughout (see Table 5.7). For these CPs we expect to find no indication of semantic specialization.

If we now turn to Scenario 4 and assume that those verbs which are in FBs 7 and 8 are more likely to have CPs which are semantically specialized or restricted than verbs that are assigned to lower frequency bands, then this prediction should hold for the six verbs hold, leave, answer, love, mention and use. For the corresponding CPs take hold of, take leave of, make answer to, make love to, make mention of and make use of, we expect to find clear indications either of semantic specialization or of semantic restrictedness in the 19th and 20th centuries. Note that Hypothesis 1 states that this prediction only applies if there is (at least) a (minimal) semantic overlap between the simple verb and the CP (for the list of CPs that have a semantic overlap with their simple verb as compared to those without, see Table 4.3 in Section 4.1). We will return to the question of semantic competition once we turn to the analyses of the individual CPs in Chapters 6 and 7.

Table 5.11 summarizes those eight CPs for which we expect to find processes of semantic specialization and/or a state of semantic restrictedness in the 19th and 20th centuries (again, provided they compete with the CP in semantic space). The table is so arranged that CPs involving the verb take are placed in the left column and those involving the verb make in the right column.

We will test the prediction of semantic specialization/restrictedness of the CPs in Chapter 6, answering the questions a) if the prediction is confirmed, b) if specialized, how strongly specialized the CPs are and c) in what ways the CPs become specialized (i.e. which type(s) of specialization occur(s)). For all the remaining CPs (with their corresponding simple verbs summarized in Tables 5.6 and 5.7), we expect to find no signs of semantic specialization.Footnote ¹³

In order to distinguish between the two types of CPs, we will, from now on, refer to those CPs that have simple verbs which are extending their semantic scope or which have an extensive semantic coverage throughout (ranging in FBs 7 or 8) as Type I-CPs and to those CPs which are decreasing in terms of their semantic scope or do not have an extensive semantic coverage as Type II-CPs.Footnote ¹⁴ We will start with the analysis of Type I-CPs, summarized in Table 5.11.