A Descriptive Survey of the Character of English Lexis in Sermons

This study investigated language use in the context of Christian sermons using the sermons of Pastor Chris Oyakhilome as a case study. The aim was to examine the lexical characteristics of language in context to establish whether or not the character of English lexis is determined by linguistic context or co-text. It therefore sought to discover whether there is a network of lexis peculiar to sermons: definite words associated with them, specific patterns of association or collocation, and peculiar meaning relations. In doing this, a corpus-computational technique was adopted in which 200 actual sermons of Pastor Chris Oyakhilome were selected, built into a corpus, then computer processed and compared with a reference corpus of contemporary English, which was chosen as a measure of normality. The results show major differences: Those words that were unusually frequent, and therefore the most significant in the sermons, were found to be unusually infrequent and therefore insignificant in general English, showing that the key lexis of the sermons is different from the key lexis of general English. This finding was strengthened by the differences recorded in the collocation patterns of the words selected for detailed examination in both contexts, and by the variations in their semantic relations.


Introduction
The need to investigate the behavior of English lexis in the specific context of religion is echoed by Crystal and Davy (1969), and Crystal (1995), who point out that the collocational idiosyncrasies which occur in religious English provide a discussion point in their own right. Claridge and Wilson (2002) also note that the sermon genre is interesting because it occupies the crossroad of orality and literacy. Yet sermons, as communicative events, have so far attracted very little linguistic attention. The paucity of research information on the linguistic significance of sermons is clearly evidenced by the minimal literature about it, as will be shown in the literature review that will follow shortly. So although mention is made of the distinctiveness of this genre, little effort has been made to investigate this claim, using a sermon corpus against the background of a reference corpus, to reveal specific differences in the character of English lexis. This investigation-of the behavior of English lexis in the context of Christian sermons-will serve the primary functions of illuminating this particular language event, enhancing its understanding and contributing to our knowledge of language as a system. clauses that follow a given word (Hunston & Francis, 2000). A pattern is closely associated with meaning in that different senses of words are distinguished by their patterns of occurrence, and words that share a pattern tend also to share a meaning. This point resonates in O'Donnell (2005), in that meanings and functions in language are found in repeated patterns of words and phrases. Hoey (1991) also believes that the concept of repetition is fundamental to language patterning. Phraseology, then, can refer to the grammatical pattern that belongs to a word, and every word has its own pattern. A parallel idea to the one above is the phraseology of lexis, lexical patterns that account for the combinatorial tendency in language (Lyons, 1981). This means that most language is a construct, not from "basic" structures and the lexicon, but from sequences of prearranged, preconstructed or formulaic language (J. Sinclair, 1998Sinclair, , 2004. Finally, the pattern of a word shall be all the words and structures that are regularly associated with the word and which contribute to its meaning.

Collocations
As is well known, the term collocation is due to Firth (1951 who used it to refer to the level of meaning, distinguishing it from both the conceptual and contextual meanings through his famous slogan: "You shall judge a word by the company it keeps." Collocation has been variously defined as a relation of the probable co-occurrence of items (Malmkjaer, 1991), "actual words in habitual company" (Firth, /1968, the co-occurrence of two or more words within a short space of each other (J. Sinclair, 1991), the associations a word acquires on account of the meanings of words in its environment (Leech, 1981), the relationship a lexical item has with items that appear with greater than random probability (Hoey, 1991), and the habitual associations of a word or the co-occurrence of two single words (Partington, 1998). What resonates from all the definitions is the idea of association-that words prefer the company of some words rather than others. Halliday (1966) and J. M. Sinclair (1966) took Firth's idea further by introducing the notion that patterns of collocation can form the basis for a lexical analysis of language that is alternative to and independent of grammatical analysis (Gabrielatos, 1994). Other important concepts in lexical studies are semantic prosody and preference which are discussed below.
The concept of semantic prosody. Stubbs (1996Stubbs ( , 2002 presents a phenomenon of collocation (identified by J. Sinclair, 1991, andnamed by Louw, 1993), the concept of semantic prosody, as referring to a situation in which some words associate with either unpleasant things/states (negative prosody) or with pleasant things/states (positive prosody), as well as with definable semantic sets. He notes that "there are always semantic relations between node and collocates and among the collocates themselves." McEnery (2006) adds that the meaning that arises from the interaction of a given node word with its collocates is its semantic prosody and observes that semantic prosodies are predominantly negative. Related to semantic prosody is the concept of semantic preference. J. Sinclair (2004) defines it as "the co-occurrence of words with semantic choices" (p. 174). He explains that semantic preference is the restriction of regular co-occurrence to items that share a semantic feature, and that this implies the identification of similarity of meaning across word boundaries. Therefore, semantic preference is the likelihood that words that share meaning domains will co-occur. According to Sinclair, the semantic prosody of a word is its dominant semantic preference.
To summarize, studies on English lexis have revealed the new role that lexical study is called upon to play in language description by showing its interconnectedness to other aspects of language and centrality to the organization of meaning. Thus, its significance in the explanation of language is no longer in doubt. Crystal and Davy (1969) note that collocational idiosyncrasies which occur in religious English provide a discussion point in their own right, because certain high-frequency collocations that occur, apart from forming the distinctiveness of the language, are quite different from the collocations for these items found elsewhere. This finding serves as justification for the present attempt to investigate the lexical nature of sermons. Thompson (2003) also examines the existence and nature of religious language and points out the existence of a specific use of language to express or describe religious experiences, practices, or beliefs, that is, to communicate religion. Samarin (1976) discusses religious language from a sociolinguistic perspective and observes that religious communities are set apart linguistically, and notes that it would be interesting to reveal the uniqueness of their use of language. Stiver (1996) too examines the significant role of language in religion which made it a prime focus in 20th century philosophy and argues that there has been a detour from metaphysics to language, what he calls a linguistic turn. This concern for language, he explains, is due in large part to the conviction that the subject of religion is one that is difficult to express, and that the power of words plays a crucial role in creating its realities. Crystal (1990) also examines liturgical language from a sociolinguistic perspective; he observes the shift in linguistics from the study of language structures to its functions, and notes that it is only by focusing on language functions or uses in different contexts that a convincing and coherent account of language can be provided.

Religious Language
In sum, the works examined reveal that religious language embodies useful insights into language and that unless this is adequately tapped, a complete account of language is impossible. In the next section, the method which is used in answering the research questions is detailed.

Method of Study
As a methodological basis, the study utilizes a corpusresearch method. The four questions that it considered are as follows: (a) Size, what proportion of language would be adequate for the corpus? (b) Content, what kinds of text should be included in the corpus? (c) Sampling, how can it ensure that the sample accounts for every part of the population? (d) Representativeness, will the corpus adequately represent the population it was drawn from? We will now discuss each step below.

Population
In view of the problems associated with taking accurate samples of language in use, Biber's (1999) suggestion of setting limits for a study's population and ensuring the inclusion of the full variability of texts in the population was a welcome relief. So the population of the sermons, for the purpose of this study, was defined at 800, as contained in selected sermon publications of Pastor Chris Oyakhilome, a high profile pastor in Nigeria. These served as the population frame because determination of the overall quantity of his sermons is unfeasible. These sermons belong in the Pentecostal fold; this is a major characteristic of this population because it is likely that the structure of sermons may differ according to denominational traditions. In the publications, there are 25 dominant topics. These became both the basis for the stratification of the sermons into text types and the guide for sampling.
Sample. Because of the nature of sermon texts, it was impossible to take equal samples in terms of number of words as the sermons were of unequal lengths. So the criterion of sampling full sermon texts, which Halliday (1991), J. Sinclair (1991), and Biber, Conrad, and Reppen (1998) recommend for the investigation of textual and association patterns, was adopted. Also, the selection of full sermons was necessary to widen the scope of search, because features are not normally evenly distributed in texts, they may cluster in some parts and be absent in others, thereby giving room for bias in the analysis. However, it was possible to sample equal numbers of texts, eight per topic. In this way, it was possible to achieve fairly balanced samples of the sermons under investigation, amounting to 200 sermons and 64,851 words.
Sampling technique. The method used to select the sermons is random stratified sampling of full sermon texts. This entails division of the population into homogeneous groups (strata), and then samples of each stratum are taken at random. The corpus of sermons built for the purpose of this research is composed of all categories of texts found in the overall population. The population of 800 sermons was stratified into 25 categories or fields, each composed of approximately 32 sermons, amounting to 800 sermons. Samples were then taken from each category. Each was saved as a text file representing a particular subtext type and named after the theme(s) it embodies.

Data Collection
As in any corpus-based study, the data are represented by the corpus to be investigated, and its reference, where one exists. In the study, there was a need to design a corpus because no such corpus was available for use. So a corpus of the sermons being investigated was built. For comparability, a reference normative corpus was needed as a standard against which to assess the corpus under study. The British National Corpus (BNC) was selected for this purpose because it is an adequate measure of contemporary British English. This study utilized a smaller version of the BNC, called the BNC sampler, consisting of two million words. This version, which contains 50% written and 50% spoken English, prepared from the larger BNC as a representative sample of British English, is ideal for this work because of both its size and composition. This is because it has some affinity with the sample text-the sermons-which themselves are written to be spoken and so may exhibit features of both genres.

Data Analysis
The analysis of corpora demands a partnership between humans and machine (Leech, 1991). The corpus method adopted for this study presupposes a quantitative approach, so some of the analyses were done on computer, using appropriate queries. However, the tasks required the expert knowledge of the researcher to verify data and, ultimately, make qualitative judgments. Basically, the following quantitative analytical processing systems were applied. And then, qualitative analysis for inferences followed.
Frequency counts. This involved first the assignment of items to particular categories, and then an arithmetic count of the number of items (or tokens) within the texts that belong to each category. In order not to create a false impression, beyond a frequency count, calculation of the frequency of occurrence of a type (item) as a percentage of the total number of tokens in the corpus, that is, the total size of the corpus, is imperative. Thus, proportional statistics presents a better approach to frequency count.
Tests of significance. Following the frequency count and the arrival at proportional statistics comes the need to ascertain that the finding-the ratio of occurrences of the type to the total tokens in a sample-is genuine and not a result of chance. To determine this, a test of significance was carried out; this determined how low or high the probability is that the difference between two texts on the type feature is due to chance. The log-likelihood was used to compare the difference between the actual frequencies observed in the corpus (observed frequency) and those expected if only chance had affected the frequencies (expected frequency).
Significant collocations. As a major objective of this study is the identification of the structural patterns of words-collocations in particular-yet another statistical measure was required to identify which co-occurrences are significant. The formula of mutual information was used. The mutual information score between any given pair of words compares the probability that the two occur together as a joint event (because they belong together) with the probability that they occur individually and so their co-occurrence is merely a factor of chance (McEnery & Wilson, 1996/2001. Therefore, pairs of words with high positive mutual information scores are more likely to constitute characteristic collocations than pairs with lower mutual information scores. The mutual information was preferred here only because it is compatible with the WordSmith software tool chosen for the study.
Analytical tools used. The study adopted the computer software tools WordSmith 5 and Wmatrix. The reason for multitool application is twofold: to enable comparability and authentication of results, and to choose the most efficient tools in each. The statistical tools employed are proportional statistics, log-likelihood statistics, and mutual information score. We shall now proceed, in subsequent sections, to discuss the results obtained from the study.

Discussion of Results and Summary of Findings
Here, the questions posed in the study and the results gained from analyses of the data corresponding to each of the investigations will be examined in detail, to uncover any insights there might be. We shall begin our discussion of the results by first restating the questions posed in the study and briefly stating how the answers were obtained. Then, the results relating to each objective of the study will be examined in relation to the question it sets out to answer. The study sets out to provide answers to the following questions:

Objective 1: To Identify Lexical Items Associated With His Sermons
To fulfill this objective, Scott's keyword method was used. This is a text-analysis method that effectively differentiates the lexis of one text from another or the style of a text from another. To find out what is peculiar to sermon texts must entail distinguishing them from English usage outside of the sermon context. This is why the sermon corpus needed to be compared with a reference corpus of general English. This comparison, using the keyword tool in Wordsmith 5, produced those words that are more frequent in sermons than in everyday English, these are called keywords. The results were confirmed by using Wmatrix software. We then used the Wordsmith software to produce lexical items beyond single word forms. These, we called multi-word items, word combinations that are bound by either syntax or semantics or both, such that they are largely unalterable. So the answer to the question-Is there lexis specific to the sermons?-seems to be yes. We shall return to this answer shortly, to justify it.

Objective 2: To Describe the Patterns of Words in the Sermons
To accomplish this objective, we had to identify collocates for the identified keywords of the sermons. So each of the keywords selected for further analysis was examined using the concordance tool in Wordsmith 5 to find those words that routinely occur with them. Subsequently, the same words were examined within the context of ordinary English to find which words are regularly found around them. These two sets of words-those that collocate with a given search word in the sermons and the ones found in general English-were then compared to identify what differences there are in their word associations. The same process was followed in the determination of the patterns of collocation, first for the sermon words, followed by the general English word patterns, and then both were compared to reveal their differences. By patterns we mean lexical patterns-recurrent word combinations. Once again, the answer to the question-Do sermon words exhibit specific patterns?-was found to be yes. This answer too will have to be proved in a little while.

Objective 3: To Analyze the Semantic Implications of Words in the Sermons
In fulfillment of this third objective, the meaning that each search word contracts from words in its immediate environment (co-text) in the sermon was analyzed and compared with its meaning in contexts other than sermons. This investigation was done using the frameworks of semantic preference and prosody, as proposed by J. Sinclair (1991) and Louw (1993). These enabled the establishment of associative, pragmatic word meanings, which according to Lewis (1997) may differ radically from literal meanings. This notion supports Firth's assertion that meaning is inherent within the partnership of words on the syntagmatic plane of language. So as words share physical space, they tend not only to share meaning but also to acquire other meanings from their co-texts. This was confirmed by the results obtained. Therefore, our third question-What are the semantic relations of words in the sermons?-is also answered.
In the next section, we shall discuss in greater detail, objective by objective, these results to see the extent to which the lexis of the sermons differs from the lexis of general English by describing their unique characteristics. This will answer the main question posed in the study and thus accomplish the goal of the research.

Results for Objective 1: Lexical Items Associated With the Sermons
The investigation revealed both words and multi-word items that are more likely to occur in sermons than in general English. These are words and multi-word items that have unusually high occurrence rates in the sermons and correspondingly low occurrence in general English. These keywords and key multi-word items make up the lexical content of the sermons. The results pertaining to the single lexical items will be discussed first, while discussion of the multiword items will occur later.
The study portrayed a total of 469 keywords with varying degrees of keyness and thus varying levels of significance. It was necessary to select a smaller sample for detailed analysis, so a cut-off point was set, based on the level of significance. All the keywords with a p value of 0 were chosen; this represents far less than a 1% error level. In the sciences and even social sciences, as high as a 5% error level is allowed as proof of the validity of results. The selection gave a total of 269 keywords; all of these were positive keywords in the sense that they were relatively overused in the sermons. Then, others were found that were negative, that is, they were highly underused in the sermons. Both groups constitute keywords because their frequencies of occurrence are similarly unusual, it is either that the frequency of use is grossly above or below normal expectations. The positive keywords represent the core lexis of the sermons, while the negative keywords mark general lexis which contrasts with the lexis of sermons and provide support for the establishment of the peculiarity of sermon words. But before discussing these keywords, the 100 most frequent lexes used in the sermons and those used in ordinary English were examined so that it will be clear what differentiates and qualifies those considered to be key and peculiar to the sermons. So, following here is, first, discussion of the top 100 generally highfrequency words, and then the top 100 most unusually high-frequency words follow. So the contrast is between normal and unusual frequencies to indicate lexical differences.

Discussion of the Most Frequent Words Found in the Sermons and General English
It is important to note that the keyword list and most frequent word list present no clear differences. They are simply words in common use in their domains and so their high frequencies are to be expected. This establishes the existence of a general or core vocabulary in contrast to a specialized vocabulary. Speaking of this kind of vocabulary, Stubbs (2002) notes that it presents a useful way for comparing texts by calculating what percentage of words from the core vocabulary they contain. Core vocabulary refers to that portion of language which its speakers cannot do without. Usually, these are the most frequent words in a language; they must also be evenly distributed across a variety of text types. When we speak of the size of individual vocabulary, these constitute a large part, while some others are somewhat peculiar to the speaker, either in terms of frequency of use or the personal choices of the user. In this study, core vocabulary is represented mainly by function words. Although the lexicon cannot be measured, Ballard (2007) has noted that words can be safely divided into lexical and functional types, and function words are known to have higher frequency because they are the mortar that binds texts together. They indicate the relationship between lexical units and other larger units in language-hence their frequent occurrence in texts. But as Biber, Johansson, Leech, Conrad, and Finegan (1999) note, their frequency is regulated by text types; some texts seem to require greater use of them than others, while also exhibiting preference for certain functional classes over others. When we compared the 100 most frequent words found in both corpora, we confirm that function words are more frequently used, and there is a reason to support preferential uses of certain classes of function words. See Table 1, Figures 1 and 2 below.
In Figure 1, we have a seemingly higher use of function words in general English in all categories. This is a misleading result because it is based on the raw frequencies without regard to the size of the data. However, we find in Figure 2 that, except for the modal auxiliaries, the functional classes are much more productive in the sermon texts than in general English. This confirms that relative frequency is a significant measure of the variability of linguistic features across text types. What, therefore, does the greater use of function words portend for the sermon texts? The basic function of grammatical words is to indicate relationships between lexical words and other larger units in language; they serve to bind texts together. So the higher frequency of function words in the sermons points us toward expectation of equally high occurrence of lexical words and greater textual cohesion in the sermons. If this is the case, then the sermon texts are expected to manifest higher information content than general English.
We shall now examine the keywords derived from the study to identify the types of lexical words they represent, and then later calculate what proportion of the texts they constitute. This will enable establishment of their lexical densities and lead us to prognosticate how much information load each text can carry. The lexical density of a text is the amount of the text made up of lexical word tokens. We shall consider first the four main classes of lexical words-nouns, verbs, adjectives, and adverbs-before measuring their densities.

Discussion of the Most Unusually Frequent Words-Keywords
It will be recalled that keywords refer to words that are statistically significant in the sermons, based on their unusual frequencies in contrast to ordinary usage. We identified two types, positive and negative, those whose frequencies are more than expected and those whose frequencies are less than expected. The measure of expectation or normality is the corpus of general English-the British National Corpus Sampler (BNCS). So positive and negative keywords do not relate to positive or negative connotations of words, rather they point to the statistical significance of words based on frequency count, which makes them pivotal in their contexts.
Although the total number of keywords found is 469, the first 269 are the most significant ones or the positive keywords which constitute our dataset. To illustrate lexical types, however, the top 100 positive keywords and the 63 negative keywords were used. The reason for this selection is that while the positive keywords represent COS (sermon text) words, the negative keywords represent BNC (general English) words. Therefore, analysis of variation between them points us toward differences in the use of lexical word types. So, we shall now classify the keywords in each category into the four groups of lexical word classes. Subsequently, we shall show their distribution in graphic form to illustrate the results.

Lexical Word Classification in Sermons and General English
Discussions on the use of lexical word classes revealed which classes are more commonly used in each text and verify the claim for preferential uses of lexical word types in texts to establish the uniqueness of the sermon texts. First, in Table 2 below, the top 100 keywords in COS are classified, and later, in Table 3, the word classes in the BNCS are also classified.
From the classification above, of the lexical word classes in the COS 100 most significant words, it can readily be seen that the sermon words are most productive in nouns (freq. 5,810 = 9.14%). This is followed by main verbs (freq. 680 = 1. 07%), adjectives (freq. 216 = 0.34%), and then adverbs (freq. 116 = 0.18%). This result, a high frequency of nouns, corroborates our finding about the function word classes because we found a dominance of the definite article among the sermon words that signals an equally high presence of nouns, as determiners make noun references. We also found that the primary auxiliaries are more frequent in the sermons, and this agrees with a higher use of main verbs among lexical words in the sermons. So there is indeed a connection between the occurrence of function and lexical words in texts. Conversely though, the very high noun frequency recorded seems to contravene previous findings that a high prevalence of pronouns in texts will correspond to a low density of nouns and vice versa (Biber, 1999). In the sermon words, we found that both nouns and pronouns were the highest in frequency, in both categories of lexis, to disprove this assertion. The implication is that sermons have proved themselves different from other text types about which this conclusion is drawn. We can say then that sermons are most lexically dense in nouns, and this shows that the information/ content level is expected to be high as nouns are the main carriers of lexical meaning. We shall now examine the results found in the BNCS.
As can be seen, in the BNC, there is no marked difference in the occurrences of the lexical word classes. For example, the occurrence of nouns (13,328 = 0.67%) is only marginally higher than that of adjectives (11,124 = 0.56%), but falls below the occurrences of verbs (19,726 = 1.00%) and adverbs (21,965 = 1.11%). This shows that in general English, adverbs and verbs may contribute more to the density of lexis in texts, unlike in sermons where verbs have a much lower occurrence rate than nouns and adverbs are found to be very rare. This suggests that in terms of information load, sermons will be heavier than general English. These results are summarized in Table 4, below, and then shown graphically in Figures 3 and 4, below the table.

Lexical Density in Sermons and General English
The distinction of content and function words is relevant to text structure because different types of texts have predictably different proportions of content and function words. While the proportion of content words indicates the lexical density of a text, the proportion of function words points to core vocabulary. On average, written texts are proved to have higher percentages of content words (ranging from 36% to 57%) than spoken texts (ranging from 24% to 43%) because they are more packed with information (Stubbs, 2002). The lexical density of a text is the number of content words expressed as a percentage of the total running words, that is, 100 × L/N, where L stands for the total number of lexical words and N stands for total running words. This factor of lexical density is an index of text variability (Moon, 1997). To calculate lexical density, we need to compute total raw frequency as against total running words. The results obtained show that a high token value does not necessarily translate into high frequency. For example, the 372 lexical word tokens found in the top 500 words manifested a total frequency of 17,638 out of a possible frequency of 52,621 for the 500 words. This frequency represents 33.51% of the top 500 words, while in BNCS, the 324 word tokens had a cumulative frequency of 333,420 out of a possible 1,379,186 frequency to represent 24.17% of the top 500 words in the BNCS. So lexical density in COS and BNCS per 500 most frequent words stand at 33.51% and 24.17%, respectively. Therefore, the density of lexical words in COS is shown to be higher than that in BNCS. This proves that even texts of a similar kind, for example, both COS and BNCS, consisting of a mixture of written and spoken data, can still vary in the number of lexical words used, depending on the information load in them. Hence, the results of previous research (Biber, 1999) on the ratio of lexical content in written versus spoken texts (above 40% for written and below 40% for spoken texts) could not apply here because of the nature of the texts. This result corroborates our earlier prediction that COS will be more lexically dense than general English and, due to the higher percentage of function words, will also manifest more textual cohesion too. Let us now examine the multi-word items found in the sermons to see how they behave in context. Note. COS = sermon text; BNC = general English. Note. COS = sermon text; BNC = general English.

Discussion of the Results of Multi-Word Items in the Sermons
The study has revealed that multi-word items exist in the lexical world of the sermons. It has shown, for instance, that  the sermons utilize a great number of word sequences, ranging from two-to ten-word combinations. Of the overall number of 2,639 word combinations found, 114 were found to be based on the sermon-religion genre and, at the same time, qualify as multi-word items based on the factors of conventionality, fixedness, and non-compositionality. These were consequently separated, and our discussion here will focus on them, because only from them can we identify the character of multi-word items in the sermons, their structural types, and possibly functions. In terms of types, we found that the majority of the sermon-based multi-words were two-to four-word sequences, and most of them were compounds or lexical collocations, for example, holy spirit, born again, first fruit, and expressions, such as blood of Jesus, body of Christ, and son of God. Among the expressions, some are fixed such that they do not allow any modification or variation, for example, the blood of Jesus, the word of God, the kingdom of God, the lion of the tribe of Judah, and the things of God, which can neither inflect nor have alternative forms. And some others are semifixed, in that they can inflect and vary in their component parts, for example, child of God/children of God, high priest/ high priests, a believer/the believers, and Christ Jesus/Jesus Christ. As the majority of the sermon-based multi-words found are compounds/lexical collocations, fixed or semifixed expressions, this suggests that the sermon texts have a preference for compounds/lexical collocations and expressions. One important observation to make here is that multiword items in the sermons are quite unproductive in their types. While previous works (Hsu, 2006;Lewis, 1997;Moon, 1997;Nattinger & DeCarrico, 1992) reveal more than a dozen types, and though there is considerable overlap in the categorizations, the sermon texts chiefly manifest two main types. This supports the claim of preferential use of multiword items across text genres.
Concerning the structures of the multi-word items found, most of them were phrasal rather than clausal. The study has shown, for example, that two-to seven-word combinations, which constitute the majority of multi-words found, were phrases rather than clauses or sentences. The reason could be that sermons are prepared primarily for listening and, as such, shorter phrases will be more effective and easier to recall. There are examples such as the father, a watchman, prayer of agreement, glory of God, the children of Israel, the  Note. COS = sermon text; BNC = general English.  Note. COS = sermon text; BNC = general English. Now in terms of pragmatic functions, the study shows less specificity of pragmatic functions. This is because the multiwords are mostly collocations that do not carry pragmatic functions. The instances and types of multi-words found agree with Leech's (1997) observation that corpora will manifest limited evidence for multi-words because multi-word items are largely genre-specific. However, Moon (1997) notes that the compound words (also lexical collocations), which are the most productive multi-word items in the sermons, typically denote and have high information content, because often they are technical terms or make specific reference. This function extends to most of the other expressions identified (at least 78%) and, interestingly, agrees with our earlier finding in the study, in relation, especially, to the discoursal functions of content and function words.
To conclude, the investigation into the types and peculiarities of sermon words yielded positive results. The findings show that there are words that are peculiar to sermons, words predominant in sermons, words having unusually high frequency, words rarely or never found elsewhere, and words that are used in ways different from their general usage.

Results for Objective 2: The Patterns of Words in the Sermons
For each of our search words, collocates were found in large numbers. However, because we are constrained by both space and time it became necessary to select a sample for closer examination. For each keyword therefore, the first 50 collocates, sorted on their Mutual Information (MI) scores, were retained; these are those words that are most strongly associated with the keywords, such that they are highly primed to co-occur. Our discussion here will center on the behavior of the keywords selected for in-depth study, in terms of the collocations they form-their strength, directionality, and phraseology-to underline any peculiarities in comparison with general English.
In terms of collocations (co-occurrences), each of the 20 keywords demonstrated a very high tendency to co-occur with certain other words, each manifested a high association strength (a MI score of 7 or above), proving that they are very much attracted to each other and, as such, are confirmed collocates of the nodes. However, the results show that when placed on the collocation strength clines of strong, medium strength, and weak, all but one of them illustrated weak collocating strength, showing that there is a lack of reciprocity in the degrees of association, as they are unpredictable in their partners and so are unreliable as indicants of fixed meaning. It was also seen that each of them established their individuality, in that each has more or less its own class of collocates, as was the case with their associates. At the same time, there was more or less uniformity in the collocating style of the nodes: a downward collocating pattern in which the node chooses collocates less frequent than itself.
The implication is that each word derives meaning from its immediate environment, defined by the words constantly in proximity to it. Therefore, the contextual meaning of a word is a function of its co-text, its syntagmatic relations rather than a function of choice alone, by paradigmatic means as standard grammars hold. This confirms Firth's assertion that meaning inheres the horizontal plane of language, and bears out previous research to be found in the literature of lexical studies, for example, Halliday (1991), J. Sinclair (2004), Louw (1993), Stubbs (2002), Hunston (2002), Biber (1999), Nattinger and DeCaricco (1992), and Hoey (1991). Some examples of the collocations found in the sermons are as follows: God partner, Christians testify, Lord crucified, salvation tragedy, praise God, salvation helmet, love responses, rising praise, faith heroes, mercy deeds, forbidden fruit, sin saved, joyful challenges, soil word, prophet anointing, root sin, miracle seed, and victorious walk.
In terms of phraseology or patterns, the study shows that the keywords exhibit both grammatical and lexical patterns. However, our interest is in the lexical patterns, to investigate constancy and peculiarity. The patterns of collocation found were 10 major kinds as shown in the Table 5 that follows.
As indicated earlier, lexical collocations are formed from combinations of lexical word classes: nouns, verbs, adverbs, and adjectives; and Benson, Benson, and Ilson (1997) identified seven types as follows: V + N/P/PP, V + N, ADJ + N, N + V, N1 of N2, ADV + ADJ, and V + ADV. Of these, the first five were found in the sermons, while the last two were not; rather, some other kinds of collocation structures occurred, for example, N + N, ART + N, NP, V + V, and V + NP. This means that collocation structures can be used as an index of text variation, as the sermon texts have proved to be productive, and also in the use of other types of collocation patterns. The results also call for extension of the index of lexical collocation types to accommodate the ones revealed in the study.

Collocation Analysis in BNCS
We sought here to examine the similarity of collocate types. Based on experience and intuition, where there is up to 50% similarity, we posit lack of substantial difference; but where the similarity is less than 50%, we posit difference of usage and, consequently, difference in behavior. For each keyword, the first 20 strongest collocates in the BNCS were compared with the first 20 collocates in COS.
The examination revealed the dissimilarities between collocates in these groups and those in COS, and immediately pointed to totally different usages. It became obvious that we cannot hypothesize the same attitudinal meanings for these words in this context. This suggests that these words differ significantly from general English words and may characterize only sermon texts. This point was evident when we compared the first 10 collocates of the historical words in the BNCS to their counterparts in COS, as shown below. We find that the asymmetry between collocates of each word in both contexts is overwhelming. For example, the collocates of God are had, one, became, sun, his with, local, great, and creator. The last two were found in the context of sermons.
This picture is largely the same as the results found in the case of other words. Therefore, in all cases, similarities of keywords, associates, and collocates were well below 50% to confirm the peculiar uses and behaviors of these words in the sermon context. This view is strengthened even more by the multi-word items retrieved for the words; for example, among the historical words, only God and salvation showed a pattern in BNCS; in the case of God, there were a sun god, a fertility god, to the god, as a god, while for salvation there were such structures as the national salvation and the national salvation front. Obviously, in both cases, two totally different lexical items are meant. Based on the extensive differences recorded in terms of collocates and their networks, we posit different usages for these words in the sermons and in ordinary English.
In conclusion, the words selected for more thorough examination have manifested sufficient differences in their lexical behaviors, in the different contexts, both in terms of the words they associate with and the patterns of their association.

Results for Objective 3: Semantic Implications of the Collocations of Words in the Sermons
The overall aim is to reveal the implications of the preferences of each item, that is, those words in whose company a word most often occurs (semantic preference), and which give it an aura of meaning (semantic prosody). The results obtained for each of the words, which form the basis of our discussion here, have been summarized for conciseness and clarity and are shown in Table 6, above.
As can be seen from this table, the concepts of semantic preference and semantic prosody apply to sermon words. It is evident too that the majority of words analyzed demonstrated positive prosody, and this contradicts previous research (Louw, 1993;J. Sinclair, 1991), which claims that semantic prosodies are predominantly negative. However, negative prosodies were also found, as were neutral ones. We can then infer that most sermon words have positive or good prosodies. This means that their attitudinal meanings are principally positive; sermon words connote more good than bad. The implication of this finding is that semantic prosodic relations, beyond their variable nature, may in fact be genrespecific. As earlier mentioned, genre investigations of the relation of semantic prosody are sparse, but we can at least speculate that different registers may demonstrate different types of prosodies. Therefore, the study of connotative language should take not only context into cognizance but also the real context, of natural language.
Notwithstanding the above, we must not lose sight of the main import of semantic prosodic relations which is, according to Sinclair, to offer the semantic analysis of a word, that is, its meaning. The question to grapple with is, "Have collocations of sermon words succeeded in creating meaning?" The answer is yes and can be illustrated using the example of faith, which demonstrates a preference for many words of verbal action such as call, speak, laments, roar, declare, ask, build, confess, cry, and demonstrate. These words, found in the co-text of faith, clearly contribute to its meaning; it is either that they show that faith is a force that empowers one to speak, act, or achieve some success, or that they enable the building of a neutral attitude toward the word as they evoke neither a bad nor good feeling toward the concept. They constitute the lexical network for the word, the knowledge of which can enhance its understanding.
Semantic prosody can also be seen in terms of word membership of definite semantic groupings or in terms of the pragmatic meaning of words. Our analysis has taken both perspectives into account; semantic preferences that prioritize collocates relations demonstrate semantic memberships, while semantic prosody that concentrates on the node word gives us the attitudinal meanings of words. The question-Do sermon words associate with definable semantic sets of words?-is then answered by the results of the semantic preferences of each word, and the question-Do they have consistent pragmatic meanings?-is also answered by their semantic prosodies.
The analyses show that almost all the words (24 of 25) demonstrated membership of some semantic sets of words; they associate with words of similar meanings and, in some cases, words were found that belonged to two different semantic sets, while some words were shown to share associations, that is, to associate with similar groups of words or semantic sets. Therefore, the findings show that semantic sets are both individualistic and general. For example, sin demonstrated membership of the semantic group of words of negativity and undesirable states by its association with words such as washing clean, unrighteousness, stain, brings out of, poverty, sickness, hurts, remission, erased, convicts, cries out, and bad. And at the same time, it was shown to share this semantic association with some other words in the sermons such as Christians, salvation, and saved. This shows that some semantic sets recur in the sermons, as Table 7 below further illustrates.
As the table shows, 22 semantic sets of words were identified for the sermon words, and the dominant ones are as follows: negative words, number words, difficult situation words, words of uncertainty, words of description, words of physical action, power/utility words, place expressions, and words of verbal action. However, this does not imply that they are the only semantic associations that can be found, or that they are even the most frequent in the sermons. We cannot make that claim because of the size of the selected sample, but we can say that the sermon words do associate with semantic sets of words that may differ in kind and content from their associations outside of sermon context, as supported by the data shown in Table 8, below.
The above table clearly shows that words in the sermons differ in their semantic relations from words in ordinary usage. Whether in terms of semantic preference or prosodies, significant differences are recorded; the attitudinal meanings varied substantially, so did their semantic membership. Where a word seems to belong to the same semantic set in both contexts, differences were noticed in the makeup of the semantic set. Take, for example, life, which maintained the same set membership in both contexts by belonging to the semantic group of descriptive words, yet it manifested different contents (lexical networks) in both, as shown in Table 9 below.
The lexical networks seem to suggest that, in COS, most of the descriptive labels refer to the same "life" but, in contrast, in BNCS there is reason to believe that kinds of life are indicated. For example, in BNCS, a technological life is certainly differentiated from a public life, an adventurous life, a hard life, a private life, and most certainly a dog's life; but in COS, the majority of the words seem to describe a particular kind of human life to indicate the attributes. We see then that the word behaves differently in both contexts, and these behaviors contribute to building a semantic frame for it.

Summary of Findings
In terms of word frequency, above 50% similarity of word types and their frequencies were found. But in terms of relative frequencies, the significant words in the sermons proved substantially different from the significant words in general English. The primary lexis of the sermons was shown to be mostly lexical words whose productivity was found to be in descending order of nouns, verbs, adjectives, and then adverbs, with nouns being of highest frequency. So the lexical density of the texts derived from these and, in consequence, high information content was recorded in the sermons.
The collocational analyses revealed differences in the usage of words in the sermons and in general English: in associations, patterns, behavior, and meaning. In the first place, the words examined portrayed noteworthy differences in the kinds of words they co-occur with; different sets of collocates were retrieved for each of the selected items in sermons and outside of sermons. In terms of patterns, the study showed both grammatical and lexical patterns; however, as our interest is collocation not colligation, discussion was based only on the lexical patterns. In the sermons, 10   main kinds of patterns of word combination were found: 5 of the 7 lexical collocation types known to occur in general English were found, and another 5 types, uncommon in ordinary usage, were also found. This suggests that, beyond meaning relations, patterns of collocation may distinguish texts, in addition to leading to some generalizations in language.
With regard to the semantic prosodic relations of words, the findings show that words in the sermons have prosody; they belong to semantic sets and possess pragmatic attitudinal meanings by virtue of their associations with other words. The results demonstrate that sermon words are predominantly positive in prosody, in contravention of previous findings on the prosodies of individual words, and point toward the likelihood of genre distribution of prosodies. The semantic memberships of the sermon words showed a dominance of types: negative words, number words, difficult situation words, words of uncertainty, words of description, words of physical action, power/utility words, place expressions, and words of verbal action. We can only speculate that the tendency is for sermon words to belong mostly to these semantic groups.

Conclusion
The study has proved that the behavior of English lexis in the sermons is considerably unlike its behavior in general English. The specific differences recorded in lexis in both contexts, we must say, would have been impossible without the corpus evidence provided by the actual sermons. It is this incontrovertible evidence that most of all validates the findings of this study and lends credibility to its conclusion. Although English in the sermon context may not, as yet, be viewed as a genre, owing to the limited, almost lack of, research in this area, there is considerable evidence of the impact of context on the content, usage, and meaning of lexis in the domain to warrant further research. This work has clear implications for descriptive linguistics, applied linguistics, semantics, and education.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) disclosed receipt of the following financial support for the research and/or authorship of this article: This research was partly funded by the American Council of learned Societies (ACLS) through a dissertation fellowship award in 2010.