**2. Materials and Methods**

## *2.1. Hybrid Meat Products in the UK Market and Review Collection*

A web search was conducted to understand the presence of hybrid meat products in the UK market. If a specific retailer was found to have launched a range, then the retailer website was further investigated to get more details on the characteristics and availability of each product. Product details were compiled into a list. Some of these products contained customer product reviews, and all available reviews were collected. This made up a total of 201 reviews from the websites of Waitrose, Ocado and Sainsbury's for three hybrid meat products. The products were Waitrose pork, chickpea and spinach sausages (79 reviews), Waitrose harissa chicken cauliflower rice & chickpea meatballs (106 reviews) and Sainsbury's Love Meat & Veg! Mediterranean beef meatballs (16 reviews). All reviews are publicly available on the retailer's website and were downloaded into an excel spreadsheet, noting down the product name, review title, review comment and score from 1 to 5. The reviews were divided into two groups: reviews with a score equal to or above 3.5 were included in the corpus of positive reviews, whereas reviews with a score below 3.5 and below formed the corpus of negative reviews. Table 1 shows the size of the two corpora, with the majority of reviews being positive (80%).


**Table 1.** Size of the corpora of online product reviews on three UK hybrid meat products.

Because the reviews were short and often included only a few words (not complete sentences), the sizes of the corpora are relatively small. Nonetheless, they were still large enough to perform frequency counts and key term analysis.

#### *2.2. Statistical Approach for Frequency Counts and Key Term Analysis*

Both positive and negative corpora were uploaded onto the linguistic software program Sketch Engine, which performed frequency counts and an extraction of key terms. Frequency counts of language items in reviews can be a useful indicator of preferences in that frequent items can signal preferred lexical choices which in turn can point to attitudes and stances. Yet, the most frequent items in English are grammatical words (e.g., articles, prepositions) that as such are used to form grammatical constructions and do not hold a lexical meaning. Because we were interested in attitudes towards the hybrid-meat products, and attitudes are likely to be revealed in the ways in which the consumers describe and evaluate the products, the analysis was focused on adjectives. Adjectives are parts of speech that function primarily as descriptors denoting a whole range of features and dimensions such as size, colour, quantity, texture, taste, judgment and affect [21]. Since all these dimensions can be relevant to hybrid-meat products, adjectives were selected as good indicators of consumers' attitudes towards specific features of the products.

Once the corpora were uploaded onto Sketch Engine, a parser was applied which tagged each word in a given corpus with its parts of speech (verbs, nouns, adjectives, etc.). In this way, adjectives were identified, and their frequencies retrieved. Because adjectives can describe a range of dimensions, subsequently all adjectives retrieved from the two corpora were grouped into their semantic domains. This allowed us to determine which dimensions of the products (e.g., texture, taste) were particularly emphasised and how they were evaluated by those who liked and disliked them. The grouping of adjectives into semantic domains was conducted first independently by the two researchers; disagreements and ambiguous meanings (e.g., the adjective 'hot' can be used to describe temperature or the level of spiciness) were resolved by checking the meanings of the adjectives in context, that is, how they were used in the reviews.

In order to gain insights into other salient themes and issues mentioned by the consumers, we also retrieved distinctive multiword items from both corpora, also known as key terms. Key terms are simply distinctive combinations of two or three words which appear more frequently in the studied corpus as compared to a reference corpus and, additionally, match the typical format of terminology in the language, that is, they are lemmatised. Key terms are good indicators of the content and distinctive topics of the studied corpus. For the purpose of this analysis, we used the EnglishTenTen corpus (available on Sketch Engine) as a reference corpus because it is a large compilation of general English collected from online sources. The key terms were retrieved using keyness scores calculated as follows:

$$\frac{fpm\_{f\text{ocus}} + n}{fpm\_{ref} + n}$$

*Fpmfocus* stands for normalised frequency (per million) of the term in the focus corpus (in our case in positive or negative reviews), while *fpmref* is the normalised frequency (per million) of the term in the reference corpus. *N* is the simple maths parameter added to account for the problem that we cannot divide by zero. Retrieved key terms were then grouped into semantic domains using the same procedure as above. The next section summarises the main findings that emerged from the search of hybrid meat products in the UK market and the results of the corpus linguistic analysis.

## **3. Results and Discussion**
