Translation data have been used extensively in contrastive corpus linguistics for "the systematic exploitation of the bilingual intuition of translators" (JOHANSSON 2007: 5). According to Johansson (ibid.), translation (or parallel) corpora have the "special advantage" that they make available texts expressing approximately the same meanings and discourse functions in two languages, such that by exploiting them we can build "paradigms of correspondences" (ibid: 9). The procedure consists in selecting a word or phrase in the target (or source) text and collecting its equivalent word(s) or phrase(s) in the source (or target) text. The correspondences can be congruent (the expected translation, sometimes referred to as "dictionary equivalent"), divergent (a more unexpected choice), or null (the word or phrase has no correspondence in the other language). By alternating between the target-to-source and the source-to-target direction, the degree of mutual correspondence, or functional similarity (ALTENBERG 2007) of single expressions and semantic categories can be established without relying on intuition or dictionaries/grammars. We can thus also evaluate the strength of each mutual correspondence (i.e., the likelihood that a given word is translated with another), and its (a)symmetry (i.e., the tendency for one correspondence to be stronger in one translation direction than in the other). Asymmetry would be an effect of translation, or translation bias which, seen from the perspective of contrastive corpus studies, is an unwanted effect, to be controlled for thanks to comparable corpus analysis.

From the perspective of translation studies, however, the focus is reversed: translation effects are the main object of interest, and contrastive differences have to be controlled for. Translation effects can be sought at the micro-level of single translator's choices (see e.g. hypotheses about translator style, MARCO 2004) or at the slightly more general level of single language pairs/directions (e.g. TEICH 2003 on German <-> English, XIAO AND HU 2015 on English -> Chinese); but they have also been hypothesised to be shared by all translators/languages (as in BAKER's (1993) proposal), or even to all instances of constrained language use (LANSTYÁK & HELTAI 2012).

Going into details about this body of work would require too long a digression, but the relevant point here is that corpus methods and resources can be used to shed light on the translation process by focusing on the ways in which text production is affected by it. Simplifying substantially, translation effects can be observed by contrasting translated texts to comparable non-translated texts in the same language, or by contrasting source and target texts in two languages. Through the former, one can focus on higher-order general textual features (as in, e.g., LAVIOSA 1998), yet relating these features causally to the translation process requires something of an act of faith, since text and corpus comparability across cultures is very difficult to ascertain (BERNARDINI & ZANETTIN 2004). Contrasting source and target texts in parallel corpora instead shows us the textual effects of tangible decisions made by translators. Generalizability may be an issue, but we get a tangible glimpse (however indirect, as is in the nature of corpus data), of the translation process. As claimed by Aston (1999: online), "[s]ince parallel concordances provide translations of each occurrence, citations are more likely to be immediately understandable for the user, diminishing the difficulties of retrieval and risks of misinterpretation associated with monolingual and comparable corpora".

Being able to efficiently mine information about word use from general and specialised text collections is one of the main competences defining professional translator expertise according to the competence framework of the European Master's in Translation Network⁵ and, I would suggest, one of the future skills needed to cope with the twists and turns of a fast-changing profession (EHLERS 2020). Yet it would be wrong to consider corpus linguistics first and foremost as a method for extracting lexical, terminological and phraseological information from text collections. The cognitive effort needed to process corpus data is substantial, and may be off-putting and ultimately unrewarding if one approaches corpora as they would approach reference resources/tools like dictionaries or even translation memories. Coherent with the assumptions of data-driven learning (JOHNS 1991), the main role of corpora in translator education would seem to be as researcher's playgrounds, where learners can come into contact with large amounts of authentic (and therefore messy) textual data, practice their budding research skills on them, and arrive at an understanding of contrastive language use, and of translation as regulated by contextually - and situationally - bound preferences and norms, rather than absolute rules (BERNARDINI ET AL. 2020).

Before we can exemplify this kind of language data exploration, however, we need some basic theoretical notions that we can rely on. A useful distinction, that goes back to Ferdinand de Saussure (1916), concerns paradigmatic and syntagmatic relations. The former, well-known and widely focused upon in linguistics and philosophy, refer to word associations: different verb forms belonging to the same lemma would constitute a verbal paradigm, and different words related to each other by means of different lexical relations (synonymy, antonymy, hyponymy, hypernymy etc.) would constitute a lexical paradigm. Paradigmatic associations are typically part of our language competence (Saussure's langue) because it is unlikely that they appear together in texts (Saussure's parole): normally, we do not need to use two verb forms belonging to the same paradigm together, or two synonyms next to each other (though of course there will be plenty of exceptions to this general principle). Paradigmatic relations are the building blocks of grammar, semantics, logic.

The other relations introduced by Saussure, syntagmatic ones, are less familiar to many of us. This is because they belong to parole, or language performance, which, prior to the advent of corpus linguistics, had not generally been the object of linguists' attention. In the words of Sinclair (2004: 10), "linguistics has been formed and shaped on inadequate evidence", namely researchers' or informants' introspection. The Neo-Firthian approach to (corpus) linguistics focuses instead primarily on syntagmatic relations observable in corpus data. It would be wrong to conclude that corpus linguistics is about language performance only, however. If we are able to recognize not only fixed expressions like proverbs and idioms, but also conventionalised word sequences formed according to standard grammar rules, it is because we have encountered them often enough for them to enter our language competence (though not necessarily our language awareness). As claimed by Saussure (1916: 173),

Syntagmatic relations are more elusive and less intuitively salient than paradigmatic ones, but fortunately they are also more easily observable in corpora - if we know what to look for. According to Sinclair (e.g. 2004: 141-42), frequency data about word co-occurrence obtained from corpora can be categorized at four levels of abstraction. The most basic level is that of collocation, which looks at what words appear together with the word(s) we are studying (we will ignore the methods for ranking collocates, but see e.g. EVERT 2009). In Artusi's Science in the kitchen, the ten most frequent nouns following the verb form cook within four words to the right are water, rice, butter, bain-marie, semolina, broth, flame, peas, milk and meat.

What do we make of this list, from a recipe book? Some nouns probably describe types of foods (rice, semolina, peas, meat), while others are more likely to refer to methods for cooking them (water, butter, bain-marie, broth, flame, milk). We could generalise over these observations by saying that cook has a semantic preference for words belonging to a set that we could label as "foods that are cooked", and another semantic preference for a set that we could label "cooking methods". Any other food or cooking method that appeared together with the verb would be added to these sets, regardless of its frequency: even single occurrences thus become relevant, by virtue of their belonging to a semantic set. Notice that semantic preferences are subjective categorizations: the sets are not given, and the grouping may differ from analyst to analyst. For instance, flame and bain-marie might be considered more similar to each other than the rest of the words in this set, and thus worthy of a separate set and label.

Colligation, describing the tendency for a word to co-occur with a set of function words, can be illustrated by the co-occurrence of cook with prepositions. If we go back to the cooking method set, and try to reconstruct their lexico-syntactic structure, we probably end up with cook + [preposition]

+ water/butter/flame etc. The most frequent grammatical class immediately following cook is indeed that of prepositions (followed by determiners and pronouns). A focused search for cook followed by a preposition would point us to different ways of expressing cooking techniques: cook in (less frequently with) butter or broth, cook on the grill or on a spit, cook over a slow fire or a high flame, and so forth.

The different syntagmatic relations that contribute to the study of units of meaning are not equally relevant for all words. The fourth and most abstract relation, semantic (or evaluative) prosody, is hard to illustrate by way of reference to a simple material process verb like cook. We shall instead use simple as an example. In Science in the Kitchen, the most frequent word following "simple [comma]", and "simple and", is healthy. Other words following these two patterns are nutritious, agreeable, natural, restful, quicker, light, civilized, and appetizing. All these adjectives express a positive evaluation by the author, and this attaches to the (more) neutral simple. Whenever we read of a simple dish, or a simple meal, we do not need other, more explicit, hints: we know that Artusi is appreciative of that dish or meal, because of the positive evaluative prosody of simple. After all, he describes himself as a "simple amateur" (ARTUSI 2003: 7), and conversely derides the "complex flavor" of Budino Gabinetto (Cabinet Pudding) that "tastes of diplomacy" and is mockingly dedicated "to the greatest of ministers, to the cynosure of the hour" who had suddenly just plunged into oblivion (ARTUSI 2003: 482-83). Evaluative prosody is more relevant for the study of socio-culturally meaningful texts, while semantic preference is more relevant when approaching scientific or technical texts. La scienza in cucina belongs to both categories, which is why it is such an interesting object of analysis, as we shall see in 2.3 below.

Communication in the culinary domain has seen a surge of academic interest in recent years, both within linguistics in general (GERHARDT ET AL. 2013, JURAFSKY 2014) and within translation studies (PARADOWSKI 2018). The socio-cultural relevance of the theme hardly needs pointing out, but there are also linguistic and translation-specific issues to consider, that make it especially interesting for translation students. Cookbook translation, for instance, has to convey culturally distinctive practices, ingredients and instruments to a wide audience of readers, from the lay person to the semi-expert, whose level of acquaintance with the culture and/or the techniques may vary considerably. The difficulties faced by technical translators add to the difficulties faced by translators of culturally-laden texts. On the one hand, success conditions for recipes dictate that the reader is able to make a satisfying dish following the instructions provided. On the other, for this success condition to be fulfilled the translator may need to apply a cultural filter, neutralizing or generalizing culture-specific items and ultimately "frustrat[ing] the expectations to taste a foreign flavor" (REBECHI AND TAGNIN 2020: 68), which had probably motivated them to seek out the recipe in the first place. If culinary translation is a challenging area for translation practice, corpora can come to the rescue. Rebechi and Tagnin (2020) offer two main reasons why corpus resources may be especially valuable for culinary translators: first, to address terminological knowledge gaps that may emerge due to the lack of reference materials, and second, to account for differences in the level of technicality of texts in different languages, which they observe in their comparison of Anglo-American and Brazilian cookbooks.

The culinary domain thus offers an excellent playground where to practice corpus-based research with translation students. But why focus on a single, rather old and difficult cookbook? First published in 1891, only three decades after Italy became a single country, Pellegrino Artusi's La scienza in cucina is without doubt the best-known Italian cookbook, both nationally and internationally. It was the first Italian cookbook to explicitly address middle-class family cooks and housewives from all over Italy, who in return contributed recipes and comments that were often included in subsequent editions. Mediating, both linguistically and gastronomically, between local traditions and local dialects, Artusi, "the gastronomic father of us all" (BALLERINI 2003), was among the greatest contributors to the development of the Italian nascent national identity (CAPATTI AND MONTANARI 1999: 237-40). This was achieved thanks to a very peculiar, almost dialogic writing style and to the inclusion of anecdotes, stories, digressions and poems alongside the actual cooking instructions. Even though the language is at times difficult for the contemporary reader, the instructions sometimes considered too vague (SCICOLONE 2003), and the recipes somewhat too heavy for current tastes (CAPATTI 2019), the book is still in print, and virtually every household in Italy has a copy. To this day, Italians read La scienza in cucina as testimony of a very significant moment in Italian history and as an utterly enjoyable literary text, as well as a cookbook.

Given the inherent difficulties of translating cooking texts discussed above, the peculiar textual features of Artusi's style, and the author's role as the father of Italian cuisine, parallel analysis of La scienza in cucina and its most highly regarded English translation (BALLERINI 2005) would seem to offer translation students a vantage point from which to appreciate the complexity of an endeavour straddling the traditional opposition between literary and technical translation (WITTMAN 2012, OLOHAN 2012).

As shown in table 1., the default equivalent CUOCERE, together with the noun COTTURA, take up 72% of all the source occurrences. The percentage is worth bearing in mind, but these cases will not be further studied, as they correspond to our expectations - they will feed back into the analysis later on, as we reverse the perspective and look for English equivalents of CUOCERE in the target text. We shall also ignore the three cases (2%) in which COOK corresponds to Italian CUCINARE (a synonym of CUOCERE), and focus instead on the more surprising choices.

Table 1

	Frequency	Percentage
CUOCERE + COTTURA [Noun]	108	72%
TIRARE, TIRARE a cottura, TIRARE a sapore	13	8.6%
no source text equivalent	8	5.3%
BOLLIRE + BOLLITURA [Noun] (BOIL/SIMMER)	5	3.3%
CUCINARE	3	2%
LESSARE (BOIL/SIMMER)	2	1.3%
RESTRINGERE (COOK down)	2	1.3%
SERVIRSI (USE)	1	0.6%
OCCORRE (NEED)	1	0.6%
BAGNARE (MOISTEN)	1	0.6%
RITIRARE (DRY)	1	0.6%
STRUGGERE (DISSOLVE)	1	0.6%
BASTARE (BE enough)	1	0.6%
PREPARARE (PREPARE)	1	0.6%
SOFFRIGGERE (FRY)	1	0.6%
METTERE in forma (PUT in a mold)	1	0.6%
Total	150	100%

⁹Glosses are provided only where congruent decontextualised equivalents (other than COOK) exist.

There are 13 occurrences of COOK (8.6% of the total) that translate the Italian term TIRARE, with its variants TIRARE a cottura and TIRARE a sapore (a rather unusual phrase in current Italian). The term describes a process whereby something is cooked by adding liquid to it from time to time, letting it dry, and then adding more. The verb focuses on the end result: if the instruction is to tirare a cottura, then the resulting state is that of food that is a) done, and b) in a dry-ish state (as in a risotto, the typical dish that is cooked in this way). The English text sometimes emphasises the end point of the process (cook thoroughly, finish cooking, simmer until cooked), but typically leaves the dry-ish state out, as in examples 1-3 below:

1	The meat stuffing, consisting of chickpea-size chunks, is best made with chicken livers, white chicken meat, sweetbreads and the like, and then cooked in brown stock and bound with a pinch of flour.	Il battuto di carne, a pezzetti grossi quanto i ceci, sarà bene farlo con fegatini, petti di pollo, animelle e cose simili tirate col sugo di carne e legate con una presa di farina;
2	All manner of vegetables, simply boiled or cooked in butter, may be used in omelettes, and one may add as well a pinch of grated Parmesan cheese, by itself or with parsley.	Ogni erbaggio o semplicemente lessato o tirato a sapore col burro, serve per le frittate, come può servire un pizzico di parmigiano solo o mescolato con prezzemolo.
3	No chard, no ink, and when the cuttlefish, as I have described above, start to turn yellow, add the rice and cook thoroughly in hot water and tomato sauce (recipe 6) or tomato paste, adding some butter for more flavor and a better appearance.	Niente bietola, niente inchiostro, e quando le seppie, come si è detto, cominciano a prendere il giallo, versate il riso e tiratelo a cottura con acqua calda e sugo di pomodoro o conserva, dandogli più grazia e sapore con un pezzetto di burro;

The next category down, with eight occurrences (5.3% of the total), is that of additions: the Italian source sentence contains no obvious term corresponding to COOK in translation. The translators have added information that remains implicit in the source, making the instructions clearer, as in examples 4-6.

4	If you have no brown stock, sauté a little onion in olive oil and butter, adding salt, pepper and the usual spices, and then use the mixture to cook the spleen.	In mancanza di questo, servitevi di un battutino di poca cipolla, olio, burro, sale, pepe e spezie per condimento. [for seasoning]
5	When this is done, immediately pour the broth over the drained, partly cooked macaroni. |#| I say immediately, because otherwise the Gruyere sinks to the bottom and sticks together.	ciò ottenuto, versatelo subito sui maccheroni già sgrondati dall'acqua e dico _subito_, perchè altrimenti il gruiera cala a fondo e si appasta. [on the drained macaroni]
6	Keep the tagliatelle al dente, and salt the water you are cooking them in very sparingly given that the sauce has prosciutto in it. Strain the pasta and dress it with the sauce and Parmesan cheese.	Le tagliatelle cuocetele poco e salatele pochissimo a motivo del prosciutto: levatele asciutte, conditele col detto intingolo e con parmigiano. [salt them very sparingly]

In example 6, for instance, the instruction is to salt the tagliatelle. While comprehensible, this is technically imprecise, since tagliatelle are boiled in water, and one does not salt them, but the water they are cooked in.

Further down the list we find five occurrences (3.3%) of COOK translating BOLLIRE and two of LESSARE (1.3%), both of which would approximately correspond to the English verbs BOIL/SIMMER. The use of BOLLIRE in some of these cases would indeed be rather unusual in modern Italian as well, since the cooking method is more akin to stewing than boiling/simmering (as in 7). In other cases, like 8, the verb co-occurs with the noun brodo (broth), and as such would seem to correspond to prototypical cases of the boiling cooking method.

7	Cover the pan with a plate, on which you will place a little water, and cook the meat at a gentle simmer for six hours.	Coprite il tegame con un piatto che lo chiuda e sul quale sia mantenuta sempre dell'acqua e fate bollire la carne per sei ore continue, ma in modo che il bollore appena apparisca. [make the meat boil]
8	Cut the mixture into little bricks or cakes as thick as a heavy coin and between 1 or 2 centimeters wide (about 1/2 inch), which you will cook in broth for a few minutes.	Tagliatelo a mattoncini o a mostaccioli della grossezza di uno scudo e della larghezza di un centimetro o due e gettateli nel brodo facendoli bollire qualche minuto. [drop them in the broth and boil them]

The tail of the list consists of nine verbal processes, each occurring once only (0.6%, the singletons taken together accounting for 6% of the total). All but one of the verbs (9-17), with their immediate co-texts, depict the action to be performed more precisely than the source expressions they translate. The one exception is number 18, where COOK translates the arguably more precise SOFFRIGGERE (meaning brown, sauté). The translators may have opted for the more generic COOK because soffriggere in this co-text would be unusual also in Italian today (in itTenTen16 it has among its objects ingredients like onion, garlic, bacon etc., rather than sliced meat). Or they may have made this choice not to repeat BROWN or SAUTÉ, both verbs having been used (of lard and onion respectively) in the preceding sentences.

9	During the hunting season, a skillful chef may cook it in a mold, stuffing it with stewed birds.	Nel tempo della cacciagione un abile cuoco può metterla in forma riempiendola di uccelletti cotti in umido. [can place it in a mold]
10	There are many people who, ignorant of the ways of food preparation, have a horror of garlic merely because they can smell it on the breath of those who have eaten it raw or poorly cooked.	Ci sono molte persone, le quali, ignare della preparazione dei cibi, hanno in orrore l'aglio per la sola ragione che lo sentono puzzare nel fiato di chi lo ha mangiato crudo o mal preparato; [unskilfully prepared]
11	Turn them often and when they are all red, baste them with tomato sauce (recipe 6) or tomato paste. Then add enough hot water to cook the rice.	Rivoltateli spesso e quando tutti saranno divenuti rossi, bagnateli con sugo di pomodoro o conserva e poco dopo versate tanta acqua calda che possa bastare pel riso. [that can be enough for the rice]
12	For cooking instructions, follow recipe 75, keeping in mind that this risotto will be more substantial and flavorful if cooked in broth.	Per la cottura regolatevi come al N. 75. Per rendere questo risotto più sostanzioso e più grato al gusto occorre il brodo. [broth is needed]
13	You may cook this dish in water instead of broth, and serve it quite dry.	Invece di brodo potete servirvi di acqua e farlo asciutto: [avail yourselves of water]
14	Chop the vegetables well and begin by sautéing the onion in oil. When the onion turns a golden brown, add the other vegetables. Season with salt and pepper, stirring constantly. Let the vegetables cook until they have reabsorbed their own water.	Tritate alquanto gli erbaggi e mettete per prima la cipolla a soffriggere nell'olio e quando questa avrà preso colore gettate giù gli altri, conditeli con sale e pepe, rimestate spesso e lasciate che ritirino l'acqua che fanno. [let them reabsorb the water they let out]
15	Cut the meat into little cubes and mince the bacon and the herbs finely with a mezzaluna. Put all the main ingredients on the fire with the butter. When the meat has browned, add the pinch of flour and then cook in broth until done.	Tagliate la carne a piccoli dadi, tritate fine colla lunetta la carnesecca, la cipolla e gli odori, poi mettete al fuoco ogni cosa insieme, compreso il burro, e quando la carne avrà preso colore aggiungete il pizzico della farina, bagnando col brodo fino a cottura intera. [moistening with broth]
16	Take a domestic duck, put it into a saucepan with some butter, season with salt and pepper, and when it begins to brown, add a battuto made with prosciutto, onion, celery and carrot, which you will place under the bird to cook. Remember to turn the duck regularly.	Prendete un'anatra domestica, mettetela in cazzaruola con un pezzetto di burro, conditela con sale e pepe e, quando avrà preso colore, aggiungete un battuto, tritato ben fine, di prosciutto, cipolla, sedano e carota. Lasciatelo struggere sotto l'anatra, rivoltandola spesso; [dissolve]
17	During the hunting season, a skillful chef may cook it in a mold, stuffing it with stewed birds.	Nel tempo della cacciagione un abile cuoco può metterla in forma riempiendola di uccelletti cotti in umido. [place it in a mold]
18	Once dry, add lard and sauté. When the onion begins to brown, add the liver, which should be cut into thin slices. Let it cook for a while, along with the onion.	asciutta che sia versateci il lardo per friggerla, e quando avrà preso il color marrone uniteci il fegato tagliato a fette sottili. Lasciatelo soffriggere alquanto, frammisto alla cipolla; [brown/sauté]

Having analyzed the source text correspondences of COOK, the next step in the analysis consists in reversing the perspective and looking at target text correspondences of the lemma CUOCERE. It will be remembered from 4.1 that this verb lemma accounts for the sheer majority of source text correspondences of COOK (68% of the total).

Table 2

	Frequency	Percentage
COOK	85	64.8%
done	22	16.8%
BOIL	6	3.8%
BOIL (CUOCERE (a) lesso)	3	2.3%
BAKE	2	1.5%
no equivalent	2	1.5%
ready	1	0.8%
STEW	1	0.8%
STEW in the oven	1	0.8%
TURN soft and translucent	1	0.8%
FRY [... ] with some butter	1	0.8%
ROAST	1	0.8%
GRILL	1	0.8%
SAUTÉ	1	0.8%
PLACE	1	0.8%
ADD	1	0.8%
PREPARE	1	0.8%
Total	131	100%

The results in table 2. show, unsurprisingly, that in the majority of cases (85 out of 131, or 64.8% of the total) CUOCERE is translated as COOK. In some of these cases, the action is not further specified (example 19). This happens especially when the past participle cotto appears in the source text, indicating a completed action or end state. In most other cases CUOCERE is further specified by way of reference to an ingredient or cooking technique (examples 20-22). In these cases, more specific verbs could have been used instead of the generic COOK (brown or sauté for cook in butter, boil or simmer for cook in water/broth), yet the target text is unambiguous thanks to the specification. In a similar case PREPARE is used instead of COOK, but once again the sense is clear thanks to the specification (cotte nell'acqua salata > prepare [...] in salted water).

19	Keep the dough on the fire until the flour is thoroughly cooked (10 minutes), stirring constantly.	Tenete la pasta sul fuoco fino a che la farina sia ben cotta (10 minuti) rimovendola sempre;
20	Cuocete i fegatini nel burro e quando l'avranno tirato bagnateli con brodo;	Cook the livers in butter, and when they have absorbed it cover with broth.
21	Per non consumar tanto brodo _imbiancate_ il riso nell'acqua e terminate di cuocerlo col brodo dei detti polli.	To avoid using too much broth, first blanch the rice in water, and then finish cooking it in the broth made by the chicken
22	Condite con questo intingolo gli spaghetti cotti in acqua poco salata, procurando che restino durettini.	Use this sauce to dress the spaghetti cooked in lightly salted water, making sure they turn out al dente.

In several other cases in which the past participle cotto appears in the source text, indicating a completed action or end state, the target text has done (22 occurrences, 16.8% of the total) or ready (1 occurrence), and in two cases the term is removed altogether, possibly because the end state can easily be inferred from the preceding instructions.

To the exclusion of two cases in which place and add translate mettere a cuocere (literally put to cooking), all remaining instances have more specific verbs than COOK, sometimes accompanied by specifications. These are worth considering in greater detail, to discern the reason for the greater specificity. We can observe two cases: one in which the target text expression does specify the technique further, and another in which the target text expression translates a unit of meaning with CUOCERE at its core.

In the first set we find six occurrences of BOIL, three of BAKE, and single occurrences of four other verbal expressions. As illustrated by example 23, boil in the broth translates cotta nel brodo. This is very similar to 21 (above), where COOK in broth translates CUOCERE col brodo. In both cases the specification (nel/col brodo > in the broth) is present in the source text as well, yet in 23 the cooking technique is also encoded through the verb BOIL. Similarly, in example 24, cotte in forno (cooked in the oven) is translated as stewed in the oven and in 25cotte in forma (mold cooked) is translated as mold-baked. In examples 26-28 the cooking technique remains implicit in the source, where it could only be inferred by reading the recipe as a whole and/or calling on cooking knowledge: the tartlet is cooked by baking it, the teal duck is cooked by sautéing it, and the mix is cooked by frying it. Finally, example 28 adds information that could not be inferred from the co-text but only from previous knowledge: the source text author assumes that the end state corresponding to cooked is known to the readers, while the translators describe it: the onion is cooked when it turns soft and translucent.

23	Cotta la pasta nel detto brodo scolatela bene e conditela a suoli con questa carne tritata, burro e parmigiano a buona misura.	Boil the pasta in the broth, then drain and season with the minced teal meat, butter and a good measure of Parmesan cheese, making a few layers.
24	Le _cottarone_, per chi non lo sa, sono mele o pere, per lo più cascaticce, cotte in forno entro una pentola nella quale si versa un gocciolo d'acqua, coprendone la bocca con un cencio bagnato.	"Cottarone," for those who do not know them, are apples and pears, mostly overripe, that are stewed in the oven in a small pan with a little water in it, while the top of the pan is covered with a moist kitchen towel.
25	Il pasticcino cuocendo rigonfia naturalmente e lascia un vuoto nell'interno;	As the tartlet is baking, it will naturally puff up with air, remaining rather hollow inside.
26	Prendete un'arzavola e, vuotata e pulita come le suddette, mettetela a cuocere insieme con un battuto di cipolla (un quarto o mezza se è piccola), un bel pezzo di sedano, mezza carota, grammi 40 di prosciutto grasso e magro e un pezzetto di burro; sale e pepe per condimento.	Take one teal, clean it as described in the preceding recipe, and sauté it in a battuto made with a quarter of an onion (half if the onion is small), a good-sized stalk of celery, half a carrot, 40 grams (about 1-1/3 ounces) of untrimmed prosciutto, and some butter, seasoning with salt and pepper.
27	Gettate questo intriso in padella per cuocerlo in bianco con burro, olio o lardo e quando è assodato da una parte voltatelo con un piatto dall'altra, ed eccovi il tondone.	Put this mix into a skillet and fry it only with some butter, olive oil or lard. When it has firmed up on the bottom, use a dish to flip it over on the other side. Now you have a tondone.
28	Quando è cotta mettetela da parte.	When the onion has turned soft and translucent, remove it from the pan and put to one side.

The six occurrences in the second set illustrate the case where the target text has a single verb that translates a unit of meaning with CUOCERE at its core. In Italian the cooking technique is encoded in the prepositional phrase that follows the verb, rather than the verb supporting it (not so in English): BOIL (CUOCERE a lesso, 3 occurrences), STEW (CUOCERE in umido, 1), ROAST (CUOCERE arrosto, 1) and GRILL (CUOCERE in gratella, 1).

At this point we can try and distill a few overall tendencies out of the chaos of authentic text, ignoring the inevitable exceptions. For the time being we will refrain from interpreting these tendencies, and leave this final step for section 5. below. The analysis of source equivalents of COOK when this verb does not translate its obvious Italian equivalents (CUOCERE and CUCINARE) points to two main tendencies: on the one hand, COOK translates more specific terms, generalizing over cooking techniques that might be unusual or difficult to understand for current source text readers as well. On the other, COOK is added or used in place of more general, vague verbs, making the instruction more easily interpretable.

The analysis of target equivalents of CUOCERE confirms that COOK is the obvious choice. We observed several cases in which CUOCERE is in fact the core of a longer unit of meaning, formed of the verb plus a preposition phrase that specifies it (e.g. CUOCERE nel brodo). These are translated by means of a corresponding expression (e.g. COOK in the broth), by a more specific verb that encapsulates the entire unit of meaning (e.g. BOIL), or by the more specific verb and a specifying prepositional phrase (e.g. BOIL in the broth). Finally, several instances of specification were also observed in the target text, whereby information that readers of the source text can only infer from the co-text, or from their disciplinary knowledge, is made explicit.