2 3. The analogy account Contrary to what is assumed within the generative framework, it has recently been argued that LD-movement (specifically LD wh-movement) does not involve a productive rule at all, but rather that these constructions are based on a general template (cf. D browska, 2004, 2008; Verhagen, 2005, 2006). This analysis will be referred to as the analogy account. The idea is that any LD-construction departing from the general template is created by analogy to this template. This hypothesis springs from the observation that naturally occurring examples of LD wh-questions show very little variation regarding their type of matrix predicate and subject. D browska and Verhagen report that in English, the construction is almost exclusively attested with the matrix verb think or say, the auxiliary do and a 2 nd person pronoun as the matrix subject. D browska (2004) looked at the Manchester corpus and found that 96% of the LD wh-questions had the matrix verb 'think' or 'say'. Furthermore, 91% of the occurrences had 'you' as the subject and 99 % had some form of 'do' in the auxiliary position. D browska (2004) further looked at the CHILDES-data and found that 47 out of 49 occurrences of long-distance wh-questions were of the form "WH do you think S?" (where 'S' stands for 'subordinate clause'). In D browska (2008), additional data from the British National Corpus (BNC) is discussed. She reports that 70 % of the LD wh-questions in the spoken part of the BNC also have the form "WH do you think S?". Similar findings are reported in Verhagen (2005) and (2006) for the Brown corpus: out of 11 occurrences, 10 had the matrix verb 'think' and 1 'say'; 9 had the matrix subject you, and 10 occurred with a form of do as the auxiliary. In Verhagen (2005) and (2006), it is furthermore pointed out that Dutch shows a similar pattern. Verhagen searched the digital version of the newspaper De Volkskrant and the Eindhoven corpus for LD wh-questions. In the Eindhoven corpus, 6 out of 6 occurrences showed up with the matrix verb denken 'think' and a 2 nd person personal pronoun. Data from the Volkskrant showed that 34 out of 43 occurrences had the matrix verb denken 'think', 5 willen 'want' and 4 zeggen 'say' or vinden 'find'. Furthermore, 36 occurrences had a 2 nd person personal pronoun as the matrix subject. Based on these observations, D browska (2004, 2008) and Verhagen (2005, 2006) argue that LD-movement constructions are stored as fixed formulas as in (5a) below for English and (5b) for Dutch, and are created by analogy to this formula. (5a) [WH do you think/say [ S … ]] (5b) [WH denk je [dat …]] WH think you that On this account, the more an LD-movement construction departs from this formula, the less acceptable it will be. The limited variation in LD wh-questions indeed suggests that the construction is not as productive as a purely formal account would predict. However, as is pointed out in the next sections, our data shows considerable variation in the type of matrix predicates and subjects that may show up in LD-mov...