1. Week 9. Sentence processing and Linger GRS LX 865 Topics in Linguistics

2. Human sentence processing When we hear a sentence, we have to assign it a structure. When we hear a sentence, we have to assign it a structure. Intuition indicates that this happens incrementally, we’re building the structure as the words come in. Intuition indicates that this happens incrementally, we’re building the structure as the words come in. Language is rife with ambiguity, but we very often don’t notice… Language is rife with ambiguity, but we very often don’t notice…

3. Ambiguity Global ambiguity Global ambiguity Most of my family lives in the Mpls area, but I sometimes don’t feel like going home for the holidays. Most of my family lives in the Mpls area, but I sometimes don’t feel like going home for the holidays. My family likes to travel in the summer, so I’m always careful to be out of town during the summer months. My family likes to travel in the summer, so I’m always careful to be out of town during the summer months. Visiting relatives can be tedious. Visiting relatives can be tedious. I sat by the bank. I sat by the bank. Nothing cleans better than Tide. Nothing cleans better than Tide.

4. Local ambiguity Because we parse things incrementally, we also face lots of local ambiguity. Because we parse things incrementally, we also face lots of local ambiguity. I saw the student… I saw the student… …yesterday. …yesterday. …leave the room yesterday. …leave the room yesterday. …from Turkey’s mother yesterday. …from Turkey’s mother yesterday. …from Turkey’s mother leave the room yesterday. …from Turkey’s mother leave the room yesterday. At the point where we hear the student, it could have any number of roles to play in the rest of the as-yet-unheard sentence. At the point where we hear the student, it could have any number of roles to play in the rest of the as-yet-unheard sentence. Object, embedded subject, possessor in an object, possessor in an embedded subject. Object, embedded subject, possessor in an object, possessor in an embedded subject.

5. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was While she was

6. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was mending the sock While she was mending the sock

7. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was mending the sock fell. While she was mending the sock fell.

8. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was mending the sock fell. While she was mending the sock fell. Since Jay always Since Jay always

9. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was mending the sock fell. While she was mending the sock fell. Since Jay always jogs a mile Since Jay always jogs a mile

10. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was mending the sock fell. While she was mending the sock fell. Since Jay always jogs a mile seems light work. Since Jay always jogs a mile seems light work.

11. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was mending the sock fell. While she was mending the sock fell. Since Jay always jogs a mile seems light work. Since Jay always jogs a mile seems light work. I gave the man I gave the man

12. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was mending the sock fell. While she was mending the sock fell. Since Jay always jogs a mile seems light work. Since Jay always jogs a mile seems light work. I gave the man the report I gave the man the report

13. Local ambiguity Generally, this isn’t a problem, but sometimes… Generally, this isn’t a problem, but sometimes… While she was mending the sock fell. While she was mending the sock fell. Since Jay always jogs a mile seems light work. Since Jay always jogs a mile seems light work. I gave the man the report criticized a demotion. I gave the man the report criticized a demotion. What these suggest is that we are processing ambiguity online and discarding a reading too early… What these suggest is that we are processing ambiguity online and discarding a reading too early…

14. Mending the sock So what goes wrong with While she was mending the sock fell? So what goes wrong with While she was mending the sock fell? Given the choice, and you seem to have to make a choice, you take the sock to be the object of mending, rather than guess that the sock might be an embedded subject. Given the choice, and you seem to have to make a choice, you take the sock to be the object of mending, rather than guess that the sock might be an embedded subject. Because you can attach the sock to mending, you do. Because you can attach the sock to mending, you do.

15. Late Closure An early principle proposed to account for this is Late Closure, a principle followed by our parsing mechanism. An early principle proposed to account for this is Late Closure, a principle followed by our parsing mechanism. Late Closure When possible, attach incoming lexical items into the clause or phrase currently being processed (i.e., the lowest possible nonterminal node dominating the last item analyzed). Late Closure When possible, attach incoming lexical items into the clause or phrase currently being processed (i.e., the lowest possible nonterminal node dominating the last item analyzed).

16. RC attachment The journalist interviewed the daughter of the colonel who had had the accident. The journalist interviewed the daughter of the colonel who had had the accident. To whom do you send flowers? To whom do you send flowers? This is ambiguous, but we have a pretty strong preference to attach the relative clause low (so, the flowers go to the colonel). This is as Late Closure would predict. This is ambiguous, but we have a pretty strong preference to attach the relative clause low (so, the flowers go to the colonel). This is as Late Closure would predict.

17. Late Closure Late Closure is usually taken to be a universal property of human sentence processing (not something English- specific). Late Closure is usually taken to be a universal property of human sentence processing (not something English- specific). #John said Bill will die yesterday. #John said Bill will die yesterday. #Juan dijo que Bill morirá ayer. #Juan dijo que Bill morirá ayer.

18. Cuetos and Mitchell (1988) In a now-famous study, Cuetos & Mitchell (1988) tested RC attachment in both Spanish and English: In a now-famous study, Cuetos & Mitchell (1988) tested RC attachment in both Spanish and English: The journalist interviewed the daughter of the colonel who had had the accident. The journalist interviewed the daughter of the colonel who had had the accident. El periodista entrevisto a la hija del coronel que tuvo el accidente. El periodista entrevisto a la hija del coronel que tuvo el accidente. Hmm. It’s Early Closure in Spanish? Hmm. It’s Early Closure in Spanish?

19. Gibson et al. (1996) Exploring what’s behind the difference. What favors high attachment in Spanish (RCs)? Exploring what’s behind the difference. What favors high attachment in Spanish (RCs)? La lámpara cerca de la pinture de la casa que fue dañada en la inundación. The lamp near the painting of the house that was damaged in the flood. La lámpara cerca de la pinture de la casa que fue dañada en la inundación. The lamp near the painting of the house that was damaged in the flood. “Early Closure” would predict high attachment is preferred, wouldn’t have much to say about relative preferences between low and middle (or would predict middle is better than low). “Early Closure” would predict high attachment is preferred, wouldn’t have much to say about relative preferences between low and middle (or would predict middle is better than low). OTOH, if there’s a second (stronger) factor favoring high attachment, but LC is still around, then low should be better than middle. OTOH, if there’s a second (stronger) factor favoring high attachment, but LC is still around, then low should be better than middle.

20. Testing attachment preferences There are a couple of ways to try to test attachment preferences. There are a couple of ways to try to test attachment preferences. Questionnaire. Present ambiguous sentences, ask how it was understood. Questionnaire. Present ambiguous sentences, ask how it was understood. Graham pointed to the photo beside the letter that was consulted in the lawsuit. Graham pointed to the photo beside the letter that was consulted in the lawsuit. What had been consulted? What had been consulted? Problem here is that these are ambiguous, there might be “late processing” involved here. What we want to know most is what the parser does automatically, right away… Problem here is that these are ambiguous, there might be “late processing” involved here. What we want to know most is what the parser does automatically, right away…

21. Testing attachment preferences Online. Present unambiguous sentences incrementally, and look at whether people slow down more in some conditions than in others (Reanalysis). Online. Present unambiguous sentences incrementally, and look at whether people slow down more in some conditions than in others (Reanalysis). Pat gestured toward… Pat gestured toward… …the lamps near the paintings of the house that was damaged in the flood …the lamps near the paintings of the house that was damaged in the flood …the lamps near the painting of the houses that was damaged in the flood …the lamps near the painting of the houses that was damaged in the flood …the lamp near the paintings of the houses that was damaged in the flood …the lamp near the paintings of the houses that was damaged in the flood Advantage: This is early processing. Advantage: This is early processing.

22. Gibson et al.’s (1996) task Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). --- -------- ------ --- ----- --- --- -------- ------ --- ----- ---

23. Gibson et al.’s (1996) task Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). Pat -------- ------ --- ----- --- Pat -------- ------ --- ----- ---

24. Gibson et al.’s (1996) task Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). --- gestured ------ --- ----- --- --- gestured ------ --- ----- ---

25. Gibson et al.’s (1996) task Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). --- -------- toward --- ----- --- --- -------- toward --- ----- ---

26. Gibson et al.’s (1996) task Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. Sentences were presented in a self-paced “moving window” display, where subjects pressed the space bar for the next word, or pressed “No” if a sentence became ungrammatical. On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). On some items (a third) the subject was asked to paraphrase the sentence (just to keep them alert). --- -------- ------ the lamps --- --- -------- ------ the lamps ---

29. Two factors Gibson et al. (1996) conclude that there are two factors at work, one that encourages high attachment, and one that encourages low attachment. Nothing encourages middle attachment, so those are the hardest. Gibson et al. (1996) conclude that there are two factors at work, one that encourages high attachment, and one that encourages low attachment. Nothing encourages middle attachment, so those are the hardest. But look—now it looks as if the Spanish speakers prefer low attachment to high attachment. But look—now it looks as if the Spanish speakers prefer low attachment to high attachment. Gibson et al. (1996) conducted essentially the same experiment in English… Gibson et al. (1996) conducted essentially the same experiment in English…

32. Hmm. English looks pretty much the same as Spanish, in these three-site sentences. English looks pretty much the same as Spanish, in these three-site sentences. Though we still believe Spanish prefers high attachment when it’s just between two sites. Though we still believe Spanish prefers high attachment when it’s just between two sites. “High attachment” in Spanish isn’t just due to “Primacy”, clearly. “High attachment” in Spanish isn’t just due to “Primacy”, clearly. Gibson et al. propose: Predicate Proximity Attach as close as possible to the head of a predicate phrase. Gibson et al. propose: Predicate Proximity Attach as close as possible to the head of a predicate phrase.

33. Predicate Proximity The lamp near the painting of the house that was damaged in the flood. The lamp near the painting of the house that was damaged in the flood. Recency (Late Closure) prefers house, then painting. Recency (Late Closure) prefers house, then painting. Predicate Proximity prefers lamp, but that’s really non-Recent in a 3-clause sentence. Predicate Proximity prefers lamp, but that’s really non-Recent in a 3-clause sentence. They speculate that Predicate Proximity is stronger in languages with more fluid word order. They speculate that Predicate Proximity is stronger in languages with more fluid word order.

34. Topics in language acquisition The (celebrated) difference between English and Spanish attachment preferences has opened up an interesting avenue in L2 acquisition. The (celebrated) difference between English and Spanish attachment preferences has opened up an interesting avenue in L2 acquisition. Where will a Spanish learner of English prefer to attach? Where will a Spanish learner of English prefer to attach? Where will an English learner of Spanish prefer to attach? Where will an English learner of Spanish prefer to attach? It there transfer of parsing principles? It there transfer of parsing principles?

35. German, Greek, and Spanish German and Greek also have a Spanish-like high attachment preference. German and Greek also have a Spanish-like high attachment preference. Papadopoulou & Clahsen (2003), Gross (2002), show that they nevertheless show a low attachment preference with “thematic prepositions”: Papadopoulou & Clahsen (2003), Gross (2002), show that they nevertheless show a low attachment preference with “thematic prepositions”: Everyone liked the actress with the servant who is always smiling. Everyone liked the actress with the servant who is always smiling. Everyone liked the servant of the actress who is always smiling. Everyone liked the servant of the actress who is always smiling. This might be thinkable-about as having an additional predicate lower down (re: Predicate Proximity). This might be thinkable-about as having an additional predicate lower down (re: Predicate Proximity).

36. L1  L2 transfer Felser et al. (2003) found German  English speakers showed low attachment for of but showed no preference for with. Felser et al. (2003) found German  English speakers showed low attachment for of but showed no preference for with. That’s not transfer. That’s not transfer. English speakers prefer low attachment for both of and with, German speakers prefer high attachment for of and low attachment for with. English speakers prefer low attachment for both of and with, German speakers prefer high attachment for of and low attachment for with. Greek (  English) comes out like German(  English). Greek (  English) comes out like German(  English).

37. Child processing Felser et al. (2003b) looked at this phenomenon in English speaking children and found: Felser et al. (2003b) looked at this phenomenon in English speaking children and found: Kids break into two groups, neither group differentiating of and with. Kids break into two groups, neither group differentiating of and with. The “low-span” kids generally attached low. The “low-span” kids generally attached low. The “high-span” kids generally attached high. The “high-span” kids generally attached high.

38. Modularity A question that is often asked is: to what extent is the parser an independent module? What kinds of grammatical (and semantic and pragmatic)) information is available immediately, as a sentence is being incrementally parsed? A question that is often asked is: to what extent is the parser an independent module? What kinds of grammatical (and semantic and pragmatic)) information is available immediately, as a sentence is being incrementally parsed? Some propose that the (first pass, automatic) parser respects only structural constraints, with semantic and pragmatic information kicking in later. Eye tracking studies and the like have suggested otherwise. Some propose that the (first pass, automatic) parser respects only structural constraints, with semantic and pragmatic information kicking in later. Eye tracking studies and the like have suggested otherwise.

39. Badecker & Straub (2002) An interesting experiment was designed in this connection by Badecker & Straub (2002), looking at whether interpretations ruled out by grammatical constraints (Binding Theory, specifically) are discarded immediately during processing. An interesting experiment was designed in this connection by Badecker & Straub (2002), looking at whether interpretations ruled out by grammatical constraints (Binding Theory, specifically) are discarded immediately during processing. John said that Mary hurt himself.*Pr.A. John said that Mary hurt himself.*Pr.A. John said that Bill i hurt him i.*Pr.B. John said that Bill i hurt him i.*Pr.B. He i said that Bill i fell.*Pr.C. He i said that Bill i fell.*Pr.C.

40. Previous results Badecker & Straub (1994) Badecker & Straub (1994) Kenny assured Lucy…fastest Kenny assured Lucy…fastest Julie assured Harry… Julie assured Harry… Kenny assured Harry… Kenny assured Harry… Julie assured Lucy…slowest Julie assured Lucy…slowest …that he was prepared for the new job. …that he was prepared for the new job. The next question is: What if the referent for he was ruled out by Binding Theory? The next question is: What if the referent for he was ruled out by Binding Theory?

41. Badecker & Straub (2002) John thought that Bill… John thought that Bill… John thought that Beth… John thought that Beth… Jane thought that Bill… Jane thought that Bill… Jane thought that Beth… Jane thought that Beth… …owed him another chance to solve the problem. …owed him another chance to solve the problem. “No candidate” was slowest in 1994. In 2002, will the last two be equivalent? Will the first two be equivalent? (Does Principle B filter out the first and third?) “No candidate” was slowest in 1994. In 2002, will the last two be equivalent? Will the first two be equivalent? (Does Principle B filter out the first and third?)

42. Well, let’s see… The program to play with today is Linger, an experiment manager (particularly useful for “moving window” experiments), written by Doug Rohde of Tedlab. The program to play with today is Linger, an experiment manager (particularly useful for “moving window” experiments), written by Doug Rohde of Tedlab. http://tedlab.mit.edu/~dr/Linger/ http://tedlab.mit.edu/~dr/Linger/

43.                       