EPISODE TRANSCRIPT
The Sound of Silence: Investigations of Implicit Prosody (Nicholas J. Van Handel)
Automatically transcribed from the episode audio. Some words and names may be inaccurate.
Listen to the episodeDownload textDownload captions
Welcome back to the Deep Dive. Today we are strapping in for a fascinating journey into one of the most hidden mechanisms of your own brain, the sound of silence. It really does sound like a paradox, doesn't it? But it's deeply real. We're talking about the unseen rhythmic, the phrasing, and the international structure that you automatically assign to sentences the moment you read them silently. And for this, we're really digging into one source. We are. We're synthesizing the monumental work from the 2022 investigation, The Sound of Silence, Investigations of Implicit Prosody. That's right. So our mission today is to explore how this silent sound, this implicit prosody, is constructed word by word in real time. And crucially, how it acts as, well, maybe the primary driver for resolving ambiguity in how we parse sentences. It turns out that the phantom voice in your head is less a quiet narrator and more a very strict foreman deciding how phrases get built. That's a great way to put it. So let's start at the beginning. To understand the silent voice, I guess we first have to understand the spoken one. Let's establish the ground truth of prosody in overt spoken language. Right. When we talk about prosody, we're basically talking about the music of speech, correct? Precisely, yeah. Music is the perfect word. Prosody covers, you could say, three main elements. First, you have the rhythm, the pattern of stressed and unstressed syllables. Like convict versus convicity.
Exactly. Right. Then there's phrasing, how we break up the speech stream into chunks, or what linguists call prosodic constituents. The pauses, the grouping of words. And finally, the international contour. That's the rise and fall of pitch, which conveys all sorts of things like focus, emotion, whether you're asking a question. And the key thing here, which the research makes very clear, is that these elements are not just decorative. They're not just flavor. Not at all. They're completely integral to comprehension. So the source highlights that prosody influences how we understand things in two main ways. Boundaries and accents. Let's tackle boundaries first. If I'm listening to you speak and you, you pause for a second. How does that pause, that boundary change how I structure the sentence in my head?
It basically builds a wall. A structural wall. I mean, decades of research show that when listeners hear a major prosodic boundary like a longer pause or a real distinct change in pitch, they're consistently less likely to attach any new information to the word that came right before that boundary. OK, so the purser in my brain just assumes that phrase is done. It's closed off. It is. The boundary, in effect, repels new information. It says, don't attach here. This part is finished. So if I'm dealing with an ambiguous sentence. Something like, the audience enjoyed the performance of the singer, pause, who was touring Europe. Right.
That pause makes me strongly prefer that who was touring Europe modifies the performance, not the singer. Exactly. The boundary isolates the singer, so the attachment has to go up higher in the structure to performance. We call that high attachment. Okay, so that's the pushing force. The boundary pushes away. But now let's talk about the pulling force. Pitch accents. The accents, right? These are all about information structure, highlighting what's new or important, what we call focus. And this is where the focus attraction hypothesis, or FAH, comes in. Focus attraction. So that sounds pretty self -explanatory.
It is. The FAH argues that focus... Well, it attracts attachment. If a sentence has two possible nouns, a phrase could modify, and you put a prominent pitch movement on one of them, the parser gets pulled toward that accented noun. So in that same sentence, if you said the audience enjoyed the performance of the singer with that emphasis. Yep. then the relative clause, who is touring Europe, is almost certainly going to be interpreted as modifying the accented singer. We call that low attachment. Focus is a very strong attractor. Okay, so this is fascinating. In spoken language, we have these two powerful forces that can sometimes compete. We have boundaries that push material away repulsion. Right. And we have accents that pull material toward the fraction. So what happens when those two forces clash in the same sentence?
Which one wins? And that. That is the exact question where the source material builds its foundation. It challenges some older assumptions and establishes what's called the visibility first hypothesis or VFH. Visibility first. OK, that name suggests there's a hierarchy. One of these is dominant. It is. The VFH argues that attachment at its core prioritizes the most visible site. And visibility here is a technical term. What does it mean? It's defined by the absence of a major prosodic boundary. So crucially. The VFH sets up a clear pecking order. Prosodic boundaries are the top priority. They are the boss. So if a boundary separates a potential attachment site from the incoming words... Then that site is basically rendered invisible. It's structurally inaccessible.
So the phrasing, the boundary, actually overrides the focus from the accent. Absolutely. The VFH posits that pitch accent of focus only get a say in the matter. They only start to modulate our preferences when both potential attachment sites are equally visible. Meaning when they're both inside the same uninterrupted phrase? Exactly. If there's a strong boundary in the way, the parser doesn't care how brightly you accent a word. It just defaults to the site that's on the near side of that boundary wall. It's a fundamental reordering of how we thought these prosodic cues work together. Okay, this is a perfect foundation.
We have the rules of engagement for spoken language. Now we have to transpose all of this onto silent reading. Right. And this is where we get to the core concept. Implicit praesity. The mental blueprint of rhythm, phrasing, and contour that the reader is just generating internally. And this internal blueprint, this silent music, is what drives the whole theory. The implicit prosody hypothesis, or IPH. What's the core principle of the IPH? It's pretty straightforward, actually. The IPH just says that when a reader encounters a sentence that could be structured in two different ways, a syntactic ambiguity, the parser favors the analysis that aligns with the most natural, the most default, the most expected prosodic contour for that string of words. So the inner voice isn't just, it's not a passive narrator just reading the words back to you. It's actively structuring the text as you go. It's structuring it. And that structure determines the meaning before the syntax is even fully resolved. Wow.
And we know that this structure, it has a cognitive cost, especially when it needs to be undone, right? A huge cost. And this is captured in what's called the prosodic constraint on reanalysis, or PCR. Let's unpack the PCR a bit. How much harder is it to read when you have to perform both a syntactic and a prosodic reanalysis? It's significantly harder. I mean, think about a classic... garden path sentence. The kind that leads you down one path and then a later word reveals you took a wrong turn. Exactly. That process of revising your interpretation is already costly. But the PCR says that cost is compounded if you also have to dismantle and rebuild the rhythmic structure your inner voice is already assigned to the first part of the sentence. So if you only have to change the grammar, that's one level of difficulty.
It's manageable. But if you have to tear down the mental phrasing and the intonation pattern you've already built, the process becomes much slower, much more error prone. And the dissertation points to some German self -paced reading studies to show this. It does. They had these ambiguous relative clauses. And when the sentence resolved in a way that required a complete shift in the expected intonational phrasing, the reading slowdowns were just much more pronounced. So the brain is really investing effort in building this.
prosodic show. And it hates wasting that investment. The big takeaway for the field, really, is that we cannot treat silent reading as a purely visual, a purely syntactic process. The prosodic that we generate is this mandatory, instantaneous layer of interpretation that actively governs how we resolve conflict. This all sounds incredibly neat in theory, but now we have to face the central challenge, right? The idea of a single default parsity. Ah, yes. I mean, if I read a sentence at a normal pace, there are often a few different ways I could phrase it that all sound perfectly fine. They're not weird or marked. So how can the parser rely on a default if several options seem possible? And you just hit on the crux of the incrementality problem. This is the core issue. The prosody that you assign during what we call first pass parsing, that's the rapid.
word -by -word decision -making process, has to operate with incomplete information. It doesn't know how the sentence ends yet. Exactly. So that initial incremental prosody you assign might be fundamentally different from the final default prosody that a speaker would assign once they can see the whole finished sentence. The parser has to make a choice now, not later. So what happens if the prosodic structure starts building in the first few words? What if that contradicts the structure that would be best for the whole sentence once you see it all? That divergence, that conflict, is exactly where parsing difficulty and reanalysis costs come from. Let's use a concrete example. The dissertation brings up Fodor's relative clause length effects. This is a classic finding.
It's a crucial one because it's one of the first big pieces of evidence that silent reading is sensitive to these subtle structural properties that we usually associate with spoken prosody. So can you refresh us on that pattern? Sure. In a complex noun phrase like, say, The professor saw the friend of the senior. What comes next is key. If the relative clause is short, like who studied, readers almost always prefer low attachment. Meaning the clause modifies senior, the closest noun. Right. But if the relative clause is long, who studied for hours and hours in the library, they suddenly flip their preference and prefer high attachment. So now it modifies friend. Correct. And the prosodic explanation for this is that readers are mentally, implicitly inserting a prosodic boundary before a long, heavy relative clause. That's the theory. That implicit boundary repels the RC, pushing it up to the higher attachment site, friend. But this brings us right back to the paradox of incrementality.
It does. Because when the parser hits that word who, It has absolutely no idea if the relative clause is going to be short or long. It has to decide right then and there whether to introduce a boundary or not. It's playing chicken with the boundary. It is. Does it just go the default, which is usually attaching to the nearest lowest site, a principle we call late closure? Just keep the phrase open. Or does it wait a few words to see if the phrase gets heavy enough to justify putting in a boundary? And this introduces all these complex timing questions. We need to figure out the rules the parser is using to build this prosody on the fly to anticipate the final structure without knowing the full input. So the goal of the whole investigation, really, is to define the rules of that inner voice, to model the constraints on how that rhythmic structure is built word by word, to predict when and why the parser decides to insert that invisible but structure -altering pause. That's the challenge. Okay, so before we can trust any experimental results about this incremental process, the dissertation has to deal with a big methodological fear. A very big one. The fear that the experimental task itself might be corrupting the very thing we're trying to measure. This brings us to a comparison between two main reading tasks, the maze task and self -paced reading. This chapter is so important for the methodology. Because as researchers, we need tools that force people to make immediate, local processing decisions. The maze task is great for that, but it's design. It poses a serious threat to validity when you're studying something as subtle as prosody. So let's detail the maze task first. What does a participant actually do? In the maze task, the sentence is shown one word at a time, but at every single step, The participant has to choose between the correct next word and a foil word, a word that's plausible but grammatically wrong. Ah, so it forces a choice point, a pause, after every single word. A mandatory pause, yes. And how does that compare to the more traditional self -paced reading or SPR task?
SPR is much simpler. You just press a button to see the next word. It measures how long you spend reading that word, and you often get what are called spillover effects. Meaning the difficulty from one word shows up in the reading time of the next word or two? Exactly. The processing cost can be delayed. So the big problem with the maze, which this research had to face head on, is the risk that all those mandatory pauses would induce a totally artificial unnatural prosody. A staccato rhythm, as they call it. Right. And if the maze forces a boundary after every word, then we can't possibly study how readers naturally generate boundaries. The task would be creating the effect. Precisely. So the first set of experiments were all about validating the maze. They had to prove it was sensitive, first to the really fine -grained parts of porosity, and then to the bigger, established phrasal effects. They started small, then, with metrical porosity, lexical stress. The smallest unit, right. This is about the rhythm inside individual words.
The classic example is the noun convict, with stress on the first syllable, versus the verb convict, with stress on the second. And the idea for the experiment here was pretty clever. It was ingenious. They built these temporarily ambiguous sentences. For example, they'd lead a reader to first interpret a word like abstract as a noun. A -B -stract. And then a few words later, they'd hit a disambiguating word that forced them to realize it was actually a verb. Abstract. So the parser has to do a syntactic reanalysis. Yeah. It has to switch from noun to verb. But critically, in some of the sentences, that syntactic reanalysis also forced a metrical reanalysis.
You had to change the stress pattern in your head. And the prediction is, if implicit prosody really matters. Then forcing someone to do both a syntactic and a metrical reanalysis should be way more costly than just the syntactic one alone. So what did they find? This is experiment one in the maze and two in SPR. The results were clear. Both tasks were sensitive to this tiny prosodic detail. There was consistently more processing difficulty, longer response times, more errors in the maze when the parser had to switch both the syntax and the stress pattern.
Which confirms that even the super localized word -by -word decision in the maze task is incorporating implicit stress. Yes. It's not ignoring it. And what I find really fascinating is that the maze showed more localized effects. What do you mean by that? The slowdown, the moment of difficulty, happened immediately. Right on the word that forced the reanalysis. In the SPR task, as expected, the cost kind of spilled over onto the next couple of words. Ah, I see. So the maze might actually be better at pinpointing the exact moment to difficulty. It supports the idea that the maze is great for studying these rapid incremental decisions because its forced choices prevent the parser from sort of hiding the cost downstream.
Okay, so the maze is good for word stress, but now they had to move up to the bigger question. Phrasal prosody. And this is where they test that core fear. Does the maze force a staccato prosody? This is the critical test for the whole methodology. And they used a classic sentence structure for it, the NPV garden path. An example would be something like, since the comedian left the manager. Exactly. The garden path effect there relies on the reader initially assuming one single uninterrupted phrase. The comedian left the manager. That's the NP interpretation, where manager is the object of left. Right. But the sentence could continue differently, making manager the subject of a new clause. That's the Z interpretation. Correct. And that Z interpretation is strongly dispreferred unless you have a really clear Q, like a comma. So the manipulation was just presence or absence of a comma. Now let's go back to that interference hypothesis. The fear was this. If the maze task forces an unnatural pause -heavy prosody on the reader, then it should automatically be inserting an implicit boundary after the verb left. And a boundary there would actually help the z -parse. It would separate manager from left, making it easy to see it as a new subject. You got it. So if the maze was flawed in the way people feared, it should have completely reversed the classic garden path effect.
It should have made the z -interpretation easier. So? Not even close. The research clearly replicated the classic garden path in the maze task. Participants showed a huge slowdown when the sentence forced that Z interpretation without a calmer. Which proves they were going down the default MP path first and getting tripped up. Exactly. And that is a huge methodological takeaway. It shows the maze does not inherently introduce some weird boundary -heavy prosody that overrides these established parsing effects. It seems to capture the default natural contour just fine. Okay, so one final check. To really cement the maze's reliability for this kind of work, they had to tackle Fodor's length effect. The cornerstone fight. Right. The idea that long relative clauses drive high attachment because of the implicit boundary they generate. If the maze can't capture that subtle length -based effect, then it's kind of useless for the rest of this investigation.
It would have been a deal breaker. But fortunately, both the maze in Experiment 5 and the SPR in Experiment 6 replicated the effect beautifully. So participants consistently chose the high attachment interpretation when the RCs were longer. Consistently. It validated the entire experimental approach. The maze is sensitive enough to pick up on the implicit boundary generated by sheer constituent size. And this is where the reading time data gets really provocative, and it links back to this idea that prosody can actually help processing. We're talking about the anti -locality effect.
Yes. This was a surprising finding. They found that the region of the sentence after the long relative clause, so the main verb of the sentence, was actually read faster than the same region after the short relative clause. Which is completely counterintuitive. It's the definition of anti -locality. Making something longer and more complex actually makes a later part of the sentence easier to process. Right. Your intuition says a longer, more complex phrase should just exhaust the parser, slowing everything down later. So why does this happen? Well, the speculation centers on a couple of possibilities. One is about reduced surprisal.
A very long, complex RC might just strongly constrain what can possibly come next grammatically. So the main verb becomes highly predictable. Okay, that's one idea. But the other possibility, the one that's really relevant to our deep dive, is implicit prosody. The mental pause button again. The mental pause button. The implicit boundary that the parser places after that long relative clause isn't just for attachment. It also acts as a signal to the parser. It says, OK, this complex dependent clause is structurally complete. We're done with it. It's like a mental restart button. That's a great way to think of it. It allows the parser to efficiently shift its focus back to the main clause and quickly process that matrix verb when it finally shows up. The cognitive cost of assigning the boundary is paid early, and the reward is a smoother, faster ride later on. That's a really powerful idea that the structural boundary we mentally impose serves this therapeutic dependency closing function, and it actually accelerates our overall parsing speed. It's a key part of the puzzle.
All right, so we've established that implicit prosody is real, it's measurable, and it doesn't get messed up by the maze task. Now we can finally get into the theory. The fun part. How does the inner voice decide, word by word, where to put those crucial boundaries? This section of the work shifts into building a sort of toy model of incremental prosodic parsing. Right. We need to define the grammatical constraints that are guiding the parser. And the core tension here is that prosodic structure, the rhythmic phrasing, it generally mirrors syntactic structure, but...
Not perfectly. You can have one grammatical structure with a few different possible rhythmic renderings. Exactly. So we need rules that govern the relationship between syntactic constituents, like a verb phrase, and prosodic constituents, like a phonological phrase. And the dissertation pulls from two major theoretical approaches here. It does. First, you have alignment theory. This theory is all about aligning the edges of syntactic constituents to the edges of prosodic constituents. For a parser that's working incrementally, this is a relatively simple calculation. Because it only needs to know where the current edge is. It doesn't need to have a picture of the whole structure yet. Correct. The alternative is match theory, which is a bit stricter. It requires a more direct one -to -one correspondence, where syntactic phrases have to directly match and cover the same material as prosodic phrases. And that would often require more complex, nested prosodic structures.
It would. So for a parser that's working in real time with limited look -ahead, alignment constraints are just. They're easier to satisfy on the fly. But beyond just aligning to syntax, the parser also has to obey its own internal rhythmic preferences. And these are critical for explaining those length effects we just talked about. Absolutely critical. The internal rhythm of the parser, it seems, prefers balance and simplicity. Prosodic phrases tend to gravitate toward a binary structure. Meaning they prefer to contain just two prosodic words. That's the ideal, yeah. The inner voice is kind of a rhythm dictator that prefers simple two -beat measures. And this preference is enforced by specific constraints. It is. Constraints like bin max and bin max sight sime. Bin min penalizes phrases that are too short, less than two words.
But the really important one for length effects is bin max, which penalizes phrases that are too long. Now, obviously, many phrases in English are longer than two words. Of course. But the idea behind bin max is that the penalty, and therefore the pressure on the parser to insert a boundary, increases exponentially as that constituent gets longer and longer. It's like pressure building up in a system. Exactly. As the current prosodic phrase keeps accumulating words, it gets heavy and starts violating binmax more and more. This violation makes it much more likely that the parser will just decide to break the phrase up and insert a boundary even before it's seen the end of the sentence. And this links directly back to Fodor's findings. A long RC is a heavy constituent. It's a huge binmax violation. It is. And on top of that, we also have what are called uniformity or balance constraints. Fodor's same -size sister principle is a good example.
The parser likes adjacent prosodic phrases to be of singular structural weight. So a heavy phrase wants to pair up with another heavy phrase. Right. A long RC, which is a heavy constituent, searches for another heavy constituent to attach to for balance, like the high noun phrase, which contains both nouns. This whole prosodic balancing act is what's driving the attachment decision. Okay, let's put these rules into motion now and see how the parser actually builds prosody word by word. We have to start with a default behavior, which you mentioned earlier, prosodic late closure.
Why is that the default? Why is it risk -averse? Well, because committing to a prosodic boundary too early is really risky. If you close off a phrase and then the sentence continues in a way that shows it should have stayed open, you've made a mistake and you have to go back and reanalyze. Which is costly. Very costly. So in the absence of a strong Q, like a comma, the parser's default is to just keep the current phonological phrase open, just keep adding material to it. This minimizes the risk of early reanalysis. Okay, let's walk through the short RC scenario first with a simple example. Let's use the brother of the psychiatrist who listened. So word by word. One, the brother.
A new phonological phrase, a figat opens. Two, of the psychiatrist. This gets added to the open figure. It's not too heavy yet. Bean Max isn't screaming. Prosodic late closure keeps the phrase open. Three, who listened? This also gets added to the open figure. And the result is a single manageable prosodic phrase that covers the whole complex NP and the short RC. Right. No boundary is inserted. And because there's no boundary, the low attachment site, psychiatrist, is the most visible and local choice. The first pass prosody matches the final preferred prosody.
Low attachment wins. Simple. Okay, now for the long RC scenario. This is the moment where the inner voice changes its mind. Let's say it's the brother of the psychiatrist who listened to the callers. The initial steps are identical. The parser is following late closure, so it's initially favoring low attachment. But as the phrase keeps growing, one, the brother of the psychiatrist, door is open. Two, who listened to the callers. This material just keeps getting added to that same open viola. The whole thing now contains many prosodic words. And the pressure is building. The pressure is building. The phrase becomes structurally heavy. The bin max constraint starts to impose a significant penalty. This makes the insertion of a boundary much, much more likely, even though the parser still hasn't seen the end of the sentence. So the sheer weight of the phrase forces the parser's hand. The parser knows, based on these internal rhythmic rules, that this long string of words is not a default.
Simple phrase. That's it. So when the parser finally hits the main verb of the sentence, the unambiguous cue that the whole RC is finally over, it decides on its final prosodic structure. And given the huge binmax violation from the length, it's going to choose to insert a boundary between the complex MP and the RC. It partitions the sentence to satisfy those balance constraints. Right. And that boundary, once it's inserted, repels the RC, which drives high attachment. But this is where you see the crucial divergence.
The first pass decision, driven by risk aversion and late closure, favored low attachment. But the final default prosody, driven by the length constraints and bin max, forces a boundary and mandates high attachment. And that potential for divergence, that's the key insight of the model. We can now hypothesize when the parser faces the highest cost of reanalysis, not just syntactically, but prosodically. It pinpoints the moment the inner voice realizes it has to change its tune. It clarifies everything. Okay, so the model successfully explains these length effects. But the dissertation, very wisely, looks ahead to the remaining mysteries. What foundational concepts did this initial model kind of put on the back burner? The initial model worked really well because it could rely mostly on alignment and binarity constraints to handle length. But future work has to grapple with deeply recursive structures. Complex sentences where phrases are nested inside other phrases. Exactly. And to model that, you really have to adopt stronger versions of match theory. And what specific constraints would come into play when you're dealing with that kind of nested structure? Well, things like strong start and equal sisters. Strong start is a constraint that basically ensures a major prosodic constituent begins with a strong metrical unit. And equal sisters is a constraint that pushes for prosodic sister constituents to be of the same prosodic category. Which reinforces that idea of structural balance and uniform recursion.
It does, but it's a big jump in complexity, right? Moving from just deciding where to put a boundary. to ensuring all these nested rhythms match up perfectly. It seems like a huge jump. It is, but it's a necessary one if we want a complete model. If the parser is building complex syntactic structure incrementally, it must be building recursive prosodic structure in parallel. These constraints are the tools we'll need to model that process. We've built the theoretical engine. Now it's time for the ultimate experimental test. Chapter 4. Right. This is where the research moves away from just manipulating the length of the relative clause itself and focuses on modifying the noun phrase before the RC. The goal is to induce an implicit boundary precisely where they want it. Yes. The goal was to pit the two big competing forces, focus and attraction from the FAH versus boundaries and visibility from the VFH against each other in a controlled written environment. Can we use length modification to force an implicit boundary and then see if that boundary overrides the informational focus of the modified noun? That was the million -dollar question. Yeah. So instead of a simple phrase like the niece of the waitress, they created complex noun phrases like the niece of the incredibly diligent waitress. Adding that two -word modifier, incredibly diligent, dramatically increases the weight and the length of that N2 constituent.
And... Based on the incremental model we just discussed, increasing the length of a constituent significantly increases the probability that the parser, under pressure from Benmax, will insert an implicit prosodic boundary right after it. Okay, so let's quickly review the three competing predictions for what would happen next. Sure. First, the repellent boundaries hypothesis, RBH. Modification creates a boundary, and that boundary repels the RC, pushing it high. Second. The modification attraction hypothesis, MA. F -A -H. The focus created by the modification attracts the RC to it. And third, the big one. The visibility -first hypothesis, the F -H. Boundaries are king. Attachment always targets the most visible, boundary -free site. Okay, let's get to the results. Experiment 7, modified noun 2, N2. So the niece of the incredibly diligent waitress. And when N2 was modified, the interpretation data showed that participants preferred more low attachment. They attached the RSC to the modified N2, the waitress. Which is ambiguous. Completely ambiguous. That result could support the MAH focus attracts attachment to N2, or it could support the VFH if you assume the boundary lands after the modified N2, but N2 is still the most local visible site. So no clear winner there. So Experiment 7 didn't settle the debate, but Experiment 8 provided the critical test. This time they modified noun 1, N1. For example, the incredibly diligent niece of the waitress. And this is the moment where everything flips. Because if the MAHFAH were correct, modifying N1 should create focus on it, pulling the RC high to N1. But that's not what happened. Not at all. The response findings were completely counterintuitive. Modifying N1 led to more low attachment. Attachment to N2, the unmodified noun. Wait, hold on. Why would focusing on the niece make me think the relative clause refers to the waitress? That seems to completely contradict the idea that focus attracts attachment. It does.
It shatters it. The MAHFEH failed this test. And this is the strong, almost undeniable evidence that elevates the visibility first hypothesis. Okay, walk us through the VFH explanation for this exact result. So when the parser encounters the incredibly diligent niece, the length of that N1 constituent goes way up. The internal rhythmic constraints, binax, are just screaming. This forces the parser to insert an implicit prosodic boundary right after N1. So there's an invisible wall after knees. Exactly. And that wall makes N1 invisible. Even though N1 carried the focus in the modification, that boundary renders it structurally inaccessible. Noun 2, waitress, is now immediately adjacent to the incoming relative pronoun who. It has become the most local and most visible attachment site.
And the parser, prioritizing visibility over focus, attaches low. It attaches low to N2. This result confirms the hierarchy. Implicit prosodic boundaries, which are driven by constituent length, are the dominant factor in written ambiguity resolution. That's just a fundamental clarifying point on how we read. It's not about what we mentally highlight with focus. It's about where our internal rhythm forces us to take a pause. And to make sure this wasn't some kind of fluke, Experiment 9 generalized the findings. They crossed both NP modification types, N1 or N2, with the classic RC length effect. The findings generalized perfectly. N1 modification and N2 modification both consistently drove attachment to the lower.
N2 site, no matter how long the RC was. It just reinforces the VFH. Length leads to boundaries. Boundaries lead to invisibility. And visibility guides attachment, period. Before we wrap up the experiments, we have to revisit the timing data because it offered this tantalizing clue about when the cognitive cost of this whole process happens. The reading times, right. Where did the parser actually slow down relative to where we think the boundary is? Across multiple experiments, there was this really suggestive pattern. The slowdown didn't happen before the boundary, which you might expect if the cost was in the decision -making. Instead, it happened on the word immediately following the hypothesized boundary. Can you give us an example of that? Sure. When N1 was modified, like in the incredibly diligent Nice, we predict a boundary right after Nice. The slowdown in reading time was observed on the very next word, the preposition of. And similarly for N2.
Same thing. When N2 was modified, the boundary was predicted after the whole complex N2 phrase, and the slowdown was observed on the relative pronoun who, the word that came right after. This timing is so fascinating. If the slowdown happens after the decision point, what does that tell us about the parser's mechanism? It really supports that concept of prosodic late closure. It suggests the slowdown might be the cognitive cost of either realizing the expected boundary has now occurred or the cost of actively assigning that boundary once the constituent length has passed some critical threshold. So the parser is basically holding off on committing to the pause until the very last second. It waits until the structural pressure from BinMax builds up and then the cost is incurred at the moment of commitment. And that registers as the processing time on the very next word. It's just highly compatible with the whole model. Risk aversion keeps the phrase open, length increases the pressure to close it, and the cost of finally closing it shows up just after that crucial decision point. It all fits together. We've covered the methodology, the theory, and the definitive experimental findings. If we take a step back, what is the integrated picture of silent reading that we're left with? Well, first, the methodological point is rock solid. Okay. The maze task is a sensitive and reliable tool for studying implicit prosody.
It gives you these localized effects that are incredibly valuable, and it does not force an unnatural rhythm. And second, the major theoretical takeaway is the hierarchy of influences. We have, I would say, overwhelming evidence now that the parser's primary guide during ambiguity resolution is the visibility -first hypothesis. Attachment preferences are governed first and foremost by implicit prosodic boundaries generated during reading. And the role of information, structure focus, modification, accent is really relegated to a secondary modulatory status. It only kicks in when prosodic visibility is neutral. That's right. The finding that just making N1 longer forces a boundary that completely overrides N1's focus. That is the defining insight of the entire investigation. Length dictates boundaries and boundaries dictate meaning. And let's connect this back to that reading time data, the anti -locality effect. Right. The faster reading after a long RC. It strongly suggests that implicit prosody isn't just affecting what the parser attaches to, but also how fast it processes what comes next. The boundary placement acts as a structural signpost. It signals the closure of a complex dependency, and that seems to ease the load on the parser as it transitions back to the main clause. It really makes that inner voice a core computational component of the brain, affecting both our interpretation and our processing speed. But this work... As successful as it is, it leaves these profound questions for future research about the inner voice. The unfinished work of prosodic parsing, what are the key areas we still need to investigate? We need better confirmation tools. I mean, while the data strongly suggests where these implicit boundaries are being placed, we don't have direct evidence of the reader's internal prosodic structure. We're inferring it from their choices. We are. So future studies need to develop new techniques, maybe something like rapid prosodic transcription tasks, to confirm precisely where readers actually place these boundaries and then correlate that data directly with their parsing choices. To verify that the binmax boundary we predict is in fact the boundary the reader is actually generating.
Exactly. And then on the theory side, we have to address those complex nested structures we touched on. We need to incorporate constraints like strong start and equal sisters and see if the parser really generates recursive prosodic structure. That seems like a massive next step to determine if our inner voice is capable of generating fully recursive nested rhythmic structures just based on visual input from a page. It is. But the goal remains to build that robust incremental model that can explain the divergence between the immediate risk -averse prosody of the first pass, and the optimal balanced final prosody of the full sentence. The work detailed in this investigation confirms that the sound of silence is loud enough to change the entire meaning of the words we read. And that brings us to our final provocation for you, the listener. If length and visual structure fundamentally dictate the unseen rhythm in your head, consider this. The next time you're struggling with an ambiguous sentence in an email or a document, Perhaps the difficulty isn't syntactic at all. Maybe your brain is simply arguing with itself about where the inner pause button should have been placed.
The parser isn't just reading words. It's constantly constructing a rhythm and then listening to the structure that it creates. Understanding this internal orchestra might just be the key to understanding comprehension itself.