Skip to content

Improve next-word predictions #255

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
astroboii47 wants to merge 2 commits into from
Closed

Improve next-word predictions #255

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
5aba308
Add bundled next-word phrase fallback
astroboii47 Jul 16, 2026
32aafd7
Advance predictions after accepting suggestion
astroboii47 Jul 16, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
129 changes: 129 additions & 0 deletions app/src/main/assets/common/predictions/en_next_words.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,129 @@
{
"fallback": [
"the",
"to",
"and",
"you",
"it",
"that",
"for",
"with",
"this",
"have",
"your",
"from"
],
"bigrams": {
"a": ["lot", "little", "few", "bit", "good", "great", "new"],
"about": ["the", "this", "that", "it", "what", "how"],
"all": ["the", "of", "you", "this", "that"],
"am": ["not", "going", "trying", "glad", "sure"],
"and": ["the", "i", "you", "we", "it", "then", "that"],
"are": ["you", "we", "they", "there", "not", "going"],
"as": ["a", "well", "soon", "long", "much", "if"],
"at": ["the", "all", "least", "first", "home", "work"],
"be": ["able", "there", "sure", "back", "careful", "good"],
"been": ["a", "there", "trying", "working", "doing"],
"but": ["i", "it", "you", "we", "the", "that"],
"by": ["the", "far", "then", "default", "my"],
"can": ["you", "we", "i", "do", "get", "see", "help"],
"could": ["you", "we", "be", "do", "get", "try"],
"did": ["you", "we", "they", "it", "not"],
"do": ["you", "we", "i", "not", "that", "this"],
"does": ["it", "that", "this", "not"],
"for": ["the", "you", "me", "this", "that", "now"],
"from": ["the", "you", "my", "there", "here"],
"get": ["the", "it", "back", "started", "to", "a"],
"good": ["luck", "idea", "morning", "night", "to"],
"got": ["it", "to", "the", "a", "some"],
"had": ["to", "a", "been", "no", "the"],
"has": ["been", "to", "a", "the", "no"],
"have": ["a", "to", "been", "the", "you", "we"],
"he": ["was", "is", "said", "has", "will"],
"hello": ["there", "everyone", "again"],
"here": ["is", "are", "we", "you", "it"],
"how": ["are", "do", "to", "can", "would", "much", "long"],
"i": ["am", "have", "think", "need", "want", "will", "can", "was"],
"if": ["you", "we", "it", "that", "there"],
"in": ["the", "a", "my", "this", "that"],
"is": ["the", "a", "it", "not", "there", "that"],
"it": ["is", "was", "would", "will", "has", "can", "should"],
"just": ["a", "the", "to", "let", "be", "wanted"],
"know": ["if", "what", "that", "how", "you"],
"let": ["me", "us", "it", "them"],
"like": ["to", "a", "the", "this", "that", "you"],
"me": ["know", "if", "when", "what", "and"],
"my": ["phone", "keyboard", "app", "friend", "device"],
"need": ["to", "a", "the", "help", "more"],
"new": ["one", "version", "build", "feature", "keyboard"],
"not": ["sure", "working", "bad", "the", "a"],
"of": ["the", "this", "that", "it", "course"],
"on": ["the", "my", "this", "that", "your"],
"or": ["the", "you", "we", "not", "maybe"],
"please": ["let", "check", "send", "help"],
"should": ["be", "have", "work", "not"],
"so": ["i", "you", "we", "it", "that"],
"thanks": ["for", "again", "so", "you"],
"that": ["is", "was", "would", "should", "you"],
"the": ["same", "best", "app", "keyboard", "phone", "issue", "other"],
"there": ["is", "are", "was", "will", "should"],
"they": ["are", "have", "will", "can", "were"],
"this": ["is", "was", "would", "should", "one"],
"to": ["the", "be", "do", "get", "make", "use", "see"],
"want": ["to", "a", "the", "it", "more"],
"was": ["the", "a", "not", "going", "just"],
"we": ["can", "should", "are", "need", "have", "will"],
"what": ["is", "do", "are", "you", "we"],
"when": ["you", "we", "it", "the", "i"],
"where": ["is", "are", "do", "can"],
"why": ["is", "does", "do", "are"],
"will": ["be", "have", "not", "the", "it"],
"with": ["the", "this", "that", "you", "a"],
"would": ["be", "like", "have", "you", "it"],
"you": ["are", "can", "should", "will", "have", "need", "know", "want"],
"your": ["phone", "keyboard", "app", "settings", "device"]
},
"trigrams": {
"a lot": ["of"],
"all of": ["the", "this", "that"],
"can i": ["get", "use", "have", "see"],
"can you": ["help", "check", "send", "see", "try"],
"do you": ["want", "know", "think", "have", "need"],
"for the": ["same", "keyboard", "app", "phone"],
"how are": ["you", "we", "things"],
"how do": ["you", "we", "i"],
"i am": ["not", "going", "trying", "glad", "sure"],
"i can": ["do", "see", "try", "help"],
"i did": ["not", "see", "try"],
"i do": ["not", "think", "want"],
"i had": ["to", "a", "been"],
"i have": ["to", "a", "been", "no"],
"i just": ["wanted", "need", "have"],
"i know": ["that", "what", "you"],
"i need": ["to", "a", "the", "help"],
"i think": ["it", "we", "you", "that"],
"i want": ["to", "a", "the", "it"],
"i will": ["be", "have", "try", "check"],
"if you": ["want", "need", "can", "have"],
"in the": ["same", "app", "keyboard", "settings"],
"is the": ["same", "best", "keyboard", "app"],
"it is": ["not", "the", "a", "just"],
"let me": ["know", "check", "see", "try"],
"need to": ["be", "do", "get", "make"],
"one of": ["the", "my", "your"],
"please let": ["me"],
"so i": ["can", "have", "know"],
"thank you": ["for", "so", "again"],
"there is": ["a", "no", "the", "nothing"],
"this is": ["the", "a", "not", "what"],
"to the": ["same", "app", "keyboard"],
"want to": ["be", "do", "get", "make"],
"we are": ["not", "going", "trying", "ready"],
"we can": ["do", "try", "see", "make"],
"what do": ["you", "we", "i"],
"you are": ["the", "not", "going"],
"you can": ["do", "see", "try", "use"],
"you have": ["to", "a", "the"],
"you need": ["to", "a", "the"]
}
}
116 changes: 116 additions & 0 deletions app/src/main/java/it/palsoftware/pastiera/core/suggestions/BundledPhrasePredictor.kt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,116 @@
package it.palsoftware.pastiera.core.suggestions

import android.content.Context
import org.json.JSONArray
import org.json.JSONObject
import java.util.Locale
import java.util.concurrent.ConcurrentHashMap

class BundledPhrasePredictor(
context: Context,
private val locale: Locale = Locale.getDefault(),
private val isValidWord: ((String) -> Boolean)? = null
) {
private val appContext = context.applicationContext

companion object {
private const val PREDICTION_ASSET_PREFIX = "common/predictions"
private val normalizeRegex = Regex("[^\\p{L}']")
private val cache = ConcurrentHashMap<String, PhraseData>()
private val emptyPhraseData = PhraseData(emptyMap(), emptyMap(), emptyList())
}

fun predictContextual(previousWord: String, previousPreviousWord: String?, limit: Int): List<String> {
if (limit <= 0) return emptyList()
val data = loadData()
if (data === emptyPhraseData) return emptyList()

val previous = normalize(previousWord) ?: return emptyList()
val previousPrevious = normalize(previousPreviousWord)
val results = linkedSetOf<String>()

if (previousPrevious != null) {
data.trigrams["$previousPrevious $previous"]
?.asSequence()
?.filter(::isPlausibleWord)
?.forEach(results::add)
}
data.bigrams[previous]
?.asSequence()
?.filter(::isPlausibleWord)
?.forEach(results::add)

return results.take(limit)
}

fun fallbackWords(limit: Int): List<String> {
if (limit <= 0) return emptyList()
return loadData().fallback
.asSequence()
.filter(::isPlausibleWord)
.take(limit)
.toList()
}

private fun loadData(): PhraseData {
val language = locale.language.lowercase(Locale.ROOT)
if (language != "en") return emptyPhraseData
return cache.getOrPut(language) {
val path = "$PREDICTION_ASSET_PREFIX/${language}_next_words.json"
runCatching {
appContext.assets.open(path).bufferedReader().use { reader ->
val root = JSONObject(reader.readText())
PhraseData(
bigrams = parseMap(root.optJSONObject("bigrams")),
trigrams = parseMap(root.optJSONObject("trigrams")),
fallback = parseArray(root.optJSONArray("fallback"))
)
}
}.getOrElse {
emptyPhraseData
}
}
}

private fun parseMap(source: JSONObject?): Map<String, List<String>> {
if (source == null) return emptyMap()
return source.keys().asSequence().associateWith { key ->
parseArray(source.optJSONArray(key))
}
}

private fun parseArray(source: JSONArray?): List<String> {
if (source == null) return emptyList()
return buildList {
for (index in 0 until source.length()) {
source.optString(index)
.takeIf { it.isNotBlank() }
?.let(::add)
}
}
}

private fun normalize(word: String?): String? {
if (word.isNullOrBlank()) return null
val normalized = word
.replace('’', '\'')
.replace('‘', '\'')
.replace('ʼ', '\'')
.lowercase(locale)
.replace(normalizeRegex, "")
.trim('\'')
return normalized.takeIf { it.isNotBlank() }
}

private fun isPlausibleWord(word: String): Boolean {
val normalized = normalize(word) ?: return false
if (normalized.length < 2) return false
return isValidWord?.invoke(normalized) ?: true
}

private data class PhraseData(
val bigrams: Map<String, List<String>>,
val trigrams: Map<String, List<String>>,
val fallback: List<String>
)
}
Loading